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 "sim_util.h"
  67 #include "update.h"
  68 #include "physics.h"
  69 #include "main.h"
  70 #include "mdatoms.h"
  71 #include "force.h"
  72 #include "bondf.h"
  73 #include "pme.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 "gmx_random.h"
  85 #include "mpelogging.h"
  86 #include "domdec.h"
  87 #include "partdec.h"
  88 #include "gmx_wallcycle.h"
  89 #include "genborn.h"
  90 #include "nbnxn_atomdata.h"
  91 #include "nbnxn_search.h"
  92 #include "nbnxn_kernels/nbnxn_kernel_ref.h"
  93 #include "nbnxn_kernels/nbnxn_kernel_x86_simd128.h"
  94 #include "nbnxn_kernels/nbnxn_kernel_x86_simd256.h"
  95 #include "nbnxn_kernels/nbnxn_kernel_gpu_ref.h"
  96
  97 #ifdef GMX_LIB_MPI
  98 #include <mpi.h>
  99 #endif
 100 #ifdef GMX_THREAD_MPI
 101 #include "tmpi.h"
 102 #endif
 103
 104 #include "adress.h"
 105 #include "qmmm.h"
 106
 107 #include "nbnxn_cuda_data_mgmt.h"
 108 #include "nbnxn_cuda/nbnxn_cuda.h"
 109
 110 #if 0
 111 typedef struct gmx_timeprint {
 112
 113 } t_gmx_timeprint;
 114 #endif
 115
 116 /* Portable version of ctime_r implemented in src/gmxlib/string2.c, but we do not want it declared in public installed headers */
 117 char *
 118 gmx_ctime_r(const time_t *clock,char *buf, int n);
 119
 120
 121 double
 122 gmx_gettime()
 123 {
 124 #ifdef HAVE_GETTIMEOFDAY
 125         struct timeval t;
 126         double seconds;
 127
 128         gettimeofday(&t,NULL);
 129
 130         seconds = (double) t.tv_sec + 1e-6*(double)t.tv_usec;
 131
 132         return seconds;
 133 #else
 134         double  seconds;
 135
 136         seconds = time(NULL);
 137
 138         return seconds;
 139 #endif
 140 }
 141
 142
 143 #define difftime(end,start) ((double)(end)-(double)(start))
 144
 145 void print_time(FILE *out,gmx_runtime_t *runtime,gmx_large_int_t step,
 146                 t_inputrec *ir, t_commrec *cr)
 147 {
 148     time_t finish;
 149     char   timebuf[STRLEN];
 150     double dt;
 151     char buf[48];
 152
 153 #ifndef GMX_THREAD_MPI
 154     if (!PAR(cr))
 155 #endif
 156     {
 157         fprintf(out,"\r");
 158     }
 159     fprintf(out,"step %s",gmx_step_str(step,buf));
 160     if ((step >= ir->nstlist))
 161     {
 162         runtime->last = gmx_gettime();
 163         dt = difftime(runtime->last,runtime->real);
 164         runtime->time_per_step = dt/(step - ir->init_step + 1);
 165
 166         dt = (ir->nsteps + ir->init_step - step)*runtime->time_per_step;
 167
 168         if (ir->nsteps >= 0)
 169         {
 170             if (dt >= 300)
 171             {
 172                 finish = (time_t) (runtime->last + dt);
 173                 gmx_ctime_r(&finish,timebuf,STRLEN);
 174                 sprintf(buf,"%s",timebuf);
 175                 buf[strlen(buf)-1]='\0';
 176                 fprintf(out,", will finish %s",buf);
 177             }
 178             else
 179                 fprintf(out,", remaining runtime: %5d s          ",(int)dt);
 180         }
 181         else
 182         {
 183             fprintf(out," performance: %.1f ns/day    ",
 184                     ir->delta_t/1000*24*60*60/runtime->time_per_step);
 185         }
 186     }
 187 #ifndef GMX_THREAD_MPI
 188     if (PAR(cr))
 189     {
 190         fprintf(out,"\n");
 191     }
 192 #endif
 193
 194     fflush(out);
 195 }
 196
 197 #ifdef NO_CLOCK
 198 #define clock() -1
 199 #endif
 200
 201 static double set_proctime(gmx_runtime_t *runtime)
 202 {
 203     double diff;
 204 #ifdef GMX_CRAY_XT3
 205     double prev;
 206
 207     prev = runtime->proc;
 208     runtime->proc = dclock();
 209
 210     diff = runtime->proc - prev;
 211 #else
 212     clock_t prev;
 213
 214     prev = runtime->proc;
 215     runtime->proc = clock();
 216
 217     diff = (double)(runtime->proc - prev)/(double)CLOCKS_PER_SEC;
 218 #endif
 219     if (diff < 0)
 220     {
 221         /* The counter has probably looped, ignore this data */
 222         diff = 0;
 223     }
 224
 225     return diff;
 226 }
 227
 228 void runtime_start(gmx_runtime_t *runtime)
 229 {
 230     runtime->real = gmx_gettime();
 231     runtime->proc          = 0;
 232     set_proctime(runtime);
 233     runtime->realtime      = 0;
 234     runtime->proctime      = 0;
 235     runtime->last          = 0;
 236     runtime->time_per_step = 0;
 237 }
 238
 239 void runtime_end(gmx_runtime_t *runtime)
 240 {
 241     double now;
 242
 243     now = gmx_gettime();
 244
 245     runtime->proctime += set_proctime(runtime);
 246     runtime->realtime  = now - runtime->real;
 247     runtime->real      = now;
 248 }
 249
 250 void runtime_upd_proc(gmx_runtime_t *runtime)
 251 {
 252     runtime->proctime += set_proctime(runtime);
 253 }
 254
 255 void print_date_and_time(FILE *fplog,int nodeid,const char *title,
 256                          const gmx_runtime_t *runtime)
 257 {
 258     int i;
 259     char timebuf[STRLEN];
 260     char time_string[STRLEN];
 261     time_t tmptime;
 262
 263     if (fplog)
 264     {
 265         if (runtime != NULL)
 266         {
 267             tmptime = (time_t) runtime->real;
 268             gmx_ctime_r(&tmptime,timebuf,STRLEN);
 269         }
 270         else
 271         {
 272             tmptime = (time_t) gmx_gettime();
 273             gmx_ctime_r(&tmptime,timebuf,STRLEN);
 274         }
 275         for(i=0; timebuf[i]>=' '; i++)
 276         {
 277             time_string[i]=timebuf[i];
 278         }
 279         time_string[i]='\0';
 280
 281         fprintf(fplog,"%s on node %d %s\n",title,nodeid,time_string);
 282     }
 283 }
 284
 285 static void sum_forces(int start,int end,rvec f[],rvec flr[])
 286 {
 287   int i;
 288
 289   if (gmx_debug_at) {
 290     pr_rvecs(debug,0,"fsr",f+start,end-start);
 291     pr_rvecs(debug,0,"flr",flr+start,end-start);
 292   }
 293   for(i=start; (i<end); i++)
 294     rvec_inc(f[i],flr[i]);
 295 }
 296
 297 /*
 298  * calc_f_el calculates forces due to an electric field.
 299  *
 300  * force is kJ mol^-1 nm^-1 = e * kJ mol^-1 nm^-1 / e
 301  *
 302  * Et[] contains the parameters for the time dependent
 303  * part of the field (not yet used).
 304  * Ex[] contains the parameters for
 305  * the spatial dependent part of the field. You can have cool periodic
 306  * fields in principle, but only a constant field is supported
 307  * now.
 308  * The function should return the energy due to the electric field
 309  * (if any) but for now returns 0.
 310  *
 311  * WARNING:
 312  * There can be problems with the virial.
 313  * Since the field is not self-consistent this is unavoidable.
 314  * For neutral molecules the virial is correct within this approximation.
 315  * For neutral systems with many charged molecules the error is small.
 316  * But for systems with a net charge or a few charged molecules
 317  * the error can be significant when the field is high.
 318  * Solution: implement a self-consitent electric field into PME.
 319  */
 320 static void calc_f_el(FILE *fp,int  start,int homenr,
 321                       real charge[],rvec x[],rvec f[],
 322                       t_cosines Ex[],t_cosines Et[],double t)
 323 {
 324     rvec Ext;
 325     real t0;
 326     int  i,m;
 327
 328     for(m=0; (m<DIM); m++)
 329     {
 330         if (Et[m].n > 0)
 331         {
 332             if (Et[m].n == 3)
 333             {
 334                 t0 = Et[m].a[1];
 335                 Ext[m] = cos(Et[m].a[0]*(t-t0))*exp(-sqr(t-t0)/(2.0*sqr(Et[m].a[2])));
 336             }
 337             else
 338             {
 339                 Ext[m] = cos(Et[m].a[0]*t);
 340             }
 341         }
 342         else
 343         {
 344             Ext[m] = 1.0;
 345         }
 346         if (Ex[m].n > 0)
 347         {
 348             /* Convert the field strength from V/nm to MD-units */
 349             Ext[m] *= Ex[m].a[0]*FIELDFAC;
 350             for(i=start; (i<start+homenr); i++)
 351                 f[i][m] += charge[i]*Ext[m];
 352         }
 353         else
 354         {
 355             Ext[m] = 0;
 356         }
 357     }
 358     if (fp != NULL)
 359     {
 360         fprintf(fp,"%10g  %10g  %10g  %10g #FIELD\n",t,
 361                 Ext[XX]/FIELDFAC,Ext[YY]/FIELDFAC,Ext[ZZ]/FIELDFAC);
 362     }
 363 }
 364
 365 static void calc_virial(FILE *fplog,int start,int homenr,rvec x[],rvec f[],
 366                         tensor vir_part,t_graph *graph,matrix box,
 367                         t_nrnb *nrnb,const t_forcerec *fr,int ePBC)
 368 {
 369   int i,j;
 370   tensor virtest;
 371
 372   /* The short-range virial from surrounding boxes */
 373   clear_mat(vir_part);
 374   calc_vir(fplog,SHIFTS,fr->shift_vec,fr->fshift,vir_part,ePBC==epbcSCREW,box);
 375   inc_nrnb(nrnb,eNR_VIRIAL,SHIFTS);
 376
 377   /* Calculate partial virial, for local atoms only, based on short range.
 378    * Total virial is computed in global_stat, called from do_md
 379    */
 380   f_calc_vir(fplog,start,start+homenr,x,f,vir_part,graph,box);
 381   inc_nrnb(nrnb,eNR_VIRIAL,homenr);
 382
 383   /* Add position restraint contribution */
 384   for(i=0; i<DIM; i++) {
 385     vir_part[i][i] += fr->vir_diag_posres[i];
 386   }
 387
 388   /* Add wall contribution */
 389   for(i=0; i<DIM; i++) {
 390     vir_part[i][ZZ] += fr->vir_wall_z[i];
 391   }
 392
 393   if (debug)
 394     pr_rvecs(debug,0,"vir_part",vir_part,DIM);
 395 }
 396
 397 static void posres_wrapper(FILE *fplog,
 398                            int flags,
 399                            gmx_bool bSepDVDL,
 400                            t_inputrec *ir,
 401                            t_nrnb *nrnb,
 402                            gmx_localtop_t *top,
 403                            matrix box,rvec x[],
 404                            rvec f[],
 405                            gmx_enerdata_t *enerd,
 406                            real *lambda,
 407                            t_forcerec *fr)
 408 {
 409     t_pbc pbc;
 410     real  v,dvdl;
 411     int   i;
 412
 413     /* Position restraints always require full pbc */
 414     set_pbc(&pbc,ir->ePBC,box);
 415     dvdl = 0;
 416     v = posres(top->idef.il[F_POSRES].nr,top->idef.il[F_POSRES].iatoms,
 417                top->idef.iparams_posres,
 418                (const rvec*)x,fr->f_novirsum,fr->vir_diag_posres,
 419                ir->ePBC==epbcNONE ? NULL : &pbc,
 420                lambda[efptRESTRAINT],&dvdl,
 421                fr->rc_scaling,fr->ePBC,fr->posres_com,fr->posres_comB);
 422     if (bSepDVDL)
 423     {
 424         fprintf(fplog,sepdvdlformat,
 425                 interaction_function[F_POSRES].longname,v,dvdl);
 426     }
 427     enerd->term[F_POSRES] += v;
 428     /* If just the force constant changes, the FEP term is linear,
 429      * but if k changes, it is not.
