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 #include "pull_rotation.h"
  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_THREAD_MPI
  94 #include "tmpi.h"
  95 #endif
  96
  97 #include "adress.h"
  98 #include "qmmm.h"
  99
 100 #if 0
 101 typedef struct gmx_timeprint {
 102
 103 } t_gmx_timeprint;
 104 #endif
 105
 106 /* Portable version of ctime_r implemented in src/gmxlib/string2.c, but we do not want it declared in public installed headers */
 107 char *
 108 gmx_ctime_r(const time_t *clock,char *buf, int n);
 109
 110
 111 double
 112 gmx_gettime()
 113 {
 114 #ifdef HAVE_GETTIMEOFDAY
 115         struct timeval t;
 116         double seconds;
 117
 118         gettimeofday(&t,NULL);
 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_THREAD_MPI
 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_THREAD_MPI
 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     gmx_bool   bDoAdressWF;
 432     matrix boxs;
 433     real   e,v,dvdl;
 434     t_pbc  pbc;
 435     float  cycles_ppdpme,cycles_pme,cycles_seppme,cycles_force;
 436
 437     start  = mdatoms->start;
 438     homenr = mdatoms->homenr;
 439
 440     bSepDVDL = (fr->bSepDVDL && do_per_step(step,inputrec->nstlog));
 441
 442     clear_mat(vir_force);
 443
 444     if (PARTDECOMP(cr))
 445     {
 446         pd_cg_range(cr,&cg0,&cg1);
 447     }
 448     else
 449     {
 450         cg0 = 0;
 451         if (DOMAINDECOMP(cr))
 452         {
 453             cg1 = cr->dd->ncg_tot;
 454         }
 455         else
 456         {
 457             cg1 = top->cgs.nr;
 458         }
 459         if (fr->n_tpi > 0)
 460         {
 461             cg1--;
 462         }
 463     }
 464
 465     bStateChanged = (flags & GMX_FORCE_STATECHANGED);
 466     bNS           = (flags & GMX_FORCE_NS) && (fr->bAllvsAll==FALSE);
 467     bFillGrid     = (bNS && bStateChanged);
 468     bCalcCGCM     = (bFillGrid && !DOMAINDECOMP(cr));
 469     bDoLongRange  = (fr->bTwinRange && bNS && (flags & GMX_FORCE_DOLR));
 470     bDoForces     = (flags & GMX_FORCE_FORCES);
 471     bSepLRF       = (bDoLongRange && bDoForces && (flags & GMX_FORCE_SEPLRF));
 472     /* should probably move this to the forcerec since it doesn't change */
 473     bDoAdressWF   = ((fr->adress_type!=eAdressOff));
 474
 475     if (bStateChanged)
 476     {
 477         update_forcerec(fplog,fr,box);
 478
 479         /* Calculate total (local) dipole moment in a temporary common array.
 480          * This makes it possible to sum them over nodes faster.
 481          */
 482         calc_mu(start,homenr,
 483                 x,mdatoms->chargeA,mdatoms->chargeB,mdatoms->nChargePerturbed,
 484                 mu,mu+DIM);
 485     }
 486
 487   if (fr->ePBC != epbcNONE) {
 488     /* Compute shift vectors every step,
 489      * because of pressure coupling or box deformation!
 490      */
 491     if ((flags & GMX_FORCE_DYNAMICBOX) && bStateChanged)
 492       calc_shifts(box,fr->shift_vec);
 493
 494     if (bCalcCGCM) {
 495       put_charge_groups_in_box(fplog,cg0,cg1,fr->ePBC,box,
 496                                &(top->cgs),x,fr->cg_cm);
 497       inc_nrnb(nrnb,eNR_CGCM,homenr);
 498       inc_nrnb(nrnb,eNR_RESETX,cg1-cg0);
 499     }
 500     else if (EI_ENERGY_MINIMIZATION(inputrec->eI) && graph) {
 501       unshift_self(graph,box,x);
 502     }
 503   }
 504   else if (bCalcCGCM) {
 505     calc_cgcm(fplog,cg0,cg1,&(top->cgs),x,fr->cg_cm);
 506     inc_nrnb(nrnb,eNR_CGCM,homenr);
 507   }
 508
 509   if (bCalcCGCM) {
 510     if (PAR(cr)) {
 511       move_cgcm(fplog,cr,fr->cg_cm);
 512     }
 513     if (gmx_debug_at)
 514       pr_rvecs(debug,0,"cgcm",fr->cg_cm,top->cgs.nr);
 515   }
 516
 517 #ifdef GMX_MPI
 518   if (!(cr->duty & DUTY_PME)) {
 519     /* Send particle coordinates to the pme nodes.
 520      * Since this is only implemented for domain decomposition
 521      * and domain decomposition does not use the graph,
 522      * we do not need to worry about shifting.
