src/mdlib/sim_util.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 1991-2000, University of Groningen, The Netherlands.
   5  * Copyright (c) 2001-2004, The GROMACS development team,
   6  * check out http://www.gromacs.org for more information.
   7  * Copyright (c) 2012,2013, by the GROMACS development team, led by
   8  * David van der Spoel, Berk Hess, Erik Lindahl, and including many
   9  * others, as listed in the AUTHORS file in the top-level source
  10  * directory and at http://www.gromacs.org.
  11  *
  12  * GROMACS is free software; you can redistribute it and/or
  13  * modify it under the terms of the GNU Lesser General Public License
  14  * as published by the Free Software Foundation; either version 2.1
  15  * of the License, or (at your option) any later version.
  16  *
  17  * GROMACS is distributed in the hope that it will be useful,
  18  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  19  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  20  * Lesser General Public License for more details.
  21  *
  22  * You should have received a copy of the GNU Lesser General Public
  23  * License along with GROMACS; if not, see
  24  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  25  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  26  *
  27  * If you want to redistribute modifications to GROMACS, please
  28  * consider that scientific software is very special. Version
  29  * control is crucial - bugs must be traceable. We will be happy to
  30  * consider code for inclusion in the official distribution, but
  31  * derived work must not be called official GROMACS. Details are found
  32  * in the README & COPYING files - if they are missing, get the
  33  * official version at http://www.gromacs.org.
  34  *
  35  * To help us fund GROMACS development, we humbly ask that you cite
  36  * the research papers on the package. Check out http://www.gromacs.org.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #ifdef GMX_CRAY_XT3
  43 #include <catamount/dclock.h>
  44 #endif
  45
  46
  47 #include <stdio.h>
  48 #include <time.h>
  49 #ifdef HAVE_SYS_TIME_H
  50 #include <sys/time.h>
  51 #endif
  52 #include <math.h>
  53 #include "visibility.h"
  54 #include "typedefs.h"
  55 #include "string2.h"
  56 #include "gmxfio.h"
  57 #include "smalloc.h"
  58 #include "names.h"
  59 #include "confio.h"
  60 #include "mvdata.h"
  61 #include "txtdump.h"
  62 #include "pbc.h"
  63 #include "chargegroup.h"
  64 #include "vec.h"
  65 #include <time.h>
  66 #include "nrnb.h"
  67 #include "mshift.h"
  68 #include "mdrun.h"
  69 #include "sim_util.h"
  70 #include "update.h"
  71 #include "physics.h"
  72 #include "main.h"
  73 #include "mdatoms.h"
  74 #include "force.h"
  75 #include "bondf.h"
  76 #include "pme.h"
  77 #include "disre.h"
  78 #include "orires.h"
  79 #include "network.h"
  80 #include "calcmu.h"
  81 #include "constr.h"
  82 #include "xvgr.h"
  83 #include "trnio.h"
  84 #include "xtcio.h"
  85 #include "copyrite.h"
  86 #include "pull_rotation.h"
  87 #include "gmx_random.h"
  88 #include "mpelogging.h"
  89 #include "domdec.h"
  90 #include "partdec.h"
  91 #include "gmx_wallcycle.h"
  92 #include "genborn.h"
  93 #include "nbnxn_atomdata.h"
  94 #include "nbnxn_search.h"
  95 #include "nbnxn_kernels/nbnxn_kernel_ref.h"
  96 #include "nbnxn_kernels/nbnxn_kernel_simd_4xn.h"
  97 #include "nbnxn_kernels/nbnxn_kernel_simd_2xnn.h"
  98 #include "nbnxn_kernels/nbnxn_kernel_gpu_ref.h"
  99
 100 #ifdef GMX_LIB_MPI
 101 #include <mpi.h>
 102 #endif
 103 #ifdef GMX_THREAD_MPI
 104 #include "tmpi.h"
 105 #endif
 106
 107 #include "adress.h"
 108 #include "qmmm.h"
 109
 110 #include "nbnxn_cuda_data_mgmt.h"
 111 #include "nbnxn_cuda/nbnxn_cuda.h"
 112
 113 #if 0
 114 typedef struct gmx_timeprint {
 115
 116 } t_gmx_timeprint;
 117 #endif
 118
 119 /* Portable version of ctime_r implemented in src/gmxlib/string2.c, but we do not want it declared in public installed headers */
 120 GMX_LIBGMX_EXPORT
 121 char *
 122 gmx_ctime_r(const time_t *clock, char *buf, int n);
 123
 124
 125 double
 126 gmx_gettime()
 127 {
 128 #ifdef HAVE_GETTIMEOFDAY
 129     struct timeval t;
 130     double         seconds;
 131
 132     gettimeofday(&t, NULL);
 133
 134     seconds = (double) t.tv_sec + 1e-6*(double)t.tv_usec;
 135
 136     return seconds;
 137 #else
 138     double  seconds;
 139
 140     seconds = time(NULL);
 141
 142     return seconds;
 143 #endif
 144 }
 145
 146
 147 #define difftime(end, start) ((double)(end)-(double)(start))
 148
 149 void print_time(FILE *out, gmx_runtime_t *runtime, gmx_large_int_t step,
 150                 t_inputrec *ir, t_commrec *cr)
 151 {
 152     time_t finish;
 153     char   timebuf[STRLEN];
 154     double dt;
 155     char   buf[48];
 156
 157 #ifndef GMX_THREAD_MPI
 158     if (!PAR(cr))
 159 #endif
 160     {