 430      */
 431     enerd->dvdl_nonlin[efptRESTRAINT] += dvdl;
 432     inc_nrnb(nrnb,eNR_POSRES,top->idef.il[F_POSRES].nr/2);
 433
 434     if ((ir->fepvals->n_lambda > 0) && (flags & GMX_FORCE_DHDL))
 435     {
 436         for(i=0; i<enerd->n_lambda; i++)
 437         {
 438             real dvdl_dum,lambda_dum;
 439
 440             lambda_dum = (i==0 ? lambda[efptRESTRAINT] : ir->fepvals->all_lambda[efptRESTRAINT][i-1]);
 441             v = posres(top->idef.il[F_POSRES].nr,top->idef.il[F_POSRES].iatoms,
 442                        top->idef.iparams_posres,
 443                        (const rvec*)x,NULL,NULL,
 444                        ir->ePBC==epbcNONE ? NULL : &pbc,lambda_dum,&dvdl,
 445                        fr->rc_scaling,fr->ePBC,fr->posres_com,fr->posres_comB);
 446             enerd->enerpart_lambda[i] += v;
 447         }
 448     }
 449 }
 450
 451 static void pull_potential_wrapper(FILE *fplog,
 452                                    gmx_bool bSepDVDL,
 453                                    t_commrec *cr,
 454                                    t_inputrec *ir,
 455                                    matrix box,rvec x[],
 456                                    rvec f[],
 457                                    tensor vir_force,
 458                                    t_mdatoms *mdatoms,
 459                                    gmx_enerdata_t *enerd,
 460                                    real *lambda,
 461                                    double t)
 462 {
 463     t_pbc  pbc;
 464     real   dvdl;
 465
 466     /* Calculate the center of mass forces, this requires communication,
 467      * which is why pull_potential is called close to other communication.
 468      * The virial contribution is calculated directly,
 469      * which is why we call pull_potential after calc_virial.
 470      */
 471     set_pbc(&pbc,ir->ePBC,box);
 472     dvdl = 0;
 473     enerd->term[F_COM_PULL] +=
 474         pull_potential(ir->ePull,ir->pull,mdatoms,&pbc,
 475                        cr,t,lambda[efptRESTRAINT],x,f,vir_force,&dvdl);
 476     if (bSepDVDL)
 477     {
 478         fprintf(fplog,sepdvdlformat,"Com pull",enerd->term[F_COM_PULL],dvdl);
 479     }
 480     enerd->dvdl_lin[efptRESTRAINT] += dvdl;
 481 }
 482
 483 static void pme_receive_force_ener(FILE *fplog,
 484                                    gmx_bool bSepDVDL,
 485                                    t_commrec *cr,
 486                                    gmx_wallcycle_t wcycle,
 487                                    gmx_enerdata_t *enerd,
 488                                    t_forcerec *fr)
 489 {
 490     real   e,v,dvdl;
 491     float  cycles_ppdpme,cycles_seppme;
 492
 493     cycles_ppdpme = wallcycle_stop(wcycle,ewcPPDURINGPME);
 494     dd_cycles_add(cr->dd,cycles_ppdpme,ddCyclPPduringPME);
 495
 496     /* In case of node-splitting, the PP nodes receive the long-range
 497      * forces, virial and energy from the PME nodes here.
 498      */
 499     wallcycle_start(wcycle,ewcPP_PMEWAITRECVF);
 500     dvdl = 0;
 501     gmx_pme_receive_f(cr,fr->f_novirsum,fr->vir_el_recip,&e,&dvdl,
 502                       &cycles_seppme);
 503     if (bSepDVDL)
 504     {
 505         fprintf(fplog,sepdvdlformat,"PME mesh",e,dvdl);
 506     }
 507     enerd->term[F_COUL_RECIP] += e;
 508     enerd->dvdl_lin[efptCOUL] += dvdl;
 509     if (wcycle)
 510     {
 511         dd_cycles_add(cr->dd,cycles_seppme,ddCyclPME);
 512     }
 513     wallcycle_stop(wcycle,ewcPP_PMEWAITRECVF);
 514 }
 515
 516 static void print_large_forces(FILE *fp,t_mdatoms *md,t_commrec *cr,
 517                                gmx_large_int_t step,real pforce,rvec *x,rvec *f)
 518 {
 519   int  i;
 520   real pf2,fn2;
 521   char buf[STEPSTRSIZE];
 522
 523   pf2 = sqr(pforce);
 524   for(i=md->start; i<md->start+md->homenr; i++) {
 525     fn2 = norm2(f[i]);
 526     /* We also catch NAN, if the compiler does not optimize this away. */
 527     if (fn2 >= pf2 || fn2 != fn2) {
 528       fprintf(fp,"step %s  atom %6d  x %8.3f %8.3f %8.3f  force %12.5e\n",
 529               gmx_step_str(step,buf),
 530               ddglatnr(cr->dd,i),x[i][XX],x[i][YY],x[i][ZZ],sqrt(fn2));
 531     }
 532   }
 533 }
 534
 535 static void post_process_forces(FILE *fplog,
 536                                 t_commrec *cr,
 537                                 gmx_large_int_t step,
 538                                 t_nrnb *nrnb,gmx_wallcycle_t wcycle,
 539                                 gmx_localtop_t *top,
 540                                 matrix box,rvec x[],
 541                                 rvec f[],
 542                                 tensor vir_force,
 543                                 t_mdatoms *mdatoms,
 544                                 t_graph *graph,
 545                                 t_forcerec *fr,gmx_vsite_t *vsite,
 546                                 int flags)
 547 {
 548     if (fr->bF_NoVirSum)
 549     {
 550         if (vsite)
 551         {
 552             /* Spread the mesh force on virtual sites to the other particles...
 553              * This is parallellized. MPI communication is performed
 554              * if the constructing atoms aren't local.
 555              */
 556             wallcycle_start(wcycle,ewcVSITESPREAD);
 557             spread_vsite_f(fplog,vsite,x,fr->f_novirsum,NULL,
 558                            (flags & GMX_FORCE_VIRIAL),fr->vir_el_recip,
 559                            nrnb,
 560                            &top->idef,fr->ePBC,fr->bMolPBC,graph,box,cr);
 561             wallcycle_stop(wcycle,ewcVSITESPREAD);
 562         }
 563         if (flags & GMX_FORCE_VIRIAL)
 564         {
 565             /* Now add the forces, this is local */
 566             if (fr->bDomDec)
 567             {
 568                 sum_forces(0,fr->f_novirsum_n,f,fr->f_novirsum);
 569             }
 570             else
 571             {
 572                 sum_forces(mdatoms->start,mdatoms->start+mdatoms->homenr,
 573                            f,fr->f_novirsum);
 574             }
 575             if (EEL_FULL(fr->eeltype))
 576             {
 577                 /* Add the mesh contribution to the virial */
 578                 m_add(vir_force,fr->vir_el_recip,vir_force);
 579             }
 580             if (debug)
 581             {
 582                 pr_rvecs(debug,0,"vir_force",vir_force,DIM);
 583             }
 584         }
 585     }
 586
 587     if (fr->print_force >= 0)
 588     {
 589         print_large_forces(stderr,mdatoms,cr,step,fr->print_force,x,f);
 590     }
 591 }
 592
 593 static void do_nb_verlet(t_forcerec *fr,
 594                          interaction_const_t *ic,
 595                          gmx_enerdata_t *enerd,
 596                          int flags, int ilocality,
 597                          int clearF,
 598                          t_nrnb *nrnb,
 599                          gmx_wallcycle_t wcycle)
 600 {
 601     int     nnbl, kernel_type, sh_e;
 602     char    *env;
 603     nonbonded_verlet_group_t  *nbvg;
 604
 605     if (!(flags & GMX_FORCE_NONBONDED))
 606     {
 607         /* skip non-bonded calculation */
 608         return;
 609     }
 610
 611     nbvg = &fr->nbv->grp[ilocality];
 612
 613     /* CUDA kernel launch overhead is already timed separately */
 614     if (fr->cutoff_scheme != ecutsVERLET)
 615     {
 616         gmx_incons("Invalid cut-off scheme passed!");
 617     }
 618
 619     if (nbvg->kernel_type != nbk8x8x8_CUDA)
 620     {
 621         wallcycle_sub_start(wcycle, ewcsNONBONDED);
 622     }
 623     switch (nbvg->kernel_type)
 624     {
 625         case nbk4x4_PlainC:
 626             nbnxn_kernel_ref(&nbvg->nbl_lists,
 627                              nbvg->nbat, ic,
 628                              fr->shift_vec,
 629                              flags,
 630                              clearF,
 631                              fr->fshift[0],
 632                              enerd->grpp.ener[egCOULSR],
 633                              fr->bBHAM ?
 634                              enerd->grpp.ener[egBHAMSR] :
 635                              enerd->grpp.ener[egLJSR]);
 636             break;
 637
 638         case nbk4xN_X86_SIMD128:
 639             nbnxn_kernel_x86_simd128(&nbvg->nbl_lists,
 640                                      nbvg->nbat, ic,
 641                                      fr->shift_vec,
 642                                      flags,
 643                                      clearF,
 644                                      fr->fshift[0],
 645                                      enerd->grpp.ener[egCOULSR],
 646                                      fr->bBHAM ?
 647                                      enerd->grpp.ener[egBHAMSR] :
 648                                      enerd->grpp.ener[egLJSR]);
 649             break;
 650         case nbk4xN_X86_SIMD256:
 651             nbnxn_kernel_x86_simd256(&nbvg->nbl_lists,
 652                                      nbvg->nbat, ic,
 653                                      fr->shift_vec,
 654                                      flags,
 655                                      clearF,
 656                                      fr->fshift[0],
 657                                      enerd->grpp.ener[egCOULSR],
 658                                      fr->bBHAM ?
 659                                      enerd->grpp.ener[egBHAMSR] :
 660                                      enerd->grpp.ener[egLJSR]);
 661             break;
 662
 663         case nbk8x8x8_CUDA:
 664             nbnxn_cuda_launch_kernel(fr->nbv->cu_nbv, nbvg->nbat, flags, ilocality);
 665             break;
 666
 667         case nbk8x8x8_PlainC:
 668             nbnxn_kernel_gpu_ref(nbvg->nbl_lists.nbl[0],
 669                                  nbvg->nbat, ic,
 670                                  fr->shift_vec,
 671                                  flags,
 672                                  clearF,
 673                                  nbvg->nbat->out[0].f,
 674                                  fr->fshift[0],
 675                                  enerd->grpp.ener[egCOULSR],
 676                                  fr->bBHAM ?
 677                                  enerd->grpp.ener[egBHAMSR] :
 678                                  enerd->grpp.ener[egLJSR]);
 679             break;
 680
 681         default:
 682             gmx_incons("Invalid nonbonded kernel type passed!");
 683
 684     }
 685     if (nbvg->kernel_type != nbk8x8x8_CUDA)
 686     {
 687         wallcycle_sub_stop(wcycle, ewcsNONBONDED);
 688     }
 689
 690     /* In eNR_??? the nbnxn F+E kernels are always the F kernel + 1 */
 691     sh_e = ((flags & GMX_FORCE_ENERGY) ? 1 : 0);
 692     inc_nrnb(nrnb,
 693              ((EEL_RF(ic->eeltype) || ic->eeltype == eelCUT) ?
 694               eNR_NBNXN_LJ_RF : eNR_NBNXN_LJ_TAB) + sh_e,
 695              nbvg->nbl_lists.natpair_ljq);
 696     inc_nrnb(nrnb,eNR_NBNXN_LJ+sh_e,nbvg->nbl_lists.natpair_lj);
 697     inc_nrnb(nrnb,
 698              ((EEL_RF(ic->eeltype) || ic->eeltype == eelCUT) ?
 699               eNR_NBNXN_RF : eNR_NBNXN_TAB)+sh_e,
 700              nbvg->nbl_lists.natpair_q);
 701 }
 702
 703 void do_force_cutsVERLET(FILE *fplog,t_commrec *cr,
 704               t_inputrec *inputrec,
 705               gmx_large_int_t step,t_nrnb *nrnb,gmx_wallcycle_t wcycle,
 706               gmx_localtop_t *top,
 707               gmx_mtop_t *mtop,
 708               gmx_groups_t *groups,
 709               matrix box,rvec x[],history_t *hist,
 710               rvec f[],
 711               tensor vir_force,
 712               t_mdatoms *mdatoms,
 713               gmx_enerdata_t *enerd,t_fcdata *fcd,
 714               real *lambda,t_graph *graph,
 715               t_forcerec *fr, interaction_const_t *ic,
 716               gmx_vsite_t *vsite,rvec mu_tot,
 717               double t,FILE *field,gmx_edsam_t ed,
 718               gmx_bool bBornRadii,
 719               int flags)
 720 {
 721     int     cg0,cg1,i,j;
 722     int     start,homenr;
 723     int     nb_kernel_type;
 724     double  mu[2*DIM];
 725     gmx_bool   bSepDVDL,bStateChanged,bNS,bFillGrid,bCalcCGCM,bBS;
 726     gmx_bool   bDoLongRange,bDoForces,bSepLRF,bUseGPU,bUseOrEmulGPU;
 727     gmx_bool   bDiffKernels=FALSE;
 728     matrix  boxs;
 729     rvec    vzero,box_diag;
 730     real    e,v,dvdl;
 731     float  cycles_pme,cycles_force;
 732     nonbonded_verlet_t *nbv;
 733
 734     cycles_force = 0;
 735     nbv = fr->nbv;
 736     nb_kernel_type = fr->nbv->grp[0].kernel_type;
 737
 738     start  = mdatoms->start;
 739     homenr = mdatoms->homenr;
 740
 741     bSepDVDL = (fr->bSepDVDL && do_per_step(step,inputrec->nstlog));
 742
 743     clear_mat(vir_force);
 744
 745     cg0 = 0;
 746     if (DOMAINDECOMP(cr))
 747     {
 748         cg1 = cr->dd->ncg_tot;
 749     }
 750     else
 751     {
 752         cg1 = top->cgs.nr;
 753     }
 754     if (fr->n_tpi > 0)
 755     {
 756         cg1--;
 757     }
 758
 759     bStateChanged = (flags & GMX_FORCE_STATECHANGED);
 760     bNS           = (flags & GMX_FORCE_NS) && (fr->bAllvsAll==FALSE);
 761     bFillGrid     = (bNS && bStateChanged);
 762     bCalcCGCM     = (bFillGrid && !DOMAINDECOMP(cr));
 763     bDoLongRange  = (fr->bTwinRange && bNS && (flags & GMX_FORCE_DO_LR));
 764     bDoForces     = (flags & GMX_FORCE_FORCES);
 765     bSepLRF       = (bDoLongRange && bDoForces && (flags & GMX_FORCE_SEPLRF));
 766     bUseGPU       = fr->nbv->bUseGPU;
 767     bUseOrEmulGPU = bUseGPU || (nbv->grp[0].kernel_type == nbk8x8x8_PlainC);
 768
 769     if (bStateChanged)
 770     {
 771         update_forcerec(fplog,fr,box);
 772
 773         if (NEED_MUTOT(*inputrec))
 774         {
 775             /* Calculate total (local) dipole moment in a temporary common array.