 523      */
 524
 525     wallcycle_start(wcycle,ewcPP_PMESENDX);
 526     GMX_MPE_LOG(ev_send_coordinates_start);
 527
 528     bBS = (inputrec->nwall == 2);
 529     if (bBS) {
 530       copy_mat(box,boxs);
 531       svmul(inputrec->wall_ewald_zfac,boxs[ZZ],boxs[ZZ]);
 532     }
 533
 534     gmx_pme_send_x(cr,bBS ? boxs : box,x,
 535                    mdatoms->nChargePerturbed,lambda,
 536                    ( flags & GMX_FORCE_VIRIAL),step);
 537
 538     GMX_MPE_LOG(ev_send_coordinates_finish);
 539     wallcycle_stop(wcycle,ewcPP_PMESENDX);
 540   }
 541 #endif /* GMX_MPI */
 542
 543     /* Communicate coordinates and sum dipole if necessary */
 544     if (PAR(cr))
 545     {
 546         wallcycle_start(wcycle,ewcMOVEX);
 547         if (DOMAINDECOMP(cr))
 548         {
 549             dd_move_x(cr->dd,box,x);
 550         }
 551         else
 552         {
 553             move_x(fplog,cr,GMX_LEFT,GMX_RIGHT,x,nrnb);
 554         }
 555         /* When we don't need the total dipole we sum it in global_stat */
 556         if (bStateChanged && NEED_MUTOT(*inputrec))
 557         {
 558             gmx_sumd(2*DIM,mu,cr);
 559         }
 560         wallcycle_stop(wcycle,ewcMOVEX);
 561     }
 562     if (bStateChanged)
 563     {
 564
 565         /* update adress weight beforehand */
 566         if(bDoAdressWF)
 567         {
 568             /* need pbc for adress weight calculation with pbc_dx */
 569             set_pbc(&pbc,inputrec->ePBC,box);
 570             if(fr->adress_site == eAdressSITEcog)
 571             {
 572                 update_adress_weights_cog(top->idef.iparams,top->idef.il,x,fr,mdatoms,
 573                                           inputrec->ePBC==epbcNONE ? NULL : &pbc);
 574             }
 575             else if (fr->adress_site == eAdressSITEcom)
 576             {
 577                 update_adress_weights_com(fplog,cg0,cg1,&(top->cgs),x,fr,mdatoms,
 578                                           inputrec->ePBC==epbcNONE ? NULL : &pbc);
 579             }
 580             else if (fr->adress_site == eAdressSITEatomatom){
 581                 update_adress_weights_atom_per_atom(cg0,cg1,&(top->cgs),x,fr,mdatoms,
 582                                           inputrec->ePBC==epbcNONE ? NULL : &pbc);
 583             }
 584             else
 585             {
 586                 update_adress_weights_atom(cg0,cg1,&(top->cgs),x,fr,mdatoms,
 587                                            inputrec->ePBC==epbcNONE ? NULL : &pbc);
 588             }
 589         }
 590
 591         for(i=0; i<2; i++)
 592         {
 593             for(j=0;j<DIM;j++)
 594             {
 595                 fr->mu_tot[i][j] = mu[i*DIM + j];
 596             }
 597         }
 598     }
 599     if (fr->efep == efepNO)
 600     {
 601         copy_rvec(fr->mu_tot[0],mu_tot);
 602     }
 603     else
 604     {
 605         for(j=0; j<DIM; j++)
 606         {
 607             mu_tot[j] =
 608                 (1.0 - lambda)*fr->mu_tot[0][j] + lambda*fr->mu_tot[1][j];
 609         }
 610     }
 611
 612     /* Reset energies */
 613     reset_enerdata(&(inputrec->opts),fr,bNS,enerd,MASTER(cr));
 614     clear_rvecs(SHIFTS,fr->fshift);
 615
 616     if (bNS)
 617     {
 618         wallcycle_start(wcycle,ewcNS);
 619
 620         if (graph && bStateChanged)
 621         {
 622             /* Calculate intramolecular shift vectors to make molecules whole */
 623             mk_mshift(fplog,graph,fr->ePBC,box,x);
 624         }
 625
 626         /* Reset long range forces if necessary */
 627         if (fr->bTwinRange)
 628         {
 629             /* Reset the (long-range) forces if necessary */
 630             clear_rvecs(fr->natoms_force_constr,bSepLRF ? fr->f_twin : f);
 631         }
 632
 633         /* Do the actual neighbour searching and if twin range electrostatics
 634          * also do the calculation of long range forces and energies.
 635          */
 636         dvdl = 0;
 637         ns(fplog,fr,x,box,
 638            groups,&(inputrec->opts),top,mdatoms,
 639            cr,nrnb,lambda,&dvdl,&enerd->grpp,bFillGrid,
 640            bDoLongRange,bDoForces,bSepLRF ? fr->f_twin : f);
 641         if (bSepDVDL)
 642         {
 643             fprintf(fplog,sepdvdlformat,"LR non-bonded",0.0,dvdl);
 644         }
 645         enerd->dvdl_lin += dvdl;
 646
 647         wallcycle_stop(wcycle,ewcNS);
 648     }
 649
 650     if (inputrec->implicit_solvent && bNS)
 651     {
 652         make_gb_nblist(cr,inputrec->gb_algorithm,inputrec->rlist,
 653                        x,box,fr,&top->idef,graph,fr->born);
 654     }
 655
 656     if (DOMAINDECOMP(cr))
 657     {
 658         if (!(cr->duty & DUTY_PME))
 659         {
 660             wallcycle_start(wcycle,ewcPPDURINGPME);
 661             dd_force_flop_start(cr->dd,nrnb);
 662         }
 663     }
 664
 665     if (inputrec->bRot)
 666     {
 667         /* Enforced rotation has its own cycle counter that starts after the collective
 668          * coordinates have been communicated. It is added to ddCyclF to allow
 669          * for proper load-balancing */
 670         wallcycle_start(wcycle,ewcROT);
 671         do_rotation(cr,inputrec,box,x,t,step,wcycle,bNS);
 672         wallcycle_stop(wcycle,ewcROT);
 673     }
 674
 675     /* Start the force cycle counter.
 676      * This counter is stopped in do_forcelow_level.
 677      * No parallel communication should occur while this counter is running,
 678      * since that will interfere with the dynamic load balancing.
 679      */
 680     wallcycle_start(wcycle,ewcFORCE);
 681
 682     if (bDoForces)
 683     {
 684         /* Reset forces for which the virial is calculated separately:
 685          * PME/Ewald forces if necessary */
 686         if (fr->bF_NoVirSum)
 687         {
 688             if (flags & GMX_FORCE_VIRIAL)
 689             {
 690                 fr->f_novirsum = fr->f_novirsum_alloc;
 691                 GMX_BARRIER(cr->mpi_comm_mygroup);
 692                 if (fr->bDomDec)
 693                 {
 694                     clear_rvecs(fr->f_novirsum_n,fr->f_novirsum);
 695                 }
 696                 else
 697                 {
 698                     clear_rvecs(homenr,fr->f_novirsum+start);
 699                 }
 700                 GMX_BARRIER(cr->mpi_comm_mygroup);
 701             }
 702             else
 703             {
 704                 /* We are not calculating the pressure so we do not need
 705                  * a separate array for forces that do not contribute
 706                  * to the pressure.
 707                  */
 708                 fr->f_novirsum = f;
 709             }
 710         }
 711
 712         if (bSepLRF)
 713         {
 714             /* Add the long range forces to the short range forces */
 715             for(i=0; i<fr->natoms_force_constr; i++)
 716             {
 717                 copy_rvec(fr->f_twin[i],f[i]);
 718             }
 719         }
 720         else if (!(fr->bTwinRange && bNS))
 721         {
 722             /* Clear the short-range forces */
 723             clear_rvecs(fr->natoms_force_constr,f);
 724         }
 725
 726         clear_rvec(fr->vir_diag_posres);
 727
 728         GMX_BARRIER(cr->mpi_comm_mygroup);
 729     }
 730     if (inputrec->ePull == epullCONSTRAINT)
 731     {
 732         clear_pull_forces(inputrec->pull);
 733     }
 734
 735     /* update QMMMrec, if necessary */
 736     if(fr->bQMMM)
 737     {
 738         update_QMMMrec(cr,fr,x,mdatoms,box,top);
 739     }
 740
 741     if ((flags & GMX_FORCE_BONDED) && top->idef.il[F_POSRES].nr > 0)
 742     {
 743         /* Position restraints always require full pbc. Check if we already did it for Adress */
 744         if(!(bStateChanged && bDoAdressWF))
 745         {
 746             set_pbc(&pbc,inputrec->ePBC,box);
 747         }
 748         v = posres(top->idef.il[F_POSRES].nr,top->idef.il[F_POSRES].iatoms,
 749                    top->idef.iparams_posres,
 750                    (const rvec*)x,fr->f_novirsum,fr->vir_diag_posres,
 751                    inputrec->ePBC==epbcNONE ? NULL : &pbc,lambda,&dvdl,
 752                    fr->rc_scaling,fr->ePBC,fr->posres_com,fr->posres_comB);
 753         if (bSepDVDL)
 754         {
 755             fprintf(fplog,sepdvdlformat,
 756                     interaction_function[F_POSRES].longname,v,dvdl);
 757         }
 758         enerd->term[F_POSRES] += v;
 759         /* This linear lambda dependence assumption is only correct
 760          * when only k depends on lambda,
 761          * not when the reference position depends on lambda.
 762          * grompp checks for this.
 763          */
 764         enerd->dvdl_lin += dvdl;
 765         inc_nrnb(nrnb,eNR_POSRES,top->idef.il[F_POSRES].nr/2);
 766     }
 767
 768     /* Compute the bonded and non-bonded energies and optionally forces */
 769     do_force_lowlevel(fplog,step,fr,inputrec,&(top->idef),
 770                       cr,nrnb,wcycle,mdatoms,&(inputrec->opts),
 771                       x,hist,f,enerd,fcd,mtop,top,fr->born,
 772                       &(top->atomtypes),bBornRadii,box,
 773                       lambda,graph,&(top->excls),fr->mu_tot,
 774                       flags,&cycles_pme);
 775
 776     cycles_force = wallcycle_stop(wcycle,ewcFORCE);
 777     GMX_BARRIER(cr->mpi_comm_mygroup);
 778
 779     if (ed)
 780     {
 781         do_flood(fplog,cr,x,f,ed,box,step);
 782     }
 783
 784     if (DOMAINDECOMP(cr))
 785     {
 786         dd_force_flop_stop(cr->dd,nrnb);
 787         if (wcycle)
 788         {
 789             dd_cycles_add(cr->dd,cycles_force-cycles_pme,ddCyclF);
 790         }
 791     }
 792
 793     if (bDoForces)
 794     {
 795         if (IR_ELEC_FIELD(*inputrec))
 796         {
 797             /* Compute forces due to electric field */
 798             calc_f_el(MASTER(cr) ? field : NULL,
 799                       start,homenr,mdatoms->chargeA,x,fr->f_novirsum,
 800                       inputrec->ex,inputrec->et,t);
 801         }
 802
 803         if (bDoAdressWF && fr->adress_icor == eAdressICThermoForce)
 804         {
 805             /* Compute thermodynamic force in hybrid AdResS region */
 806             adress_thermo_force(start,homenr,&(top->cgs),x,fr->f_novirsum,fr,mdatoms,
 807                                 inputrec->ePBC==epbcNONE ? NULL : &pbc);
 808         }
 809
 810         /* Communicate the forces */
 811         if (PAR(cr))
 812         {
 813             wallcycle_start(wcycle,ewcMOVEF);
 814             if (DOMAINDECOMP(cr))
 815             {
 816                 dd_move_f(cr->dd,f,fr->fshift);
 817                 /* Do we need to communicate the separate force array
 818                  * for terms that do not contribute to the single sum virial?
 819                  * Position restraints and electric fields do not introduce
 820                  * inter-cg forces, only full electrostatics methods do.
 821                  * When we do not calculate the virial, fr->f_novirsum = f,
 822                  * so we have already communicated these forces.
 823                  */
 824                 if (EEL_FULL(fr->eeltype) && cr->dd->n_intercg_excl &&
 825                     (flags & GMX_FORCE_VIRIAL))
 826                 {
 827                     dd_move_f(cr->dd,fr->f_novirsum,NULL);
 828                 }
 829                 if (bSepLRF)
 830                 {
 831                     /* We should not update the shift forces here,
 832                      * since f_twin is already included in f.
 833                      */
 834                     dd_move_f(cr->dd,fr->f_twin,NULL);
 835                 }
 836             }
 837             else
 838             {
 839                 pd_move_f(cr,f,nrnb);
 840                 if (bSepLRF)
 841                 {
 842                     pd_move_f(cr,fr->f_twin,nrnb);
 843                 }
 844             }
 845             wallcycle_stop(wcycle,ewcMOVEF);
 846         }
 847
 848         /* If we have NoVirSum forces, but we do not calculate the virial,
 849          * we sum fr->f_novirum=f later.