 776              * This makes it possible to sum them over nodes faster.
 777              */
 778             calc_mu(start,homenr,
 779                     x,mdatoms->chargeA,mdatoms->chargeB,mdatoms->nChargePerturbed,
 780                     mu,mu+DIM);
 781         }
 782     }
 783
 784     if (fr->ePBC != epbcNONE) {
 785         /* Compute shift vectors every step,
 786          * because of pressure coupling or box deformation!
 787          */
 788         if ((flags & GMX_FORCE_DYNAMICBOX) && bStateChanged)
 789             calc_shifts(box,fr->shift_vec);
 790
 791         if (bCalcCGCM) {
 792             put_atoms_in_box_omp(fr->ePBC,box,homenr,x);
 793             inc_nrnb(nrnb,eNR_SHIFTX,homenr);
 794         }
 795         else if (EI_ENERGY_MINIMIZATION(inputrec->eI) && graph) {
 796             unshift_self(graph,box,x);
 797         }
 798     }
 799
 800     nbnxn_atomdata_copy_shiftvec(flags & GMX_FORCE_DYNAMICBOX,
 801                                   fr->shift_vec,nbv->grp[0].nbat);
 802
 803 #ifdef GMX_MPI
 804     if (!(cr->duty & DUTY_PME)) {
 805         /* Send particle coordinates to the pme nodes.
 806          * Since this is only implemented for domain decomposition
 807          * and domain decomposition does not use the graph,
 808          * we do not need to worry about shifting.
 809          */
 810
 811         wallcycle_start(wcycle,ewcPP_PMESENDX);
 812         GMX_MPE_LOG(ev_send_coordinates_start);
 813
 814         bBS = (inputrec->nwall == 2);
 815         if (bBS) {
 816             copy_mat(box,boxs);
 817             svmul(inputrec->wall_ewald_zfac,boxs[ZZ],boxs[ZZ]);
 818         }
 819
 820         gmx_pme_send_x(cr,bBS ? boxs : box,x,
 821                        mdatoms->nChargePerturbed,lambda[efptCOUL],
 822                        (flags & (GMX_FORCE_VIRIAL | GMX_FORCE_ENERGY)),step);
 823
 824         GMX_MPE_LOG(ev_send_coordinates_finish);
 825         wallcycle_stop(wcycle,ewcPP_PMESENDX);
 826     }
 827 #endif /* GMX_MPI */
 828
 829     /* do gridding for pair search */
 830     if (bNS)
 831     {
 832         if (graph && bStateChanged)
 833         {
 834             /* Calculate intramolecular shift vectors to make molecules whole */
 835             mk_mshift(fplog,graph,fr->ePBC,box,x);
 836         }
 837
 838         clear_rvec(vzero);
 839         box_diag[XX] = box[XX][XX];
 840         box_diag[YY] = box[YY][YY];
 841         box_diag[ZZ] = box[ZZ][ZZ];
 842
 843         wallcycle_start(wcycle,ewcNS);
 844         if (!fr->bDomDec)
 845         {
 846             wallcycle_sub_start(wcycle,ewcsNBS_GRID_LOCAL);
 847             nbnxn_put_on_grid(nbv->nbs,fr->ePBC,box,
 848                               0,vzero,box_diag,
 849                               0,mdatoms->homenr,-1,fr->cginfo,x,
 850                               0,NULL,
 851                               nbv->grp[eintLocal].kernel_type,
 852                               nbv->grp[eintLocal].nbat);
 853             wallcycle_sub_stop(wcycle,ewcsNBS_GRID_LOCAL);
 854         }
 855         else
 856         {
 857             wallcycle_sub_start(wcycle,ewcsNBS_GRID_NONLOCAL);
 858             nbnxn_put_on_grid_nonlocal(nbv->nbs,domdec_zones(cr->dd),
 859                                        fr->cginfo,x,
 860                                        nbv->grp[eintNonlocal].kernel_type,
 861                                        nbv->grp[eintNonlocal].nbat);
 862             wallcycle_sub_stop(wcycle,ewcsNBS_GRID_NONLOCAL);
 863         }
 864
 865         if (nbv->ngrp == 1 ||
 866             nbv->grp[eintNonlocal].nbat == nbv->grp[eintLocal].nbat)
 867         {
 868             nbnxn_atomdata_set(nbv->grp[eintLocal].nbat,eatAll,
 869                                 nbv->nbs,mdatoms,fr->cginfo);
 870         }
 871         else
 872         {
 873             nbnxn_atomdata_set(nbv->grp[eintLocal].nbat,eatLocal,
 874                                 nbv->nbs,mdatoms,fr->cginfo);
 875             nbnxn_atomdata_set(nbv->grp[eintNonlocal].nbat,eatAll,
 876                                 nbv->nbs,mdatoms,fr->cginfo);
 877         }
 878         wallcycle_stop(wcycle, ewcNS);
 879     }
 880
 881     /* initialize the GPU atom data and copy shift vector */
 882     if (bUseGPU)
 883     {
 884         if (bNS)
 885         {
 886             wallcycle_start_nocount(wcycle, ewcLAUNCH_GPU_NB);
 887             nbnxn_cuda_init_atomdata(nbv->cu_nbv, nbv->grp[eintLocal].nbat);
 888             wallcycle_stop(wcycle, ewcLAUNCH_GPU_NB);
 889         }
 890
 891         wallcycle_start_nocount(wcycle, ewcLAUNCH_GPU_NB);
 892         nbnxn_cuda_upload_shiftvec(nbv->cu_nbv, nbv->grp[eintLocal].nbat);
 893         wallcycle_stop(wcycle, ewcLAUNCH_GPU_NB);
 894     }
 895
 896     /* do local pair search */
 897     if (bNS)
 898     {
 899         wallcycle_start_nocount(wcycle,ewcNS);
 900         wallcycle_sub_start(wcycle,ewcsNBS_SEARCH_LOCAL);
 901         nbnxn_make_pairlist(nbv->nbs,nbv->grp[eintLocal].nbat,
 902                             &top->excls,
 903                             ic->rlist,
 904                             nbv->min_ci_balanced,
 905                             &nbv->grp[eintLocal].nbl_lists,
 906                             eintLocal,
 907                             nbv->grp[eintLocal].kernel_type,
 908                             nrnb);
 909         wallcycle_sub_stop(wcycle,ewcsNBS_SEARCH_LOCAL);
 910
 911         if (bUseGPU)
 912         {
 913             /* initialize local pair-list on the GPU */
 914             nbnxn_cuda_init_pairlist(nbv->cu_nbv,
 915                                      nbv->grp[eintLocal].nbl_lists.nbl[0],
 916                                      eintLocal);
 917         }
 918         wallcycle_stop(wcycle, ewcNS);
 919     }
 920     else
 921     {
 922         wallcycle_start(wcycle, ewcNB_XF_BUF_OPS);
 923         wallcycle_sub_start(wcycle, ewcsNB_X_BUF_OPS);
 924         nbnxn_atomdata_copy_x_to_nbat_x(nbv->nbs,eatLocal,FALSE,x,
 925                                         nbv->grp[eintLocal].nbat);
 926         wallcycle_sub_stop(wcycle, ewcsNB_X_BUF_OPS);
 927         wallcycle_stop(wcycle, ewcNB_XF_BUF_OPS);
 928     }
 929
 930     if (bUseGPU)
 931     {
 932         wallcycle_start(wcycle,ewcLAUNCH_GPU_NB);
 933         /* launch local nonbonded F on GPU */
 934         do_nb_verlet(fr, ic, enerd, flags, eintLocal, enbvClearFNo,
 935                      nrnb, wcycle);
 936         wallcycle_stop(wcycle,ewcLAUNCH_GPU_NB);
 937     }
 938
 939     /* Communicate coordinates and sum dipole if necessary +
 940        do non-local pair search */
 941     if (DOMAINDECOMP(cr))
 942     {
 943         bDiffKernels = (nbv->grp[eintNonlocal].kernel_type !=
 944                         nbv->grp[eintLocal].kernel_type);
 945
 946         if (bDiffKernels)
 947         {
 948             /* With GPU+CPU non-bonded calculations we need to copy
 949              * the local coordinates to the non-local nbat struct
 950              * (in CPU format) as the non-local kernel call also
 951              * calculates the local - non-local interactions.
 952              */
 953             wallcycle_start(wcycle, ewcNB_XF_BUF_OPS);
 954             wallcycle_sub_start(wcycle, ewcsNB_X_BUF_OPS);
 955             nbnxn_atomdata_copy_x_to_nbat_x(nbv->nbs,eatLocal,TRUE,x,
 956                                              nbv->grp[eintNonlocal].nbat);
 957             wallcycle_sub_stop(wcycle, ewcsNB_X_BUF_OPS);
 958             wallcycle_stop(wcycle, ewcNB_XF_BUF_OPS);
 959         }
 960
 961         if (bNS)
 962         {
 963             wallcycle_start_nocount(wcycle,ewcNS);
 964             wallcycle_sub_start(wcycle,ewcsNBS_SEARCH_NONLOCAL);
 965
 966             if (bDiffKernels)
 967             {
 968                 nbnxn_grid_add_simple(nbv->nbs,nbv->grp[eintNonlocal].nbat);
 969             }
 970
 971             nbnxn_make_pairlist(nbv->nbs,nbv->grp[eintNonlocal].nbat,
 972                                 &top->excls,
 973                                 ic->rlist,
 974                                 nbv->min_ci_balanced,
 975                                 &nbv->grp[eintNonlocal].nbl_lists,
 976                                 eintNonlocal,
 977                                 nbv->grp[eintNonlocal].kernel_type,
 978                                 nrnb);
 979
 980             wallcycle_sub_stop(wcycle,ewcsNBS_SEARCH_NONLOCAL);
 981
 982             if (nbv->grp[eintNonlocal].kernel_type == nbk8x8x8_CUDA)
 983             {
 984                 /* initialize non-local pair-list on the GPU */
 985                 nbnxn_cuda_init_pairlist(nbv->cu_nbv,
 986                                          nbv->grp[eintNonlocal].nbl_lists.nbl[0],
 987                                          eintNonlocal);
 988             }
 989             wallcycle_stop(wcycle,ewcNS);
 990         }
 991         else
 992         {
 993             wallcycle_start(wcycle,ewcMOVEX);
 994             dd_move_x(cr->dd,box,x);
 995
 996             /* When we don't need the total dipole we sum it in global_stat */
 997             if (bStateChanged && NEED_MUTOT(*inputrec))
 998             {
 999                 gmx_sumd(2*DIM,mu,cr);
1000             }
1001             wallcycle_stop(wcycle,ewcMOVEX);
1002
1003             wallcycle_start(wcycle, ewcNB_XF_BUF_OPS);
1004             wallcycle_sub_start(wcycle, ewcsNB_X_BUF_OPS);
1005             nbnxn_atomdata_copy_x_to_nbat_x(nbv->nbs,eatNonlocal,FALSE,x,
1006                                             nbv->grp[eintNonlocal].nbat);
1007             wallcycle_sub_stop(wcycle, ewcsNB_X_BUF_OPS);
1008             cycles_force += wallcycle_stop(wcycle, ewcNB_XF_BUF_OPS);
1009         }
1010
1011         if (bUseGPU && !bDiffKernels)
1012         {
1013             wallcycle_start(wcycle,ewcLAUNCH_GPU_NB);
1014             /* launch non-local nonbonded F on GPU */
1015             do_nb_verlet(fr, ic, enerd, flags, eintNonlocal, enbvClearFNo,
1016                          nrnb, wcycle);
1017             cycles_force += wallcycle_stop(wcycle,ewcLAUNCH_GPU_NB);
1018         }
1019     }
1020
1021     if (bUseGPU)
1022     {
1023         /* launch D2H copy-back F */
1024         wallcycle_start_nocount(wcycle, ewcLAUNCH_GPU_NB);
1025         if (DOMAINDECOMP(cr) && !bDiffKernels)
1026         {
1027             nbnxn_cuda_launch_cpyback(nbv->cu_nbv, nbv->grp[eintNonlocal].nbat,
1028                                       flags, eatNonlocal);
1029         }
1030         nbnxn_cuda_launch_cpyback(nbv->cu_nbv, nbv->grp[eintLocal].nbat,
1031                                   flags, eatLocal);
1032         cycles_force += wallcycle_stop(wcycle,ewcLAUNCH_GPU_NB);
1033     }
1034
1035     if (bStateChanged && NEED_MUTOT(*inputrec))
1036     {
1037         if (PAR(cr))
1038         {
1039             gmx_sumd(2*DIM,mu,cr);
1040         }
1041
1042         for(i=0; i<2; i++)
1043         {
1044             for(j=0;j<DIM;j++)
1045             {
1046                 fr->mu_tot[i][j] = mu[i*DIM + j];
1047             }
1048         }
1049     }
1050     if (fr->efep == efepNO)
1051     {
1052         copy_rvec(fr->mu_tot[0],mu_tot);
1053     }
1054     else
1055     {
1056         for(j=0; j<DIM; j++)
1057         {
1058             mu_tot[j] =
1059                 (1.0 - lambda[efptCOUL])*fr->mu_tot[0][j] +
1060                 lambda[efptCOUL]*fr->mu_tot[1][j];
1061         }
1062     }
1063
1064     /* Reset energies */
1065     reset_enerdata(&(inputrec->opts),fr,bNS,enerd,MASTER(cr));
1066     clear_rvecs(SHIFTS,fr->fshift);
1067
1068     if (DOMAINDECOMP(cr))
1069     {
1070         if (!(cr->duty & DUTY_PME))
1071         {
1072             wallcycle_start(wcycle,ewcPPDURINGPME);
1073             dd_force_flop_start(cr->dd,nrnb);
1074         }
1075     }
1076
1077     /* Start the force cycle counter.