 850          */
 851         if (vsite && !(fr->bF_NoVirSum && !(flags & GMX_FORCE_VIRIAL)))
 852         {
 853             wallcycle_start(wcycle,ewcVSITESPREAD);
 854             spread_vsite_f(fplog,vsite,x,f,fr->fshift,FALSE,NULL,nrnb,
 855                            &top->idef,fr->ePBC,fr->bMolPBC,graph,box,cr);
 856             wallcycle_stop(wcycle,ewcVSITESPREAD);
 857
 858             if (bSepLRF)
 859             {
 860                 wallcycle_start(wcycle,ewcVSITESPREAD);
 861                 spread_vsite_f(fplog,vsite,x,fr->f_twin,NULL,FALSE,NULL,
 862                                nrnb,
 863                                &top->idef,fr->ePBC,fr->bMolPBC,graph,box,cr);
 864                 wallcycle_stop(wcycle,ewcVSITESPREAD);
 865             }
 866         }
 867
 868         if (flags & GMX_FORCE_VIRIAL)
 869         {
 870             /* Calculation of the virial must be done after vsites! */
 871             calc_virial(fplog,mdatoms->start,mdatoms->homenr,x,f,
 872                         vir_force,graph,box,nrnb,fr,inputrec->ePBC);
 873         }
 874     }
 875
 876     enerd->term[F_COM_PULL] = 0;
 877     if (inputrec->ePull == epullUMBRELLA || inputrec->ePull == epullCONST_F)
 878     {
 879         /* Calculate the center of mass forces, this requires communication,
 880          * which is why pull_potential is called close to other communication.
 881          * The virial contribution is calculated directly,
 882          * which is why we call pull_potential after calc_virial.
 883          */
 884         set_pbc(&pbc,inputrec->ePBC,box);
 885         dvdl = 0;
 886         enerd->term[F_COM_PULL] +=
 887             pull_potential(inputrec->ePull,inputrec->pull,mdatoms,&pbc,
 888                            cr,t,lambda,x,f,vir_force,&dvdl);
 889         if (bSepDVDL)
 890         {
 891             fprintf(fplog,sepdvdlformat,"Com pull",enerd->term[F_COM_PULL],dvdl);
 892         }
 893         enerd->dvdl_lin += dvdl;
 894     }
 895
 896     /* Add the forces from enforced rotation potentials (if any) */
 897     if (inputrec->bRot)
 898     {
 899         wallcycle_start(wcycle,ewcROTadd);
 900         enerd->term[F_COM_PULL] += add_rot_forces(inputrec->rot, f, cr,step,t);
 901         wallcycle_stop(wcycle,ewcROTadd);
 902     }
 903
 904     if (PAR(cr) && !(cr->duty & DUTY_PME))
 905     {
 906         cycles_ppdpme = wallcycle_stop(wcycle,ewcPPDURINGPME);
 907         dd_cycles_add(cr->dd,cycles_ppdpme,ddCyclPPduringPME);
 908
 909         /* In case of node-splitting, the PP nodes receive the long-range
 910          * forces, virial and energy from the PME nodes here.
 911          */
 912         wallcycle_start(wcycle,ewcPP_PMEWAITRECVF);
 913         dvdl = 0;
 914         gmx_pme_receive_f(cr,fr->f_novirsum,fr->vir_el_recip,&e,&dvdl,
 915                           &cycles_seppme);
 916         if (bSepDVDL)
 917         {
 918             fprintf(fplog,sepdvdlformat,"PME mesh",e,dvdl);
 919         }
 920         enerd->term[F_COUL_RECIP] += e;
 921         enerd->dvdl_lin += dvdl;
 922         if (wcycle)
 923         {
 924             dd_cycles_add(cr->dd,cycles_seppme,ddCyclPME);
 925         }
 926         wallcycle_stop(wcycle,ewcPP_PMEWAITRECVF);
 927     }
 928
 929     if (bDoForces && fr->bF_NoVirSum)
 930     {
 931         if (vsite)
 932         {
 933             /* Spread the mesh force on virtual sites to the other particles...
 934              * This is parallellized. MPI communication is performed
 935              * if the constructing atoms aren't local.
 936              */
 937             wallcycle_start(wcycle,ewcVSITESPREAD);
 938             spread_vsite_f(fplog,vsite,x,fr->f_novirsum,NULL,
 939                            (flags & GMX_FORCE_VIRIAL),fr->vir_el_recip,
 940                            nrnb,
 941                            &top->idef,fr->ePBC,fr->bMolPBC,graph,box,cr);
 942             wallcycle_stop(wcycle,ewcVSITESPREAD);
 943         }
 944         if (flags & GMX_FORCE_VIRIAL)
 945         {
 946             /* Now add the forces, this is local */
 947             if (fr->bDomDec)
 948             {
 949                 sum_forces(0,fr->f_novirsum_n,f,fr->f_novirsum);
 950             }
 951             else
 952             {
 953                 sum_forces(start,start+homenr,f,fr->f_novirsum);
 954             }
 955             if (EEL_FULL(fr->eeltype))
 956             {
 957                 /* Add the mesh contribution to the virial */
 958                 m_add(vir_force,fr->vir_el_recip,vir_force);
 959             }
 960             if (debug)
 961             {
 962                 pr_rvecs(debug,0,"vir_force",vir_force,DIM);
 963             }
 964         }
 965     }
 966
 967     /* Sum the potential energy terms from group contributions */
 968     sum_epot(&(inputrec->opts),enerd);
 969
 970     if (fr->print_force >= 0 && bDoForces)
 971     {
 972         print_large_forces(stderr,mdatoms,cr,step,fr->print_force,x,f);
 973     }
 974 }
 975
 976 void do_constrain_first(FILE *fplog,gmx_constr_t constr,
 977                         t_inputrec *ir,t_mdatoms *md,
 978                         t_state *state,rvec *f,
 979                         t_graph *graph,t_commrec *cr,t_nrnb *nrnb,
 980                         t_forcerec *fr, gmx_localtop_t *top, tensor shake_vir)
 981 {
 982     int    i,m,start,end;
 983     gmx_large_int_t step;
 984     double mass,tmass,vcm[4];
 985     real   dt=ir->delta_t;
 986     real   dvdlambda;
 987     rvec   *savex;
 988
 989     snew(savex,state->natoms);
 990
 991     start = md->start;
 992     end   = md->homenr + start;
 993
 994     if (debug)
 995         fprintf(debug,"vcm: start=%d, homenr=%d, end=%d\n",
 996                 start,md->homenr,end);
 997     /* Do a first constrain to reset particles... */
 998     step = ir->init_step;
 999     if (fplog)
1000     {
1001         char buf[STEPSTRSIZE];
1002         fprintf(fplog,"\nConstraining the starting coordinates (step %s)\n",
1003                 gmx_step_str(step,buf));
1004     }
1005     dvdlambda = 0;
1006
1007     /* constrain the current position */
1008     constrain(NULL,TRUE,FALSE,constr,&(top->idef),
1009               ir,NULL,cr,step,0,md,
1010               state->x,state->x,NULL,
1011               state->box,state->lambda,&dvdlambda,
1012               NULL,NULL,nrnb,econqCoord,ir->epc==epcMTTK,state->veta,state->veta);
1013     if (EI_VV(ir->eI))
1014     {
1015         /* constrain the inital velocity, and save it */
1016         /* also may be useful if we need the ekin from the halfstep for velocity verlet */
1017         /* might not yet treat veta correctly */
1018         constrain(NULL,TRUE,FALSE,constr,&(top->idef),
1019                   ir,NULL,cr,step,0,md,
1020                   state->x,state->v,state->v,
1021                   state->box,state->lambda,&dvdlambda,
1022                   NULL,NULL,nrnb,econqVeloc,ir->epc==epcMTTK,state->veta,state->veta);
1023     }
1024     /* constrain the inital velocities at t-dt/2 */
1025     if (EI_STATE_VELOCITY(ir->eI) && ir->eI!=eiVV)
1026     {
1027         for(i=start; (i<end); i++)
1028         {
1029             for(m=0; (m<DIM); m++)
1030             {
1031                 /* Reverse the velocity */
1032                 state->v[i][m] = -state->v[i][m];
1033                 /* Store the position at t-dt in buf */
1034                 savex[i][m] = state->x[i][m] + dt*state->v[i][m];
1035             }
1036         }
1037     /* Shake the positions at t=-dt with the positions at t=0
1038      * as reference coordinates.
1039          */
1040         if (fplog)
1041         {
1042             char buf[STEPSTRSIZE];
1043             fprintf(fplog,"\nConstraining the coordinates at t0-dt (step %s)\n",
1044                     gmx_step_str(step,buf));
1045         }
1046         dvdlambda = 0;
1047         constrain(NULL,TRUE,FALSE,constr,&(top->idef),
1048                   ir,NULL,cr,step,-1,md,
1049                   state->x,savex,NULL,
1050                   state->box,state->lambda,&dvdlambda,
1051                   state->v,NULL,nrnb,econqCoord,ir->epc==epcMTTK,state->veta,state->veta);
1052
1053         for(i=start; i<end; i++) {
1054             for(m=0; m<DIM; m++) {
1055                 /* Re-reverse the velocities */
1056                 state->v[i][m] = -state->v[i][m];
1057             }
1058         }
1059     }
1060
1061     for(m=0; (m<4); m++)
1062         vcm[m] = 0;
1063     for(i=start; i<end; i++) {
1064         mass = md->massT[i];
1065         for(m=0; m<DIM; m++) {
1066             vcm[m] += state->v[i][m]*mass;
1067         }
1068         vcm[3] += mass;
1069     }
1070
1071     if (ir->nstcomm != 0 || debug) {
1072         /* Compute the global sum of vcm */
1073         if (debug)
1074             fprintf(debug,"vcm: %8.3f  %8.3f  %8.3f,"
1075                     " total mass = %12.5e\n",vcm[XX],vcm[YY],vcm[ZZ],vcm[3]);
1076         if (PAR(cr))
1077             gmx_sumd(4,vcm,cr);
1078         tmass = vcm[3];
1079         for(m=0; (m<DIM); m++)
1080             vcm[m] /= tmass;
1081         if (debug)
1082             fprintf(debug,"vcm: %8.3f  %8.3f  %8.3f,"
1083                     " total mass = %12.5e\n",vcm[XX],vcm[YY],vcm[ZZ],tmass);
1084         if (ir->nstcomm != 0) {
1085             /* Now we have the velocity of center of mass, let's remove it */
1086             for(i=start; (i<end); i++) {
1087                 for(m=0; (m<DIM); m++)
1088                     state->v[i][m] -= vcm[m];
1089             }
1090
1091         }
1092     }
1093     sfree(savex);
1094 }
1095
1096 void calc_enervirdiff(FILE *fplog,int eDispCorr,t_forcerec *fr)
1097 {
1098   double eners[2],virs[2],enersum,virsum,y0,f,g,h;
1099   double r0,r1,r,rc3,rc9,ea,eb,ec,pa,pb,pc,pd;
1100   double invscale,invscale2,invscale3;
1101   int    ri0,ri1,ri,i,offstart,offset;
1102   real   scale,*vdwtab;
1103
1104   fr->enershiftsix = 0;
1105   fr->enershifttwelve = 0;
1106   fr->enerdiffsix = 0;
1107   fr->enerdifftwelve = 0;
1108   fr->virdiffsix = 0;
1109   fr->virdifftwelve = 0;
1110
1111   if (eDispCorr != edispcNO) {
1112     for(i=0; i<2; i++) {
1113       eners[i] = 0;
1114       virs[i]  = 0;
1115     }
1116     if ((fr->vdwtype == evdwSWITCH) || (fr->vdwtype == evdwSHIFT)) {
1117       if (fr->rvdw_switch == 0)
1118         gmx_fatal(FARGS,
1119                   "With dispersion correction rvdw-switch can not be zero "
1120                   "for vdw-type = %s",evdw_names[fr->vdwtype]);
1121
1122       scale  = fr->nblists[0].tab.scale;
1123       vdwtab = fr->nblists[0].vdwtab;
1124
1125       /* Round the cut-offs to exact table values for precision */
1126       ri0 = floor(fr->rvdw_switch*scale);
1127       ri1 = ceil(fr->rvdw*scale);
1128       r0  = ri0/scale;
1129       r1  = ri1/scale;
1130       rc3 = r0*r0*r0;
1131       rc9  = rc3*rc3*rc3;
1132
1133       if (fr->vdwtype == evdwSHIFT) {
1134         /* Determine the constant energy shift below rvdw_switch */
1135         fr->enershiftsix    = (real)(-1.0/(rc3*rc3)) - vdwtab[8*ri0];
1136         fr->enershifttwelve = (real)( 1.0/(rc9*rc3)) - vdwtab[8*ri0 + 4];
1137       }
1138       /* Add the constant part from 0 to rvdw_switch.
1139        * This integration from 0 to rvdw_switch overcounts the number
1140        * of interactions by 1, as it also counts the self interaction.