1078      * This counter is stopped in do_forcelow_level.
1079      * No parallel communication should occur while this counter is running,
1080      * since that will interfere with the dynamic load balancing.
1081      */
1082     wallcycle_start(wcycle,ewcFORCE);
1083     if (bDoForces)
1084     {
1085         /* Reset forces for which the virial is calculated separately:
1086          * PME/Ewald forces if necessary */
1087         if (fr->bF_NoVirSum)
1088         {
1089             if (flags & GMX_FORCE_VIRIAL)
1090             {
1091                 fr->f_novirsum = fr->f_novirsum_alloc;
1092                 GMX_BARRIER(cr->mpi_comm_mygroup);
1093                 if (fr->bDomDec)
1094                 {
1095                     clear_rvecs(fr->f_novirsum_n,fr->f_novirsum);
1096                 }
1097                 else
1098                 {
1099                     clear_rvecs(homenr,fr->f_novirsum+start);
1100                 }
1101                 GMX_BARRIER(cr->mpi_comm_mygroup);
1102             }
1103             else
1104             {
1105                 /* We are not calculating the pressure so we do not need
1106                  * a separate array for forces that do not contribute
1107                  * to the pressure.
1108                  */
1109                 fr->f_novirsum = f;
1110             }
1111         }
1112
1113         /* Clear the short- and long-range forces */
1114         clear_rvecs(fr->natoms_force_constr,f);
1115         if(bSepLRF && do_per_step(step,inputrec->nstcalclr))
1116         {
1117             clear_rvecs(fr->natoms_force_constr,fr->f_twin);
1118         }
1119
1120         clear_rvec(fr->vir_diag_posres);
1121
1122         GMX_BARRIER(cr->mpi_comm_mygroup);
1123     }
1124     if (inputrec->ePull == epullCONSTRAINT)
1125     {
1126         clear_pull_forces(inputrec->pull);
1127     }
1128
1129     /* update QMMMrec, if necessary */
1130     if(fr->bQMMM)
1131     {
1132         update_QMMMrec(cr,fr,x,mdatoms,box,top);
1133     }
1134
1135     if ((flags & GMX_FORCE_BONDED) && top->idef.il[F_POSRES].nr > 0)
1136     {
1137         posres_wrapper(fplog,flags,bSepDVDL,inputrec,nrnb,top,box,x,
1138                        f,enerd,lambda,fr);
1139     }
1140
1141     /* Compute the bonded and non-bonded energies and optionally forces */
1142     do_force_lowlevel(fplog,step,fr,inputrec,&(top->idef),
1143                       cr,nrnb,wcycle,mdatoms,&(inputrec->opts),
1144                       x,hist,f, bSepLRF ? fr->f_twin : f,enerd,fcd,mtop,top,fr->born,
1145                       &(top->atomtypes),bBornRadii,box,
1146                       inputrec->fepvals,lambda,graph,&(top->excls),fr->mu_tot,
1147                       flags, &cycles_pme);
1148
1149     if(bSepLRF)
1150     {
1151         if (do_per_step(step,inputrec->nstcalclr))
1152         {
1153             /* Add the long range forces to the short range forces */
1154             for(i=0; i<fr->natoms_force_constr; i++)
1155             {
1156                 rvec_add(fr->f_twin[i],f[i],f[i]);
1157             }
1158         }
1159     }
1160
1161     if (!bUseOrEmulGPU)
1162     {
1163         /* Maybe we should move this into do_force_lowlevel */
1164         do_nb_verlet(fr, ic, enerd, flags, eintLocal, enbvClearFYes,
1165                      nrnb, wcycle);
1166     }
1167
1168
1169     if (!bUseOrEmulGPU || bDiffKernels)
1170     {
1171         int aloc;
1172
1173         if (DOMAINDECOMP(cr))
1174         {
1175             do_nb_verlet(fr, ic, enerd, flags, eintNonlocal,
1176                          bDiffKernels ? enbvClearFYes : enbvClearFNo,
1177                          nrnb, wcycle);
1178         }
1179
1180         if (!bUseOrEmulGPU)
1181         {
1182             aloc = eintLocal;
1183         }
1184         else
1185         {
1186             aloc = eintNonlocal;
1187         }
1188
1189         /* Add all the non-bonded force to the normal force array.
1190          * This can be split into a local a non-local part when overlapping
1191          * communication with calculation with domain decomposition.
1192          */
1193         cycles_force += wallcycle_stop(wcycle,ewcFORCE);
1194         wallcycle_start(wcycle, ewcNB_XF_BUF_OPS);
1195         wallcycle_sub_start(wcycle, ewcsNB_F_BUF_OPS);
1196         nbnxn_atomdata_add_nbat_f_to_f(nbv->nbs,eatAll,nbv->grp[aloc].nbat,f);
1197         wallcycle_sub_stop(wcycle, ewcsNB_F_BUF_OPS);
1198         cycles_force += wallcycle_stop(wcycle, ewcNB_XF_BUF_OPS);
1199         wallcycle_start_nocount(wcycle,ewcFORCE);
1200
1201         /* if there are multiple fshift output buffers reduce them */
1202         if ((flags & GMX_FORCE_VIRIAL) &&
1203             nbv->grp[aloc].nbl_lists.nnbl > 1)
1204         {
1205             nbnxn_atomdata_add_nbat_fshift_to_fshift(nbv->grp[aloc].nbat,
1206                                                       fr->fshift);
1207         }
1208     }
1209
1210     cycles_force += wallcycle_stop(wcycle,ewcFORCE);
1211     GMX_BARRIER(cr->mpi_comm_mygroup);
1212
1213     if (ed)
1214     {
1215         do_flood(fplog,cr,x,f,ed,box,step,bNS);
1216     }
1217
1218     if (bUseOrEmulGPU && !bDiffKernels)
1219     {
1220         /* wait for non-local forces (or calculate in emulation mode) */
1221         if (DOMAINDECOMP(cr))
1222         {
1223             if (bUseGPU)
1224             {
1225                 wallcycle_start(wcycle,ewcWAIT_GPU_NB_NL);
1226                 nbnxn_cuda_wait_gpu(nbv->cu_nbv,
1227                                     nbv->grp[eintNonlocal].nbat,
1228                                     flags, eatNonlocal,
1229                                     enerd->grpp.ener[egLJSR], enerd->grpp.ener[egCOULSR],
1230                                     fr->fshift);
1231                 cycles_force += wallcycle_stop(wcycle,ewcWAIT_GPU_NB_NL);
1232             }
1233             else
1234             {
1235                 wallcycle_start_nocount(wcycle,ewcFORCE);
1236                 do_nb_verlet(fr, ic, enerd, flags, eintNonlocal, enbvClearFYes,
1237                              nrnb, wcycle);
1238                 cycles_force += wallcycle_stop(wcycle,ewcFORCE);
1239             }
1240             wallcycle_start(wcycle, ewcNB_XF_BUF_OPS);
1241             wallcycle_sub_start(wcycle, ewcsNB_F_BUF_OPS);
1242             /* skip the reduction if there was no non-local work to do */
1243             if (nbv->grp[eintLocal].nbl_lists.nbl[0]->nsci > 0)
1244             {
1245                 nbnxn_atomdata_add_nbat_f_to_f(nbv->nbs,eatNonlocal,
1246                                                nbv->grp[eintNonlocal].nbat,f);
1247             }
1248             wallcycle_sub_stop(wcycle, ewcsNB_F_BUF_OPS);
1249             cycles_force += wallcycle_stop(wcycle, ewcNB_XF_BUF_OPS);
1250         }
1251     }
1252
1253     if (bDoForces)
1254     {
1255         /* Communicate the forces */
1256         if (PAR(cr))
1257         {
1258             wallcycle_start(wcycle,ewcMOVEF);
1259             if (DOMAINDECOMP(cr))
1260             {
1261                 dd_move_f(cr->dd,f,fr->fshift);
1262                 /* Do we need to communicate the separate force array
1263                  * for terms that do not contribute to the single sum virial?
1264                  * Position restraints and electric fields do not introduce
1265                  * inter-cg forces, only full electrostatics methods do.
1266                  * When we do not calculate the virial, fr->f_novirsum = f,
1267                  * so we have already communicated these forces.
1268                  */
1269                 if (EEL_FULL(fr->eeltype) && cr->dd->n_intercg_excl &&
1270                     (flags & GMX_FORCE_VIRIAL))
1271                 {
1272                     dd_move_f(cr->dd,fr->f_novirsum,NULL);
1273                 }
1274                 if (bSepLRF)
1275                 {
1276                     /* We should not update the shift forces here,
1277                      * since f_twin is already included in f.
1278                      */
1279                     dd_move_f(cr->dd,fr->f_twin,NULL);
1280                 }
1281             }
1282             wallcycle_stop(wcycle,ewcMOVEF);
1283         }
1284     }
1285
1286     if (bUseOrEmulGPU)
1287     {
1288         /* wait for local forces (or calculate in emulation mode) */
1289         if (bUseGPU)
1290         {
1291             wallcycle_start(wcycle,ewcWAIT_GPU_NB_L);
1292             nbnxn_cuda_wait_gpu(nbv->cu_nbv,
1293                                 nbv->grp[eintLocal].nbat,
1294                                 flags, eatLocal,
1295                                 enerd->grpp.ener[egLJSR], enerd->grpp.ener[egCOULSR],
1296                                 fr->fshift);
1297             wallcycle_stop(wcycle,ewcWAIT_GPU_NB_L);
1298
1299             /* now clear the GPU outputs while we finish the step on the CPU */
1300             nbnxn_cuda_clear_outputs(nbv->cu_nbv, flags);
1301         }
1302         else
1303         {
1304             wallcycle_start_nocount(wcycle,ewcFORCE);
1305             do_nb_verlet(fr, ic, enerd, flags, eintLocal,
1306                          DOMAINDECOMP(cr) ? enbvClearFNo : enbvClearFYes,
1307                          nrnb, wcycle);
1308             wallcycle_stop(wcycle,ewcFORCE);
1309         }
1310         wallcycle_start(wcycle, ewcNB_XF_BUF_OPS);
1311         wallcycle_sub_start(wcycle, ewcsNB_F_BUF_OPS);
1312         if (nbv->grp[eintLocal].nbl_lists.nbl[0]->nsci > 0)
1313         {
1314             /* skip the reduction if there was no non-local work to do */
1315             nbnxn_atomdata_add_nbat_f_to_f(nbv->nbs,eatLocal,
1316                                            nbv->grp[eintLocal].nbat,f);
1317         }
1318         wallcycle_sub_stop(wcycle, ewcsNB_F_BUF_OPS);
1319         wallcycle_stop(wcycle, ewcNB_XF_BUF_OPS);
1320     }
1321
1322     if (DOMAINDECOMP(cr))
1323     {
1324         dd_force_flop_stop(cr->dd,nrnb);
1325         if (wcycle)
1326         {
1327             dd_cycles_add(cr->dd,cycles_force-cycles_pme,ddCyclF);
1328         }
1329     }
1330
1331     if (bDoForces)
1332     {
1333         if (IR_ELEC_FIELD(*inputrec))
1334         {
1335             /* Compute forces due to electric field */
1336             calc_f_el(MASTER(cr) ? field : NULL,
1337                       start,homenr,mdatoms->chargeA,x,fr->f_novirsum,
1338                       inputrec->ex,inputrec->et,t);
1339         }
1340
1341         /* If we have NoVirSum forces, but we do not calculate the virial,
1342          * we sum fr->f_novirum=f later.