1141        * We will correct for this later.
1142        */
1143       eners[0] += 4.0*M_PI*fr->enershiftsix*rc3/3.0;
1144       eners[1] += 4.0*M_PI*fr->enershifttwelve*rc3/3.0;
1145
1146       invscale = 1.0/(scale);
1147       invscale2 = invscale*invscale;
1148       invscale3 = invscale*invscale2;
1149
1150       /* following summation derived from cubic spline definition,
1151         Numerical Recipies in C, second edition, p. 113-116.  Exact
1152         for the cubic spline.  We first calculate the negative of
1153         the energy from rvdw to rvdw_switch, assuming that g(r)=1,
1154         and then add the more standard, abrupt cutoff correction to
1155         that result, yielding the long-range correction for a
1156         switched function.  We perform both the pressure and energy
1157         loops at the same time for simplicity, as the computational
1158         cost is low. */
1159
1160       for (i=0;i<2;i++) {
1161         enersum = 0.0; virsum = 0.0;
1162         if (i==0)
1163           offstart = 0;
1164         else
1165           offstart = 4;
1166         for (ri=ri0; ri<ri1; ri++) {
1167           r = ri*invscale;
1168           ea = invscale3;
1169           eb = 2.0*invscale2*r;
1170           ec = invscale*r*r;
1171
1172           pa = invscale3;
1173           pb = 3.0*invscale2*r;
1174           pc = 3.0*invscale*r*r;
1175           pd = r*r*r;
1176
1177           /* this "8" is from the packing in the vdwtab array - perhaps
1178             should be #define'ed? */
1179           offset = 8*ri + offstart;
1180           y0 = vdwtab[offset];
1181           f = vdwtab[offset+1];
1182           g = vdwtab[offset+2];
1183           h = vdwtab[offset+3];
1184
1185           enersum += y0*(ea/3 + eb/2 + ec) + f*(ea/4 + eb/3 + ec/2)+
1186             g*(ea/5 + eb/4 + ec/3) + h*(ea/6 + eb/5 + ec/4);
1187           virsum  +=  f*(pa/4 + pb/3 + pc/2 + pd) +
1188             2*g*(pa/5 + pb/4 + pc/3 + pd/2) + 3*h*(pa/6 + pb/5 + pc/4 + pd/3);
1189
1190         }
1191         enersum *= 4.0*M_PI;
1192         virsum  *= 4.0*M_PI;
1193         eners[i] -= enersum;
1194         virs[i]  -= virsum;
1195       }
1196
1197       /* now add the correction for rvdw_switch to infinity */
1198       eners[0] += -4.0*M_PI/(3.0*rc3);
1199       eners[1] +=  4.0*M_PI/(9.0*rc9);
1200       virs[0]  +=  8.0*M_PI/rc3;
1201       virs[1]  += -16.0*M_PI/(3.0*rc9);
1202     }
1203     else if ((fr->vdwtype == evdwCUT) || (fr->vdwtype == evdwUSER)) {
1204       if (fr->vdwtype == evdwUSER && fplog)
1205         fprintf(fplog,
1206                 "WARNING: using dispersion correction with user tables\n");
1207       rc3  = fr->rvdw*fr->rvdw*fr->rvdw;
1208       rc9  = rc3*rc3*rc3;
1209       eners[0] += -4.0*M_PI/(3.0*rc3);
1210       eners[1] +=  4.0*M_PI/(9.0*rc9);
1211       virs[0]  +=  8.0*M_PI/rc3;
1212       virs[1]  += -16.0*M_PI/(3.0*rc9);
1213     } else {
1214       gmx_fatal(FARGS,
1215                 "Dispersion correction is not implemented for vdw-type = %s",
1216                 evdw_names[fr->vdwtype]);
1217     }
1218     fr->enerdiffsix    = eners[0];
1219     fr->enerdifftwelve = eners[1];
1220     /* The 0.5 is due to the Gromacs definition of the virial */
1221     fr->virdiffsix     = 0.5*virs[0];
1222     fr->virdifftwelve  = 0.5*virs[1];
1223   }
1224 }
1225
1226 void calc_dispcorr(FILE *fplog,t_inputrec *ir,t_forcerec *fr,
1227                    gmx_large_int_t step,int natoms,
1228                    matrix box,real lambda,tensor pres,tensor virial,
1229                    real *prescorr, real *enercorr, real *dvdlcorr)
1230 {
1231     gmx_bool bCorrAll,bCorrPres;
1232     real dvdlambda,invvol,dens,ninter,avcsix,avctwelve,enerdiff,svir=0,spres=0;
1233     int  m;
1234
1235     *prescorr = 0;
1236     *enercorr = 0;
1237     *dvdlcorr = 0;
1238
1239     clear_mat(virial);
1240     clear_mat(pres);
1241
1242     if (ir->eDispCorr != edispcNO) {
1243         bCorrAll  = (ir->eDispCorr == edispcAllEner ||
1244                      ir->eDispCorr == edispcAllEnerPres);
1245         bCorrPres = (ir->eDispCorr == edispcEnerPres ||
1246                      ir->eDispCorr == edispcAllEnerPres);
1247
1248         invvol = 1/det(box);
1249         if (fr->n_tpi)
1250         {
1251             /* Only correct for the interactions with the inserted molecule */
1252             dens = (natoms - fr->n_tpi)*invvol;
1253             ninter = fr->n_tpi;
1254         }
1255         else
1256         {
1257             dens = natoms*invvol;
1258             ninter = 0.5*natoms;
1259         }
1260
1261         if (ir->efep == efepNO)
1262         {
1263             avcsix    = fr->avcsix[0];
1264             avctwelve = fr->avctwelve[0];
1265         }
1266         else
1267         {
1268             avcsix    = (1 - lambda)*fr->avcsix[0]    + lambda*fr->avcsix[1];
1269             avctwelve = (1 - lambda)*fr->avctwelve[0] + lambda*fr->avctwelve[1];
1270         }
1271
1272         enerdiff = ninter*(dens*fr->enerdiffsix - fr->enershiftsix);
1273         *enercorr += avcsix*enerdiff;
1274         dvdlambda = 0.0;
1275         if (ir->efep != efepNO)
1276         {
1277             dvdlambda += (fr->avcsix[1] - fr->avcsix[0])*enerdiff;
1278         }
1279         if (bCorrAll)
1280         {
1281             enerdiff = ninter*(dens*fr->enerdifftwelve - fr->enershifttwelve);
1282             *enercorr += avctwelve*enerdiff;
1283             if (fr->efep != efepNO)
1284             {
1285                 dvdlambda += (fr->avctwelve[1] - fr->avctwelve[0])*enerdiff;
1286             }
1287         }
1288
1289         if (bCorrPres)
1290         {
1291             svir = ninter*dens*avcsix*fr->virdiffsix/3.0;
1292             if (ir->eDispCorr == edispcAllEnerPres)
1293             {
1294                 svir += ninter*dens*avctwelve*fr->virdifftwelve/3.0;
1295             }
1296             /* The factor 2 is because of the Gromacs virial definition */
1297             spres = -2.0*invvol*svir*PRESFAC;
1298
1299             for(m=0; m<DIM; m++) {
1300                 virial[m][m] += svir;
1301                 pres[m][m] += spres;
1302             }
1303             *prescorr += spres;
1304         }
1305
1306         /* Can't currently control when it prints, for now, just print when degugging */
1307         if (debug)
1308         {
1309             if (bCorrAll) {
1310                 fprintf(debug,"Long Range LJ corr.: <C6> %10.4e, <C12> %10.4e\n",
1311                         avcsix,avctwelve);
1312             }
1313             if (bCorrPres)
1314             {
1315                 fprintf(debug,
1316                         "Long Range LJ corr.: Epot %10g, Pres: %10g, Vir: %10g\n",
1317                         *enercorr,spres,svir);
1318             }
1319             else
1320             {
1321                 fprintf(debug,"Long Range LJ corr.: Epot %10g\n",*enercorr);
1322             }
1323         }
1324
1325         if (fr->bSepDVDL && do_per_step(step,ir->nstlog))
1326         {
1327             fprintf(fplog,sepdvdlformat,"Dispersion correction",
1328                     *enercorr,dvdlambda);
1329         }
1330         if (fr->efep != efepNO)
1331         {
1332             *dvdlcorr += dvdlambda;
1333         }
1334     }
1335 }
1336
1337 void do_pbc_first(FILE *fplog,matrix box,t_forcerec *fr,
1338                   t_graph *graph,rvec x[])
1339 {
1340   if (fplog)
1341     fprintf(fplog,"Removing pbc first time\n");
1342   calc_shifts(box,fr->shift_vec);
1343   if (graph) {
1344     mk_mshift(fplog,graph,fr->ePBC,box,x);
1345     if (gmx_debug_at)
1346       p_graph(debug,"do_pbc_first 1",graph);
1347     shift_self(graph,box,x);
1348     /* By doing an extra mk_mshift the molecules that are broken
1349      * because they were e.g. imported from another software
1350      * will be made whole again. Such are the healing powers
1351      * of GROMACS.
1352      */
1353     mk_mshift(fplog,graph,fr->ePBC,box,x);
1354     if (gmx_debug_at)
1355       p_graph(debug,"do_pbc_first 2",graph);
1356   }
1357   if (fplog)
1358     fprintf(fplog,"Done rmpbc\n");
1359 }
1360
1361 static void low_do_pbc_mtop(FILE *fplog,int ePBC,matrix box,
1362                             gmx_mtop_t *mtop,rvec x[],
1363                             gmx_bool bFirst)
1364 {
1365   t_graph *graph;
1366   int mb,as,mol;
1367   gmx_molblock_t *molb;
1368
1369   if (bFirst && fplog)
1370     fprintf(fplog,"Removing pbc first time\n");
1371
1372   snew(graph,1);
1373   as = 0;
1374   for(mb=0; mb<mtop->nmolblock; mb++) {
1375     molb = &mtop->molblock[mb];
1376     if (molb->natoms_mol == 1 ||
1377         (!bFirst && mtop->moltype[molb->type].cgs.nr == 1)) {
1378       /* Just one atom or charge group in the molecule, no PBC required */
1379       as += molb->nmol*molb->natoms_mol;
1380     } else {
1381       /* Pass NULL iso fplog to avoid graph prints for each molecule type */
1382       mk_graph_ilist(NULL,mtop->moltype[molb->type].ilist,
1383                      0,molb->natoms_mol,FALSE,FALSE,graph);
1384
1385       for(mol=0; mol<molb->nmol; mol++) {
1386         mk_mshift(fplog,graph,ePBC,box,x+as);
1387
1388         shift_self(graph,box,x+as);
1389         /* The molecule is whole now.
1390          * We don't need the second mk_mshift call as in do_pbc_first,
1391          * since we no longer need this graph.