1343          */
1344         if (vsite && !(fr->bF_NoVirSum && !(flags & GMX_FORCE_VIRIAL)))
1345         {
1346             wallcycle_start(wcycle,ewcVSITESPREAD);
1347             spread_vsite_f(fplog,vsite,x,f,fr->fshift,FALSE,NULL,nrnb,
1348                            &top->idef,fr->ePBC,fr->bMolPBC,graph,box,cr);
1349             wallcycle_stop(wcycle,ewcVSITESPREAD);
1350
1351             if (bSepLRF)
1352             {
1353                 wallcycle_start(wcycle,ewcVSITESPREAD);
1354                 spread_vsite_f(fplog,vsite,x,fr->f_twin,NULL,FALSE,NULL,
1355                                nrnb,
1356                                &top->idef,fr->ePBC,fr->bMolPBC,graph,box,cr);
1357                 wallcycle_stop(wcycle,ewcVSITESPREAD);
1358             }
1359         }
1360
1361         if (flags & GMX_FORCE_VIRIAL)
1362         {
1363             /* Calculation of the virial must be done after vsites! */
1364             calc_virial(fplog,mdatoms->start,mdatoms->homenr,x,f,
1365                         vir_force,graph,box,nrnb,fr,inputrec->ePBC);
1366         }
1367     }
1368
1369     if (inputrec->ePull == epullUMBRELLA || inputrec->ePull == epullCONST_F)
1370     {
1371         pull_potential_wrapper(fplog,bSepDVDL,cr,inputrec,box,x,
1372                                f,vir_force,mdatoms,enerd,lambda,t);
1373     }
1374
1375     if (PAR(cr) && !(cr->duty & DUTY_PME))
1376     {
1377         /* In case of node-splitting, the PP nodes receive the long-range
1378          * forces, virial and energy from the PME nodes here.
1379          */
1380         pme_receive_force_ener(fplog,bSepDVDL,cr,wcycle,enerd,fr);
1381     }
1382
1383     if (bDoForces)
1384     {
1385         post_process_forces(fplog,cr,step,nrnb,wcycle,
1386                             top,box,x,f,vir_force,mdatoms,graph,fr,vsite,
1387                             flags);
1388     }
1389
1390     /* Sum the potential energy terms from group contributions */
1391     sum_epot(&(inputrec->opts),enerd);
1392 }
1393
1394 void do_force_cutsGROUP(FILE *fplog,t_commrec *cr,
1395               t_inputrec *inputrec,
1396               gmx_large_int_t step,t_nrnb *nrnb,gmx_wallcycle_t wcycle,
1397               gmx_localtop_t *top,
1398               gmx_mtop_t *mtop,
1399               gmx_groups_t *groups,
1400               matrix box,rvec x[],history_t *hist,
1401               rvec f[],
1402               tensor vir_force,
1403               t_mdatoms *mdatoms,
1404               gmx_enerdata_t *enerd,t_fcdata *fcd,
1405               real *lambda,t_graph *graph,
1406               t_forcerec *fr,gmx_vsite_t *vsite,rvec mu_tot,
1407               double t,FILE *field,gmx_edsam_t ed,
1408               gmx_bool bBornRadii,
1409               int flags)
1410 {
1411     int    cg0,cg1,i,j;
1412     int    start,homenr;
1413     double mu[2*DIM];
1414     gmx_bool   bSepDVDL,bStateChanged,bNS,bFillGrid,bCalcCGCM,bBS;
1415     gmx_bool   bDoLongRangeNS,bDoForces,bDoPotential,bSepLRF;
1416     gmx_bool   bDoAdressWF;
1417     matrix boxs;
1418     rvec   vzero,box_diag;
1419     real   e,v,dvdlambda[efptNR];
1420     t_pbc  pbc;
1421     float  cycles_pme,cycles_force;
1422
1423     start  = mdatoms->start;
1424     homenr = mdatoms->homenr;
1425
1426     bSepDVDL = (fr->bSepDVDL && do_per_step(step,inputrec->nstlog));
1427
1428     clear_mat(vir_force);
1429
1430     if (PARTDECOMP(cr))
1431     {
1432         pd_cg_range(cr,&cg0,&cg1);
1433     }
1434     else
1435     {
1436         cg0 = 0;
1437         if (DOMAINDECOMP(cr))
1438         {
1439             cg1 = cr->dd->ncg_tot;
1440         }
1441         else
1442         {
1443             cg1 = top->cgs.nr;
1444         }
1445         if (fr->n_tpi > 0)
1446         {
1447             cg1--;
1448         }
1449     }
1450
1451     bStateChanged  = (flags & GMX_FORCE_STATECHANGED);
1452     bNS            = (flags & GMX_FORCE_NS) && (fr->bAllvsAll==FALSE);
1453     /* Should we update the long-range neighborlists at this step? */
1454     bDoLongRangeNS = fr->bTwinRange && bNS;
1455     /* Should we perform the long-range nonbonded evaluation inside the neighborsearching? */
1456     bFillGrid      = (bNS && bStateChanged);
1457     bCalcCGCM      = (bFillGrid && !DOMAINDECOMP(cr));
1458     bDoForces      = (flags & GMX_FORCE_FORCES);
1459     bDoPotential   = (flags & GMX_FORCE_ENERGY);
1460     bSepLRF        = ((inputrec->nstcalclr>1) && bDoForces &&
1461                       (flags & GMX_FORCE_SEPLRF) && (flags & GMX_FORCE_DO_LR));
1462
1463     /* should probably move this to the forcerec since it doesn't change */
1464     bDoAdressWF   = ((fr->adress_type!=eAdressOff));
1465
1466     if (bStateChanged)
1467     {
1468         update_forcerec(fplog,fr,box);
1469
1470         if (NEED_MUTOT(*inputrec))
1471         {
1472             /* Calculate total (local) dipole moment in a temporary common array.
1473              * This makes it possible to sum them over nodes faster.
1474              */
1475             calc_mu(start,homenr,
1476                     x,mdatoms->chargeA,mdatoms->chargeB,mdatoms->nChargePerturbed,
1477                     mu,mu+DIM);
1478         }
1479     }
1480
1481     if (fr->ePBC != epbcNONE) {
1482         /* Compute shift vectors every step,
1483          * because of pressure coupling or box deformation!
1484          */
1485         if ((flags & GMX_FORCE_DYNAMICBOX) && bStateChanged)
1486             calc_shifts(box,fr->shift_vec);
1487
1488         if (bCalcCGCM) {
1489             put_charge_groups_in_box(fplog,cg0,cg1,fr->ePBC,box,
1490                     &(top->cgs),x,fr->cg_cm);
1491             inc_nrnb(nrnb,eNR_CGCM,homenr);
1492             inc_nrnb(nrnb,eNR_RESETX,cg1-cg0);
1493         }
1494         else if (EI_ENERGY_MINIMIZATION(inputrec->eI) && graph) {
1495             unshift_self(graph,box,x);
1496         }
1497     }
1498     else if (bCalcCGCM) {
1499         calc_cgcm(fplog,cg0,cg1,&(top->cgs),x,fr->cg_cm);
1500         inc_nrnb(nrnb,eNR_CGCM,homenr);
1501     }
1502
1503     if (bCalcCGCM) {
1504         if (PAR(cr)) {
1505             move_cgcm(fplog,cr,fr->cg_cm);
1506         }
1507         if (gmx_debug_at)
1508             pr_rvecs(debug,0,"cgcm",fr->cg_cm,top->cgs.nr);
1509     }
1510
1511 #ifdef GMX_MPI
1512     if (!(cr->duty & DUTY_PME)) {
1513         /* Send particle coordinates to the pme nodes.
1514          * Since this is only implemented for domain decomposition
1515          * and domain decomposition does not use the graph,
1516          * we do not need to worry about shifting.
1517          */
1518
1519         wallcycle_start(wcycle,ewcPP_PMESENDX);
1520         GMX_MPE_LOG(ev_send_coordinates_start);
1521
1522         bBS = (inputrec->nwall == 2);
1523         if (bBS) {
1524             copy_mat(box,boxs);
1525             svmul(inputrec->wall_ewald_zfac,boxs[ZZ],boxs[ZZ]);
1526         }
1527
1528         gmx_pme_send_x(cr,bBS ? boxs : box,x,
1529                        mdatoms->nChargePerturbed,lambda[efptCOUL],
1530                        (flags & (GMX_FORCE_VIRIAL | GMX_FORCE_ENERGY)),step);
1531
1532         GMX_MPE_LOG(ev_send_coordinates_finish);
1533         wallcycle_stop(wcycle,ewcPP_PMESENDX);
1534     }
1535 #endif /* GMX_MPI */
1536
1537     /* Communicate coordinates and sum dipole if necessary */
1538     if (PAR(cr))
1539     {
1540         wallcycle_start(wcycle,ewcMOVEX);
1541         if (DOMAINDECOMP(cr))
1542         {
1543             dd_move_x(cr->dd,box,x);
1544         }
1545         else
1546         {
1547             move_x(fplog,cr,GMX_LEFT,GMX_RIGHT,x,nrnb);
1548         }
1549         wallcycle_stop(wcycle,ewcMOVEX);
1550     }
1551
1552     /* update adress weight beforehand */
1553     if(bStateChanged && bDoAdressWF)
1554     {
1555         /* need pbc for adress weight calculation with pbc_dx */
1556         set_pbc(&pbc,inputrec->ePBC,box);
1557         if(fr->adress_site == eAdressSITEcog)
1558         {
1559             update_adress_weights_cog(top->idef.iparams,top->idef.il,x,fr,mdatoms,
1560                                       inputrec->ePBC==epbcNONE ? NULL : &pbc);
1561         }
1562         else if (fr->adress_site == eAdressSITEcom)
1563         {
1564             update_adress_weights_com(fplog,cg0,cg1,&(top->cgs),x,fr,mdatoms,
1565                                       inputrec->ePBC==epbcNONE ? NULL : &pbc);
1566         }
1567         else if (fr->adress_site == eAdressSITEatomatom){
1568             update_adress_weights_atom_per_atom(cg0,cg1,&(top->cgs),x,fr,mdatoms,
1569                                                 inputrec->ePBC==epbcNONE ? NULL : &pbc);
1570         }
1571         else
1572         {
1573             update_adress_weights_atom(cg0,cg1,&(top->cgs),x,fr,mdatoms,
1574                                        inputrec->ePBC==epbcNONE ? NULL : &pbc);
1575         }
1576     }
1577
1578     if (NEED_MUTOT(*inputrec))
1579     {
1580
1581         if (bStateChanged)
1582         {
1583             if (PAR(cr))
1584             {
1585                 gmx_sumd(2*DIM,mu,cr);
1586             }
1587             for(i=0; i<2; i++)
1588             {
1589                 for(j=0;j<DIM;j++)
1590                 {
1591                     fr->mu_tot[i][j] = mu[i*DIM + j];
1592                 }
1593             }
1594         }
1595         if (fr->efep == efepNO)
1596         {
1597             copy_rvec(fr->mu_tot[0],mu_tot);
1598         }
1599         else
1600         {
1601             for(j=0; j<DIM; j++)
1602             {
1603                 mu_tot[j] =
1604                     (1.0 - lambda[efptCOUL])*fr->mu_tot[0][j] + lambda[efptCOUL]*fr->mu_tot[1][j];
1605             }
1606         }
1607     }
1608
1609     /* Reset energies */
1610     reset_enerdata(&(inputrec->opts),fr,bNS,enerd,MASTER(cr));
1611     clear_rvecs(SHIFTS,fr->fshift);
1612
1613     if (bNS)
1614     {
1615         wallcycle_start(wcycle,ewcNS);
1616
1617         if (graph && bStateChanged)
1618         {
1619             /* Calculate intramolecular shift vectors to make molecules whole */
1620             mk_mshift(fplog,graph,fr->ePBC,box,x);
1621         }
1622
1623         /* Do the actual neighbour searching and if twin range electrostatics
1624          * also do the calculation of long range forces and energies.
1625          */
1626         for (i=0;i<efptNR;i++) {dvdlambda[i] = 0;}
1627         ns(fplog,fr,x,box,
1628            groups,&(inputrec->opts),top,mdatoms,
1629            cr,nrnb,lambda,dvdlambda,&enerd->grpp,bFillGrid,
1630            bDoLongRangeNS);
1631         if (bSepDVDL)
1632         {
1633             fprintf(fplog,sepdvdlformat,"LR non-bonded",0.0,dvdlambda);
1634         }
1635         enerd->dvdl_lin[efptVDW] += dvdlambda[efptVDW];
1636         enerd->dvdl_lin[efptCOUL] += dvdlambda[efptCOUL];
1637
1638         wallcycle_stop(wcycle,ewcNS);
1639     }
1640
1641     if (inputrec->implicit_solvent && bNS)
1642     {
1643         make_gb_nblist(cr,inputrec->gb_algorithm,inputrec->rlist,
1644                        x,box,fr,&top->idef,graph,fr->born);
1645     }
1646
1647     if (DOMAINDECOMP(cr))
1648     {
1649         if (!(cr->duty & DUTY_PME))
1650         {
1651             wallcycle_start(wcycle,ewcPPDURINGPME);
1652             dd_force_flop_start(cr->dd,nrnb);
1653         }
1654     }
1655
1656     if (inputrec->bRot)
1657     {
1658         /* Enforced rotation has its own cycle counter that starts after the collective
1659          * coordinates have been communicated. It is added to ddCyclF to allow
1660          * for proper load-balancing */
1661         wallcycle_start(wcycle,ewcROT);
1662         do_rotation(cr,inputrec,box,x,t,step,wcycle,bNS);
1663         wallcycle_stop(wcycle,ewcROT);
1664     }
1665
1666     /* Start the force cycle counter.
1667      * This counter is stopped in do_forcelow_level.
1668      * No parallel communication should occur while this counter is running,
1669      * since that will interfere with the dynamic load balancing.
1670      */
1671     wallcycle_start(wcycle,ewcFORCE);
1672
1673     if (bDoForces)
1674     {
1675         /* Reset forces for which the virial is calculated separately:
1676          * PME/Ewald forces if necessary */
1677         if (fr->bF_NoVirSum)
1678         {
1679             if (flags & GMX_FORCE_VIRIAL)
1680             {
1681                 fr->f_novirsum = fr->f_novirsum_alloc;
1682                 GMX_BARRIER(cr->mpi_comm_mygroup);
1683                 if (fr->bDomDec)
1684                 {
1685                     clear_rvecs(fr->f_novirsum_n,fr->f_novirsum);
1686                 }
1687                 else
1688                 {
1689                     clear_rvecs(homenr,fr->f_novirsum+start);
1690                 }
1691                 GMX_BARRIER(cr->mpi_comm_mygroup);
1692             }
1693             else
1694             {
1695                 /* We are not calculating the pressure so we do not need
1696                  * a separate array for forces that do not contribute
1697                  * to the pressure.
1698                  */
1699                 fr->f_novirsum = f;
1700             }
1701         }
1702
1703         /* Clear the short- and long-range forces */
1704         clear_rvecs(fr->natoms_force_constr,f);
1705         if(bSepLRF && do_per_step(step,inputrec->nstcalclr))
1706         {
1707             clear_rvecs(fr->natoms_force_constr,fr->f_twin);
1708         }
1709
1710         clear_rvec(fr->vir_diag_posres);
1711
1712         GMX_BARRIER(cr->mpi_comm_mygroup);
1713     }
1714     if (inputrec->ePull == epullCONSTRAINT)
1715     {
1716         clear_pull_forces(inputrec->pull);
1717     }
1718
1719     /* update QMMMrec, if necessary */
1720     if(fr->bQMMM)
1721     {
1722         update_QMMMrec(cr,fr,x,mdatoms,box,top);
1723     }
1724
1725     if ((flags & GMX_FORCE_BONDED) && top->idef.il[F_POSRES].nr > 0)
1726     {
1727         posres_wrapper(fplog,flags,bSepDVDL,inputrec,nrnb,top,box,x,
1728                        f,enerd,lambda,fr);
1729     }
1730
1731     /* Compute the bonded and non-bonded energies and optionally forces */
1732     do_force_lowlevel(fplog,step,fr,inputrec,&(top->idef),
1733                       cr,nrnb,wcycle,mdatoms,&(inputrec->opts),
1734                       x,hist,f, bSepLRF ? fr->f_twin : f,enerd,fcd,mtop,top,fr->born,
1735                       &(top->atomtypes),bBornRadii,box,
1736                       inputrec->fepvals,lambda,
1737                       graph,&(top->excls),fr->mu_tot,
1738                       flags,
1739                       &cycles_pme);
1740
1741     if(bSepLRF)
1742     {
1743         if (do_per_step(step,inputrec->nstcalclr))
1744         {
1745             /* Add the long range forces to the short range forces */
1746             for(i=0; i<fr->natoms_force_constr; i++)
1747             {
1748                 rvec_add(fr->f_twin[i],f[i],f[i]);
1749             }
1750         }
1751     }
1752
1753     cycles_force = wallcycle_stop(wcycle,ewcFORCE);
1754     GMX_BARRIER(cr->mpi_comm_mygroup);
1755
1756     if (ed)
1757     {
1758         do_flood(fplog,cr,x,f,ed,box,step,bNS);
1759     }
1760
1761     if (DOMAINDECOMP(cr))
1762     {
1763         dd_force_flop_stop(cr->dd,nrnb);
1764         if (wcycle)
1765         {
1766             dd_cycles_add(cr->dd,cycles_force-cycles_pme,ddCyclF);
1767         }
1768     }
1769
1770     if (bDoForces)
1771     {
1772         if (IR_ELEC_FIELD(*inputrec))
1773         {
1774             /* Compute forces due to electric field */
1775             calc_f_el(MASTER(cr) ? field : NULL,
1776                       start,homenr,mdatoms->chargeA,x,fr->f_novirsum,
1777                       inputrec->ex,inputrec->et,t);
1778         }
1779
1780         if (bDoAdressWF && fr->adress_icor == eAdressICThermoForce)
1781         {
1782             /* Compute thermodynamic force in hybrid AdResS region */
1783             adress_thermo_force(start,homenr,&(top->cgs),x,fr->f_novirsum,fr,mdatoms,
1784                                 inputrec->ePBC==epbcNONE ? NULL : &pbc);
1785         }
1786
1787         /* Communicate the forces */
1788         if (PAR(cr))
1789         {
1790             wallcycle_start(wcycle,ewcMOVEF);
1791             if (DOMAINDECOMP(cr))
1792             {
1793                 dd_move_f(cr->dd,f,fr->fshift);
1794                 /* Do we need to communicate the separate force array
1795                  * for terms that do not contribute to the single sum virial?
1796                  * Position restraints and electric fields do not introduce
1797                  * inter-cg forces, only full electrostatics methods do.
1798                  * When we do not calculate the virial, fr->f_novirsum = f,
1799                  * so we have already communicated these forces.
1800                  */
1801                 if (EEL_FULL(fr->eeltype) && cr->dd->n_intercg_excl &&
1802                     (flags & GMX_FORCE_VIRIAL))
1803                 {
1804                     dd_move_f(cr->dd,fr->f_novirsum,NULL);
1805                 }
1806                 if (bSepLRF)
1807                 {
1808                     /* We should not update the shift forces here,
1809                      * since f_twin is already included in f.
1810                      */
1811                     dd_move_f(cr->dd,fr->f_twin,NULL);
1812                 }
1813             }
1814             else
1815             {
1816                 pd_move_f(cr,f,nrnb);
1817                 if (bSepLRF)
1818                 {
1819                     pd_move_f(cr,fr->f_twin,nrnb);
1820                 }
1821             }
1822             wallcycle_stop(wcycle,ewcMOVEF);
1823         }
1824
1825         /* If we have NoVirSum forces, but we do not calculate the virial,
1826          * we sum fr->f_novirum=f later.
1827          */
1828         if (vsite && !(fr->bF_NoVirSum && !(flags & GMX_FORCE_VIRIAL)))
1829         {
1830             wallcycle_start(wcycle,ewcVSITESPREAD);
1831             spread_vsite_f(fplog,vsite,x,f,fr->fshift,FALSE,NULL,nrnb,
1832                            &top->idef,fr->ePBC,fr->bMolPBC,graph,box,cr);
1833             wallcycle_stop(wcycle,ewcVSITESPREAD);
1834
1835             if (bSepLRF)
1836             {
1837                 wallcycle_start(wcycle,ewcVSITESPREAD);
1838                 spread_vsite_f(fplog,vsite,x,fr->f_twin,NULL,FALSE,NULL,
1839                                nrnb,
1840                                &top->idef,fr->ePBC,fr->bMolPBC,graph,box,cr);
1841                 wallcycle_stop(wcycle,ewcVSITESPREAD);
1842             }
1843         }
1844
1845         if (flags & GMX_FORCE_VIRIAL)
1846         {
1847             /* Calculation of the virial must be done after vsites! */
1848             calc_virial(fplog,mdatoms->start,mdatoms->homenr,x,f,
1849                         vir_force,graph,box,nrnb,fr,inputrec->ePBC);
1850         }
1851     }
1852
1853     if (inputrec->ePull == epullUMBRELLA || inputrec->ePull == epullCONST_F)
1854     {
1855         pull_potential_wrapper(fplog,bSepDVDL,cr,inputrec,box,x,
1856                                f,vir_force,mdatoms,enerd,lambda,t);
1857     }
1858
1859     /* Add the forces from enforced rotation potentials (if any) */
1860     if (inputrec->bRot)
1861     {
1862         wallcycle_start(wcycle,ewcROTadd);
1863         enerd->term[F_COM_PULL] += add_rot_forces(inputrec->rot, f, cr,step,t);
1864         wallcycle_stop(wcycle,ewcROTadd);
1865     }
1866
1867     if (PAR(cr) && !(cr->duty & DUTY_PME))
1868     {
1869         /* In case of node-splitting, the PP nodes receive the long-range
1870          * forces, virial and energy from the PME nodes here.
1871          */
1872         pme_receive_force_ener(fplog,bSepDVDL,cr,wcycle,enerd,fr);
1873     }
1874
1875     if (bDoForces)
1876     {
1877         post_process_forces(fplog,cr,step,nrnb,wcycle,
1878                             top,box,x,f,vir_force,mdatoms,graph,fr,vsite,
1879                             flags);
1880     }
1881
1882     /* Sum the potential energy terms from group contributions */
1883     sum_epot(&(inputrec->opts),enerd);
1884 }
1885
1886 void do_force(FILE *fplog,t_commrec *cr,
1887               t_inputrec *inputrec,
1888               gmx_large_int_t step,t_nrnb *nrnb,gmx_wallcycle_t wcycle,
1889               gmx_localtop_t *top,
1890               gmx_mtop_t *mtop,
1891               gmx_groups_t *groups,
1892               matrix box,rvec x[],history_t *hist,
1893               rvec f[],
1894               tensor vir_force,
1895               t_mdatoms *mdatoms,
1896               gmx_enerdata_t *enerd,t_fcdata *fcd,
1897               real *lambda,t_graph *graph,
1898               t_forcerec *fr,
1899               gmx_vsite_t *vsite,rvec mu_tot,
1900               double t,FILE *field,gmx_edsam_t ed,
1901               gmx_bool bBornRadii,
1902               int flags)
1903 {
1904     /* modify force flag if not doing nonbonded */
1905     if (!fr->bNonbonded)
1906     {
1907         flags &= ~GMX_FORCE_NONBONDED;
1908     }
1909
1910     switch (inputrec->cutoff_scheme)
1911     {
1912         case ecutsVERLET:
1913             do_force_cutsVERLET(fplog, cr, inputrec,
1914                                 step, nrnb, wcycle,
1915                                 top, mtop,
1916                                 groups,
1917                                 box, x, hist,
1918                                 f, vir_force,
1919                                 mdatoms,
1920                                 enerd, fcd,
1921                                 lambda, graph,
1922                                 fr, fr->ic,
1923                                 vsite, mu_tot,
1924                                 t, field, ed,
1925                                 bBornRadii,
1926                                 flags);
1927             break;
1928         case ecutsGROUP:
1929              do_force_cutsGROUP(fplog, cr, inputrec,
1930                                 step, nrnb, wcycle,
1931                                 top, mtop,
1932                                 groups,
1933                                 box, x, hist,
1934                                 f, vir_force,
1935                                 mdatoms,
1936                                 enerd, fcd,
1937                                 lambda, graph,
1938                                 fr, vsite, mu_tot,
1939                                 t, field, ed,
1940                                 bBornRadii,
1941                                 flags);
1942             break;
1943         default:
1944             gmx_incons("Invalid cut-off scheme passed!");
1945     }
1946 }
1947
1948
1949 void do_constrain_first(FILE *fplog,gmx_constr_t constr,
1950                         t_inputrec *ir,t_mdatoms *md,
1951                         t_state *state,rvec *f,
1952                         t_graph *graph,t_commrec *cr,t_nrnb *nrnb,
1953                         t_forcerec *fr, gmx_localtop_t *top, tensor shake_vir)
1954 {
1955     int    i,m,start,end;
1956     gmx_large_int_t step;
1957     real   dt=ir->delta_t;
1958     real   dvdl_dum;
1959     rvec   *savex;
1960
1961     snew(savex,state->natoms);
1962
1963     start = md->start;
1964     end   = md->homenr + start;
1965
1966     if (debug)
1967         fprintf(debug,"vcm: start=%d, homenr=%d, end=%d\n",
1968                 start,md->homenr,end);
1969     /* Do a first constrain to reset particles... */
1970     step = ir->init_step;
1971     if (fplog)
1972     {
1973         char buf[STEPSTRSIZE];
1974         fprintf(fplog,"\nConstraining the starting coordinates (step %s)\n",
1975                 gmx_step_str(step,buf));
1976     }
1977     dvdl_dum = 0;
1978
1979     /* constrain the current position */
1980     constrain(NULL,TRUE,FALSE,constr,&(top->idef),
1981               ir,NULL,cr,step,0,md,
1982               state->x,state->x,NULL,
1983               fr->bMolPBC,state->box,
1984               state->lambda[efptBONDED],&dvdl_dum,
1985               NULL,NULL,nrnb,econqCoord,
1986               ir->epc==epcMTTK,state->veta,state->veta);
1987     if (EI_VV(ir->eI))
1988     {
1989         /* constrain the inital velocity, and save it */
1990         /* also may be useful if we need the ekin from the halfstep for velocity verlet */
1991         /* might not yet treat veta correctly */
1992         constrain(NULL,TRUE,FALSE,constr,&(top->idef),
1993                   ir,NULL,cr,step,0,md,
1994                   state->x,state->v,state->v,
1995                   fr->bMolPBC,state->box,
1996                   state->lambda[efptBONDED],&dvdl_dum,
1997                   NULL,NULL,nrnb,econqVeloc,
1998                   ir->epc==epcMTTK,state->veta,state->veta);
1999     }
2000     /* constrain the inital velocities at t-dt/2 */
2001     if (EI_STATE_VELOCITY(ir->eI) && ir->eI!=eiVV)
2002     {
2003         for(i=start; (i<end); i++)
2004         {
2005             for(m=0; (m<DIM); m++)
2006             {
2007                 /* Reverse the velocity */
2008                 state->v[i][m] = -state->v[i][m];
2009                 /* Store the position at t-dt in buf */
2010                 savex[i][m] = state->x[i][m] + dt*state->v[i][m];
2011             }
2012         }
2013     /* Shake the positions at t=-dt with the positions at t=0
2014      * as reference coordinates.