1392          */
1393
1394         as += molb->natoms_mol;
1395       }
1396       done_graph(graph);
1397     }
1398   }
1399   sfree(graph);
1400 }
1401
1402 void do_pbc_first_mtop(FILE *fplog,int ePBC,matrix box,
1403                        gmx_mtop_t *mtop,rvec x[])
1404 {
1405   low_do_pbc_mtop(fplog,ePBC,box,mtop,x,TRUE);
1406 }
1407
1408 void do_pbc_mtop(FILE *fplog,int ePBC,matrix box,
1409                  gmx_mtop_t *mtop,rvec x[])
1410 {
1411   low_do_pbc_mtop(fplog,ePBC,box,mtop,x,FALSE);
1412 }
1413
1414 void finish_run(FILE *fplog,t_commrec *cr,const char *confout,
1415                 t_inputrec *inputrec,
1416                 t_nrnb nrnb[],gmx_wallcycle_t wcycle,
1417                 gmx_runtime_t *runtime,
1418                 gmx_bool bWriteStat)
1419 {
1420   int    i,j;
1421   t_nrnb *nrnb_tot=NULL;
1422   real   delta_t;
1423   double nbfs,mflop;
1424   double cycles[ewcNR];
1425
1426   wallcycle_sum(cr,wcycle,cycles);
1427
1428   if (cr->nnodes > 1) {
1429     if (SIMMASTER(cr))
1430       snew(nrnb_tot,1);
1431 #ifdef GMX_MPI
1432     MPI_Reduce(nrnb->n,nrnb_tot->n,eNRNB,MPI_DOUBLE,MPI_SUM,
1433                MASTERRANK(cr),cr->mpi_comm_mysim);
1434 #endif
1435   } else {
1436     nrnb_tot = nrnb;
1437   }
1438
1439   if (SIMMASTER(cr)) {
1440     print_flop(fplog,nrnb_tot,&nbfs,&mflop);
1441     if (cr->nnodes > 1) {
1442       sfree(nrnb_tot);
1443     }
1444   }
1445
1446   if ((cr->duty & DUTY_PP) && DOMAINDECOMP(cr)) {
1447     print_dd_statistics(cr,inputrec,fplog);
1448   }
1449
1450 #ifdef GMX_MPI
1451     if (PARTDECOMP(cr))
1452     {
1453         if (MASTER(cr))
1454         {
1455             t_nrnb     *nrnb_all;
1456             int        s;
1457             MPI_Status stat;
1458
1459             snew(nrnb_all,cr->nnodes);
1460             nrnb_all[0] = *nrnb;
1461             for(s=1; s<cr->nnodes; s++)
1462             {
1463                 MPI_Recv(nrnb_all[s].n,eNRNB,MPI_DOUBLE,s,0,
1464                          cr->mpi_comm_mysim,&stat);
1465             }
1466             pr_load(fplog,cr,nrnb_all);
1467             sfree(nrnb_all);
1468         }
1469         else
1470         {
1471             MPI_Send(nrnb->n,eNRNB,MPI_DOUBLE,MASTERRANK(cr),0,
1472                      cr->mpi_comm_mysim);
1473         }
1474     }
1475 #endif
1476
1477   if (SIMMASTER(cr)) {
1478     wallcycle_print(fplog,cr->nnodes,cr->npmenodes,runtime->realtime,
1479                     wcycle,cycles);
1480
1481     if (EI_DYNAMICS(inputrec->eI)) {
1482       delta_t = inputrec->delta_t;
1483     } else {
1484       delta_t = 0;
1485     }
1486
1487     if (fplog) {
1488         print_perf(fplog,runtime->proctime,runtime->realtime,
1489                    cr->nnodes-cr->npmenodes,
1490                    runtime->nsteps_done,delta_t,nbfs,mflop);
1491     }
1492     if (bWriteStat) {
1493         print_perf(stderr,runtime->proctime,runtime->realtime,
1494                    cr->nnodes-cr->npmenodes,
1495                    runtime->nsteps_done,delta_t,nbfs,mflop);
1496     }
1497
1498     /*
1499     runtime=inputrec->nsteps*inputrec->delta_t;
1500     if (bWriteStat) {
1501       if (cr->nnodes == 1)
1502         fprintf(stderr,"\n\n");
1503       print_perf(stderr,nodetime,realtime,runtime,&ntot,
1504                  cr->nnodes-cr->npmenodes,FALSE);
1505     }
1506     wallcycle_print(fplog,cr->nnodes,cr->npmenodes,realtime,wcycle,cycles);
1507     print_perf(fplog,nodetime,realtime,runtime,&ntot,cr->nnodes-cr->npmenodes,
1508                TRUE);
1509     if (PARTDECOMP(cr))
1510       pr_load(fplog,cr,nrnb_all);
1511     if (cr->nnodes > 1)
1512       sfree(nrnb_all);
1513     */
1514   }
1515 }
1516
1517 void init_md(FILE *fplog,
1518              t_commrec *cr,t_inputrec *ir,const output_env_t oenv,
1519              double *t,double *t0,
1520              real *lambda,double *lam0,
1521              t_nrnb *nrnb,gmx_mtop_t *mtop,
1522              gmx_update_t *upd,
1523              int nfile,const t_filenm fnm[],
1524              gmx_mdoutf_t **outf,t_mdebin **mdebin,
1525              tensor force_vir,tensor shake_vir,rvec mu_tot,
1526              gmx_bool *bSimAnn,t_vcm **vcm, t_state *state, unsigned long Flags)
1527 {
1528     int  i,j,n;
1529     real tmpt,mod;
1530
1531     /* Initial values */
1532     *t = *t0       = ir->init_t;
1533     if (ir->efep != efepNO)
1534     {
1535         *lam0 = ir->init_lambda;
1536         *lambda = *lam0 + ir->init_step*ir->delta_lambda;
1537     }
1538     else
1539     {
1540         *lambda = *lam0   = 0.0;
1541     }
1542
1543     *bSimAnn=FALSE;
1544     for(i=0;i<ir->opts.ngtc;i++)
1545     {
1546         /* set bSimAnn if any group is being annealed */
1547         if(ir->opts.annealing[i]!=eannNO)
1548         {
1549             *bSimAnn = TRUE;
1550         }
1551     }
1552     if (*bSimAnn)
1553     {
1554         update_annealing_target_temp(&(ir->opts),ir->init_t);
1555     }
1556
1557     if (upd)
1558     {
1559         *upd = init_update(fplog,ir);
1560     }
1561
1562     if (vcm != NULL)
1563     {
1564         *vcm = init_vcm(fplog,&mtop->groups,ir);
1565     }
1566
1567     if (EI_DYNAMICS(ir->eI) && !(Flags & MD_APPENDFILES))
1568     {
1569         if (ir->etc == etcBERENDSEN)
1570         {
1571             please_cite(fplog,"Berendsen84a");
1572         }
1573         if (ir->etc == etcVRESCALE)
1574         {
1575             please_cite(fplog,"Bussi2007a");
1576         }
1577     }
1578
1579     init_nrnb(nrnb);
1580
1581     if (nfile != -1)
1582     {
1583         *outf = init_mdoutf(nfile,fnm,Flags,cr,ir,oenv);
1584
1585         *mdebin = init_mdebin((Flags & MD_APPENDFILES) ? NULL : (*outf)->fp_ene,
1586                               mtop,ir, (*outf)->fp_dhdl);
1587     }
1588
1589     if (ir->bAdress)
1590     {
1591       please_cite(fplog,"Fritsch12");
1592       please_cite(fplog,"Junghans10");
1593     }
1594     /* Initiate variables */
1595     clear_mat(force_vir);
1596     clear_mat(shake_vir);
1597     clear_rvec(mu_tot);
1598
1599     debug_gmx();
1600 }
1601