2015          */
2016         if (fplog)
2017         {
2018             char buf[STEPSTRSIZE];
2019             fprintf(fplog,"\nConstraining the coordinates at t0-dt (step %s)\n",
2020                     gmx_step_str(step,buf));
2021         }
2022         dvdl_dum = 0;
2023         constrain(NULL,TRUE,FALSE,constr,&(top->idef),
2024                   ir,NULL,cr,step,-1,md,
2025                   state->x,savex,NULL,
2026                   fr->bMolPBC,state->box,
2027                   state->lambda[efptBONDED],&dvdl_dum,
2028                   state->v,NULL,nrnb,econqCoord,
2029                   ir->epc==epcMTTK,state->veta,state->veta);
2030
2031         for(i=start; i<end; i++) {
2032             for(m=0; m<DIM; m++) {
2033                 /* Re-reverse the velocities */
2034                 state->v[i][m] = -state->v[i][m];
2035             }
2036         }
2037     }
2038     sfree(savex);
2039 }
2040
2041 void calc_enervirdiff(FILE *fplog,int eDispCorr,t_forcerec *fr)
2042 {
2043   double eners[2],virs[2],enersum,virsum,y0,f,g,h;
2044   double r0,r1,r,rc3,rc9,ea,eb,ec,pa,pb,pc,pd;
2045   double invscale,invscale2,invscale3;
2046   int    ri0,ri1,ri,i,offstart,offset;
2047   real   scale,*vdwtab,tabfactor,tmp;
2048
2049   fr->enershiftsix = 0;
2050   fr->enershifttwelve = 0;
2051   fr->enerdiffsix = 0;
2052   fr->enerdifftwelve = 0;
2053   fr->virdiffsix = 0;
2054   fr->virdifftwelve = 0;
2055
2056   if (eDispCorr != edispcNO) {
2057     for(i=0; i<2; i++) {
2058       eners[i] = 0;
2059       virs[i]  = 0;
2060     }
2061     if ((fr->vdwtype == evdwSWITCH) || (fr->vdwtype == evdwSHIFT)) {
2062       if (fr->rvdw_switch == 0)
2063         gmx_fatal(FARGS,
2064                   "With dispersion correction rvdw-switch can not be zero "
2065                   "for vdw-type = %s",evdw_names[fr->vdwtype]);
2066
2067       scale  = fr->nblists[0].table_elec_vdw.scale;
2068       vdwtab = fr->nblists[0].table_vdw.data;
2069
2070       /* Round the cut-offs to exact table values for precision */
2071       ri0 = floor(fr->rvdw_switch*scale);
2072       ri1 = ceil(fr->rvdw*scale);
2073       r0  = ri0/scale;
2074       r1  = ri1/scale;
2075       rc3 = r0*r0*r0;
2076       rc9  = rc3*rc3*rc3;
2077
2078       if (fr->vdwtype == evdwSHIFT)
2079       {
2080           /* Determine the constant energy shift below rvdw_switch.
2081            * Table has a scale factor since we have scaled it down to compensate
2082            * for scaling-up c6/c12 with the derivative factors to save flops in analytical kernels.
2083            */
2084           fr->enershiftsix    = (real)(-1.0/(rc3*rc3)) - 6.0*vdwtab[8*ri0];
2085           fr->enershifttwelve = (real)( 1.0/(rc9*rc3)) - 12.0*vdwtab[8*ri0 + 4];
2086       }
2087       /* Add the constant part from 0 to rvdw_switch.
2088        * This integration from 0 to rvdw_switch overcounts the number
2089        * of interactions by 1, as it also counts the self interaction.
2090        * We will correct for this later.
2091        */
2092       eners[0] += 4.0*M_PI*fr->enershiftsix*rc3/3.0;
2093       eners[1] += 4.0*M_PI*fr->enershifttwelve*rc3/3.0;
2094
2095       invscale = 1.0/(scale);
2096       invscale2 = invscale*invscale;
2097       invscale3 = invscale*invscale2;
2098
2099       /* following summation derived from cubic spline definition,
2100         Numerical Recipies in C, second edition, p. 113-116.  Exact
2101         for the cubic spline.  We first calculate the negative of
2102         the energy from rvdw to rvdw_switch, assuming that g(r)=1,
2103         and then add the more standard, abrupt cutoff correction to
2104         that result, yielding the long-range correction for a
2105         switched function.  We perform both the pressure and energy
2106         loops at the same time for simplicity, as the computational
2107         cost is low. */
2108
2109       for (i=0;i<2;i++) {
2110         enersum = 0.0; virsum = 0.0;
2111         if (i==0)
2112         {
2113             offstart = 0;
2114             /* Since the dispersion table has been scaled down a factor 6.0 and the repulsion
2115              * a factor 12.0 to compensate for the c6/c12 parameters inside nbfp[] being scaled
2116              * up (to save flops in kernels), we need to correct for this.
2117              */
2118             tabfactor = 6.0;
2119         }
2120         else
2121         {
2122             offstart = 4;
2123             tabfactor = 12.0;
2124         }
2125         for (ri=ri0; ri<ri1; ri++) {
2126           r = ri*invscale;
2127           ea = invscale3;
2128           eb = 2.0*invscale2*r;
2129           ec = invscale*r*r;
2130
2131           pa = invscale3;
2132           pb = 3.0*invscale2*r;
2133           pc = 3.0*invscale*r*r;
2134           pd = r*r*r;
2135
2136           /* this "8" is from the packing in the vdwtab array - perhaps should be #define'ed? */
2137           offset = 8*ri + offstart;
2138           y0 = vdwtab[offset];
2139           f  = vdwtab[offset+1];
2140           g  = vdwtab[offset+2];
2141           h  = vdwtab[offset+3];
2142
2143           enersum += y0*(ea/3 + eb/2 + ec) + f*(ea/4 + eb/3 + ec/2) + g*(ea/5 + eb/4 + ec/3) + h*(ea/6 + eb/5 + ec/4);
2144           virsum  += f*(pa/4 + pb/3 + pc/2 + pd) + 2*g*(pa/5 + pb/4 + pc/3 + pd/2) + 3*h*(pa/6 + pb/5 + pc/4 + pd/3);
2145         }
2146
2147         enersum *= 4.0*M_PI*tabfactor;
2148         virsum  *= 4.0*M_PI*tabfactor;
2149         eners[i] -= enersum;
2150         virs[i]  -= virsum;
2151       }
2152
2153       /* now add the correction for rvdw_switch to infinity */
2154       eners[0] += -4.0*M_PI/(3.0*rc3);
2155       eners[1] +=  4.0*M_PI/(9.0*rc9);
2156       virs[0]  +=  8.0*M_PI/rc3;
2157       virs[1]  += -16.0*M_PI/(3.0*rc9);
2158     }
2159     else if ((fr->vdwtype == evdwCUT) || (fr->vdwtype == evdwUSER)) {
2160       if (fr->vdwtype == evdwUSER && fplog)
2161         fprintf(fplog,
2162                 "WARNING: using dispersion correction with user tables\n");
2163       rc3  = fr->rvdw*fr->rvdw*fr->rvdw;
2164       rc9  = rc3*rc3*rc3;
2165       /* Contribution beyond the cut-off */
2166       eners[0] += -4.0*M_PI/(3.0*rc3);
2167       eners[1] +=  4.0*M_PI/(9.0*rc9);
2168       if (fr->vdw_modifier==eintmodPOTSHIFT) {
2169           /* Contribution within the cut-off */
2170           eners[0] += -4.0*M_PI/(3.0*rc3);
2171           eners[1] +=  4.0*M_PI/(3.0*rc9);
2172       }
2173       /* Contribution beyond the cut-off */
2174       virs[0]  +=  8.0*M_PI/rc3;
2175       virs[1]  += -16.0*M_PI/(3.0*rc9);
2176     } else {
2177       gmx_fatal(FARGS,
2178                 "Dispersion correction is not implemented for vdw-type = %s",
2179                 evdw_names[fr->vdwtype]);
2180     }
2181     fr->enerdiffsix    = eners[0];
2182     fr->enerdifftwelve = eners[1];
2183     /* The 0.5 is due to the Gromacs definition of the virial */
2184     fr->virdiffsix     = 0.5*virs[0];
2185     fr->virdifftwelve  = 0.5*virs[1];
2186   }
2187 }
2188
2189 void calc_dispcorr(FILE *fplog,t_inputrec *ir,t_forcerec *fr,
2190                    gmx_large_int_t step,int natoms,
2191                    matrix box,real lambda,tensor pres,tensor virial,
2192                    real *prescorr, real *enercorr, real *dvdlcorr)
2193 {
2194     gmx_bool bCorrAll,bCorrPres;
2195     real dvdlambda,invvol,dens,ninter,avcsix,avctwelve,enerdiff,svir=0,spres=0;
2196     int  m;
2197
2198     *prescorr = 0;
2199     *enercorr = 0;
2200     *dvdlcorr = 0;
2201
2202     clear_mat(virial);
2203     clear_mat(pres);
2204
2205     if (ir->eDispCorr != edispcNO) {
2206         bCorrAll  = (ir->eDispCorr == edispcAllEner ||
2207                      ir->eDispCorr == edispcAllEnerPres);
2208         bCorrPres = (ir->eDispCorr == edispcEnerPres ||
2209                      ir->eDispCorr == edispcAllEnerPres);
2210
2211         invvol = 1/det(box);
2212         if (fr->n_tpi)
2213         {
2214             /* Only correct for the interactions with the inserted molecule */
2215             dens = (natoms - fr->n_tpi)*invvol;
2216             ninter = fr->n_tpi;
2217         }
2218         else
2219         {
2220             dens = natoms*invvol;
2221             ninter = 0.5*natoms;
2222         }
2223
2224         if (ir->efep == efepNO)
2225         {
2226             avcsix    = fr->avcsix[0];
2227             avctwelve = fr->avctwelve[0];
2228         }
2229         else
2230         {
2231             avcsix    = (1 - lambda)*fr->avcsix[0]    + lambda*fr->avcsix[1];
2232             avctwelve = (1 - lambda)*fr->avctwelve[0] + lambda*fr->avctwelve[1];
2233         }
2234
2235         enerdiff = ninter*(dens*fr->enerdiffsix - fr->enershiftsix);
2236         *enercorr += avcsix*enerdiff;
2237         dvdlambda = 0.0;
2238         if (ir->efep != efepNO)
2239         {
2240             dvdlambda += (fr->avcsix[1] - fr->avcsix[0])*enerdiff;
2241         }
2242         if (bCorrAll)
2243         {
2244             enerdiff = ninter*(dens*fr->enerdifftwelve - fr->enershifttwelve);
2245             *enercorr += avctwelve*enerdiff;
2246             if (fr->efep != efepNO)
2247             {
2248                 dvdlambda += (fr->avctwelve[1] - fr->avctwelve[0])*enerdiff;
2249             }
2250         }
2251
2252         if (bCorrPres)
2253         {
2254             svir = ninter*dens*avcsix*fr->virdiffsix/3.0;
2255             if (ir->eDispCorr == edispcAllEnerPres)
2256             {
2257                 svir += ninter*dens*avctwelve*fr->virdifftwelve/3.0;
2258             }
2259             /* The factor 2 is because of the Gromacs virial definition */
2260             spres = -2.0*invvol*svir*PRESFAC;
2261
2262             for(m=0; m<DIM; m++) {
2263                 virial[m][m] += svir;
2264                 pres[m][m] += spres;
2265             }
2266             *prescorr += spres;
2267         }
2268
2269         /* Can't currently control when it prints, for now, just print when degugging */
2270         if (debug)
2271         {
2272             if (bCorrAll) {
2273                 fprintf(debug,"Long Range LJ corr.: <C6> %10.4e, <C12> %10.4e\n",
2274                         avcsix,avctwelve);
2275             }
2276             if (bCorrPres)
2277             {
2278                 fprintf(debug,
2279                         "Long Range LJ corr.: Epot %10g, Pres: %10g, Vir: %10g\n",
2280                         *enercorr,spres,svir);
2281             }
2282             else
2283             {
2284                 fprintf(debug,"Long Range LJ corr.: Epot %10g\n",*enercorr);
2285             }
2286         }
2287
2288         if (fr->bSepDVDL && do_per_step(step,ir->nstlog))
2289         {
2290             fprintf(fplog,sepdvdlformat,"Dispersion correction",
2291                     *enercorr,dvdlambda);
2292         }
2293         if (fr->efep != efepNO)
2294         {
2295             *dvdlcorr += dvdlambda;
2296         }
2297     }
2298 }
2299
2300 void do_pbc_first(FILE *fplog,matrix box,t_forcerec *fr,
2301                   t_graph *graph,rvec x[])
2302 {
2303   if (fplog)
2304     fprintf(fplog,"Removing pbc first time\n");
2305   calc_shifts(box,fr->shift_vec);
2306   if (graph) {
2307     mk_mshift(fplog,graph,fr->ePBC,box,x);
2308     if (gmx_debug_at)
2309       p_graph(debug,"do_pbc_first 1",graph);
2310     shift_self(graph,box,x);
2311     /* By doing an extra mk_mshift the molecules that are broken
2312      * because they were e.g. imported from another software
2313      * will be made whole again. Such are the healing powers
2314      * of GROMACS.
2315      */
2316     mk_mshift(fplog,graph,fr->ePBC,box,x);
2317     if (gmx_debug_at)
2318       p_graph(debug,"do_pbc_first 2",graph);
2319   }
2320   if (fplog)
2321     fprintf(fplog,"Done rmpbc\n");
2322 }
2323
2324 static void low_do_pbc_mtop(FILE *fplog,int ePBC,matrix box,
2325                             gmx_mtop_t *mtop,rvec x[],
2326                             gmx_bool bFirst)
2327 {
2328   t_graph *graph;
2329   int mb,as,mol;
2330   gmx_molblock_t *molb;
2331
2332   if (bFirst && fplog)
2333     fprintf(fplog,"Removing pbc first time\n");
2334
2335   snew(graph,1);
2336   as = 0;
2337   for(mb=0; mb<mtop->nmolblock; mb++) {
2338     molb = &mtop->molblock[mb];
2339     if (molb->natoms_mol == 1 ||
2340         (!bFirst && mtop->moltype[molb->type].cgs.nr == 1)) {
2341       /* Just one atom or charge group in the molecule, no PBC required */
2342       as += molb->nmol*molb->natoms_mol;
2343     } else {
2344       /* Pass NULL iso fplog to avoid graph prints for each molecule type */
2345       mk_graph_ilist(NULL,mtop->moltype[molb->type].ilist,
2346                      0,molb->natoms_mol,FALSE,FALSE,graph);
2347
2348       for(mol=0; mol<molb->nmol; mol++) {
2349         mk_mshift(fplog,graph,ePBC,box,x+as);
2350
2351         shift_self(graph,box,x+as);
2352         /* The molecule is whole now.
2353          * We don't need the second mk_mshift call as in do_pbc_first,
2354          * since we no longer need this graph.
2355          */
2356
2357         as += molb->natoms_mol;
2358       }
2359       done_graph(graph);
2360     }
2361   }
2362   sfree(graph);
2363 }
2364
2365 void do_pbc_first_mtop(FILE *fplog,int ePBC,matrix box,
2366                        gmx_mtop_t *mtop,rvec x[])
2367 {
2368   low_do_pbc_mtop(fplog,ePBC,box,mtop,x,TRUE);
2369 }
2370
2371 void do_pbc_mtop(FILE *fplog,int ePBC,matrix box,
2372                  gmx_mtop_t *mtop,rvec x[])
2373 {
2374   low_do_pbc_mtop(fplog,ePBC,box,mtop,x,FALSE);
2375 }
2376
2377 void finish_run(FILE *fplog,t_commrec *cr,const char *confout,
2378                 t_inputrec *inputrec,
2379                 t_nrnb nrnb[],gmx_wallcycle_t wcycle,
2380                 gmx_runtime_t *runtime,
2381                 wallclock_gpu_t *gputimes,
2382                 int omp_nth_pp,
2383                 gmx_bool bWriteStat)
2384 {
2385     int    i,j;
2386     t_nrnb *nrnb_tot=NULL;
2387     real   delta_t;
2388     double nbfs,mflop;
2389
2390     wallcycle_sum(cr,wcycle);
2391
2392     if (cr->nnodes > 1)
2393     {
2394         snew(nrnb_tot,1);
2395 #ifdef GMX_MPI
2396         MPI_Allreduce(nrnb->n,nrnb_tot->n,eNRNB,MPI_DOUBLE,MPI_SUM,
2397                       cr->mpi_comm_mysim);
2398 #endif
2399     }
2400     else
2401     {
2402         nrnb_tot = nrnb;
2403     }
2404
2405 #if defined(GMX_MPI) && !defined(GMX_THREAD_MPI)
2406     if (cr->nnodes > 1)
2407     {
2408         /* reduce nodetime over all MPI processes in the current simulation */
2409         double sum;
2410         MPI_Allreduce(&runtime->proctime,&sum,1,MPI_DOUBLE,MPI_SUM,
2411                       cr->mpi_comm_mysim);
2412         runtime->proctime = sum;
2413     }
2414 #endif
2415
2416     if (SIMMASTER(cr))
2417     {
2418         print_flop(fplog,nrnb_tot,&nbfs,&mflop);
2419     }
2420     if (cr->nnodes > 1)
2421     {
2422         sfree(nrnb_tot);
2423     }
2424
2425     if ((cr->duty & DUTY_PP) && DOMAINDECOMP(cr))
2426     {
2427         print_dd_statistics(cr,inputrec,fplog);
2428     }
2429
2430 #ifdef GMX_MPI
2431     if (PARTDECOMP(cr))
2432     {
2433         if (MASTER(cr))
2434         {
2435             t_nrnb     *nrnb_all;
2436             int        s;
2437             MPI_Status stat;
2438
2439             snew(nrnb_all,cr->nnodes);
2440             nrnb_all[0] = *nrnb;
2441             for(s=1; s<cr->nnodes; s++)
2442             {
2443                 MPI_Recv(nrnb_all[s].n,eNRNB,MPI_DOUBLE,s,0,
2444                          cr->mpi_comm_mysim,&stat);
2445             }
2446             pr_load(fplog,cr,nrnb_all);
2447             sfree(nrnb_all);
2448         }
2449         else
2450         {
2451             MPI_Send(nrnb->n,eNRNB,MPI_DOUBLE,MASTERRANK(cr),0,
2452                      cr->mpi_comm_mysim);
2453         }
2454     }
2455 #endif
2456
2457     if (SIMMASTER(cr))
2458     {
2459         wallcycle_print(fplog,cr->nnodes,cr->npmenodes,runtime->realtime,
2460                         wcycle,gputimes);
2461
2462         if (EI_DYNAMICS(inputrec->eI))
2463         {
2464             delta_t = inputrec->delta_t;
2465         }
2466         else
2467         {
2468             delta_t = 0;
2469         }
2470
2471         if (fplog)
2472         {
2473             print_perf(fplog,runtime->proctime,runtime->realtime,
2474                        cr->nnodes-cr->npmenodes,
2475                        runtime->nsteps_done,delta_t,nbfs,mflop,
2476                        omp_nth_pp);
2477         }
2478         if (bWriteStat)
2479         {
2480             print_perf(stderr,runtime->proctime,runtime->realtime,
2481                        cr->nnodes-cr->npmenodes,
2482                        runtime->nsteps_done,delta_t,nbfs,mflop,
2483                        omp_nth_pp);
2484         }
2485     }
2486 }
2487
2488 extern void initialize_lambdas(FILE *fplog,t_inputrec *ir,int *fep_state,real *lambda,double *lam0)
2489 {
2490     /* this function works, but could probably use a logic rewrite to keep all the different
2491        types of efep straight. */
2492
2493     int i;
2494     t_lambda *fep = ir->fepvals;
2495
2496     if ((ir->efep==efepNO) && (ir->bSimTemp == FALSE)) {
2497         for (i=0;i<efptNR;i++)  {
2498             lambda[i] = 0.0;
2499             if (lam0)
2500             {
2501                 lam0[i] = 0.0;
2502             }
2503         }
2504         return;
2505     } else {
2506         *fep_state = fep->init_fep_state; /* this might overwrite the checkpoint
2507                                              if checkpoint is set -- a kludge is in for now
2508                                              to prevent this.*/
2509         for (i=0;i<efptNR;i++)
2510         {
2511             /* overwrite lambda state with init_lambda for now for backwards compatibility */
2512             if (fep->init_lambda>=0) /* if it's -1, it was never initializd */
2513             {
2514                 lambda[i] = fep->init_lambda;
2515                 if (lam0) {
2516                     lam0[i] = lambda[i];
2517                 }
2518             }
2519             else
2520             {
2521                 lambda[i] = fep->all_lambda[i][*fep_state];
2522                 if (lam0) {
2523                     lam0[i] = lambda[i];
2524                 }
2525             }
2526         }
2527         if (ir->bSimTemp) {
2528             /* need to rescale control temperatures to match current state */
2529             for (i=0;i<ir->opts.ngtc;i++) {
2530                 if (ir->opts.ref_t[i] > 0) {
2531                     ir->opts.ref_t[i] = ir->simtempvals->temperatures[*fep_state];
2532                 }
2533             }
2534         }
2535     }
2536
2537     /* Send to the log the information on the current lambdas */
2538     if (fplog != NULL)
2539     {
2540         fprintf(fplog,"Initial vector of lambda components:[ ");
2541         for (i=0;i<efptNR;i++)
2542         {
2543             fprintf(fplog,"%10.4f ",lambda[i]);
2544         }
2545         fprintf(fplog,"]\n");
2546     }
2547     return;
2548 }
2549
2550
2551 void init_md(FILE *fplog,
2552              t_commrec *cr,t_inputrec *ir,const output_env_t oenv,
2553              double *t,double *t0,
2554              real *lambda, int *fep_state, double *lam0,
2555              t_nrnb *nrnb,gmx_mtop_t *mtop,
2556              gmx_update_t *upd,
2557              int nfile,const t_filenm fnm[],
2558              gmx_mdoutf_t **outf,t_mdebin **mdebin,
2559              tensor force_vir,tensor shake_vir,rvec mu_tot,
2560              gmx_bool *bSimAnn,t_vcm **vcm, t_state *state, unsigned long Flags)
2561 {
2562     int  i,j,n;
2563     real tmpt,mod;
2564
2565     /* Initial values */
2566     *t = *t0       = ir->init_t;
2567
2568     *bSimAnn=FALSE;
2569     for(i=0;i<ir->opts.ngtc;i++)
2570     {
2571         /* set bSimAnn if any group is being annealed */
2572         if(ir->opts.annealing[i]!=eannNO)
2573         {
2574             *bSimAnn = TRUE;
2575         }
2576     }
2577     if (*bSimAnn)
2578     {
2579         update_annealing_target_temp(&(ir->opts),ir->init_t);
2580     }
2581
2582     /* Initialize lambda variables */
2583     initialize_lambdas(fplog,ir,fep_state,lambda,lam0);
2584
2585     if (upd)
2586     {
2587         *upd = init_update(fplog,ir);
2588     }
2589
2590
2591     if (vcm != NULL)
2592     {
2593         *vcm = init_vcm(fplog,&mtop->groups,ir);
2594     }
2595
2596     if (EI_DYNAMICS(ir->eI) && !(Flags & MD_APPENDFILES))
2597     {
2598         if (ir->etc == etcBERENDSEN)
2599         {
2600             please_cite(fplog,"Berendsen84a");
2601         }
2602         if (ir->etc == etcVRESCALE)
2603         {
2604             please_cite(fplog,"Bussi2007a");
2605         }
2606     }
2607
2608     init_nrnb(nrnb);
2609
2610     if (nfile != -1)
2611     {
2612         *outf = init_mdoutf(nfile,fnm,Flags,cr,ir,oenv);
2613
2614         *mdebin = init_mdebin((Flags & MD_APPENDFILES) ? NULL : (*outf)->fp_ene,
2615                               mtop,ir, (*outf)->fp_dhdl);
2616     }
2617
2618     if (ir->bAdress)
2619     {
2620       please_cite(fplog,"Fritsch12");
2621       please_cite(fplog,"Junghans10");
2622     }
2623     /* Initiate variables */
2624     clear_mat(force_vir);
2625     clear_mat(shake_vir);
2626     clear_rvec(mu_tot);
2627
2628     debug_gmx();
2629 }
2630