 161         fprintf(out, "\r");
 162     }
 163     fprintf(out, "step %s", gmx_step_str(step, buf));
 164     if ((step >= ir->nstlist))
 165     {
 166         runtime->last          = gmx_gettime();
 167         dt                     = difftime(runtime->last, runtime->real);
 168         runtime->time_per_step = dt/(step - ir->init_step + 1);
 169
 170         dt = (ir->nsteps + ir->init_step - step)*runtime->time_per_step;
 171
 172         if (ir->nsteps >= 0)
 173         {
 174             if (dt >= 300)
 175             {
 176                 finish = (time_t) (runtime->last + dt);
 177                 gmx_ctime_r(&finish, timebuf, STRLEN);
 178                 sprintf(buf, "%s", timebuf);
 179                 buf[strlen(buf)-1] = '\0';
 180                 fprintf(out, ", will finish %s", buf);
 181             }
 182             else
 183             {
 184                 fprintf(out, ", remaining runtime: %5d s          ", (int)dt);
 185             }
 186         }
 187         else
 188         {
 189             fprintf(out, " performance: %.1f ns/day    ",
 190                     ir->delta_t/1000*24*60*60/runtime->time_per_step);
 191         }
 192     }
 193 #ifndef GMX_THREAD_MPI
 194     if (PAR(cr))
 195     {
 196         fprintf(out, "\n");
 197     }
 198 #endif
 199
 200     fflush(out);
 201 }
 202
 203 #ifdef NO_CLOCK
 204 #define clock() -1
 205 #endif
 206
 207 static double set_proctime(gmx_runtime_t *runtime)
 208 {
 209     double diff;
 210 #ifdef GMX_CRAY_XT3
 211     double prev;
 212
 213     prev          = runtime->proc;
 214     runtime->proc = dclock();
 215
 216     diff = runtime->proc - prev;
 217 #else
 218     clock_t prev;
 219
 220     prev          = runtime->proc;
 221     runtime->proc = clock();
 222
 223     diff = (double)(runtime->proc - prev)/(double)CLOCKS_PER_SEC;
 224 #endif
 225     if (diff < 0)
 226     {
 227         /* The counter has probably looped, ignore this data */
 228         diff = 0;
 229     }
 230
 231     return diff;
 232 }
 233
 234 void runtime_start(gmx_runtime_t *runtime)
 235 {
 236     runtime->real          = gmx_gettime();
 237     runtime->proc          = 0;
 238     set_proctime(runtime);
 239     runtime->realtime      = 0;
 240     runtime->proctime      = 0;
 241     runtime->last          = 0;
 242     runtime->time_per_step = 0;
 243 }
 244
 245 void runtime_end(gmx_runtime_t *runtime)
 246 {
 247     double now;
 248
 249     now = gmx_gettime();
 250
 251     runtime->proctime += set_proctime(runtime);
 252     runtime->realtime  = now - runtime->real;
 253     runtime->real      = now;
 254 }
 255
 256 void runtime_upd_proc(gmx_runtime_t *runtime)
 257 {
 258     runtime->proctime += set_proctime(runtime);
 259 }
 260
 261 void print_date_and_time(FILE *fplog, int nodeid, const char *title,
 262                          const gmx_runtime_t *runtime)
 263 {
 264     int    i;
 265     char   timebuf[STRLEN];
 266     char   time_string[STRLEN];
 267     time_t tmptime;
 268
 269     if (fplog)
 270     {
 271         if (runtime != NULL)
 272         {
 273             tmptime = (time_t) runtime->real;
 274             gmx_ctime_r(&tmptime, timebuf, STRLEN);
 275         }
 276         else
 277         {
 278             tmptime = (time_t) gmx_gettime();
 279             gmx_ctime_r(&tmptime, timebuf, STRLEN);
 280         }
 281         for (i = 0; timebuf[i] >= ' '; i++)
 282         {
 283             time_string[i] = timebuf[i];
 284         }
 285         time_string[i] = '\0';
 286
 287         fprintf(fplog, "%s on node %d %s\n", title, nodeid, time_string);
 288     }
 289 }
 290
 291 static void sum_forces(int start, int end, rvec f[], rvec flr[])
 292 {
 293     int i;
 294
 295     if (gmx_debug_at)
 296     {
 297         pr_rvecs(debug, 0, "fsr", f+start, end-start);
 298         pr_rvecs(debug, 0, "flr", flr+start, end-start);
 299     }
 300     for (i = start; (i < end); i++)
 301     {
 302         rvec_inc(f[i], flr[i]);
 303     }
 304 }
 305
 306 /*
 307  * calc_f_el calculates forces due to an electric field.
 308  *
 309  * force is kJ mol^-1 nm^-1 = e * kJ mol^-1 nm^-1 / e
 310  *
 311  * Et[] contains the parameters for the time dependent
 312  * part of the field (not yet used).
 313  * Ex[] contains the parameters for
 314  * the spatial dependent part of the field. You can have cool periodic
 315  * fields in principle, but only a constant field is supported
 316  * now.
 317  * The function should return the energy due to the electric field
 318  * (if any) but for now returns 0.
 319  *
 320  * WARNING:
 321  * There can be problems with the virial.
 322  * Since the field is not self-consistent this is unavoidable.
 323  * For neutral molecules the virial is correct within this approximation.
 324  * For neutral systems with many charged molecules the error is small.
 325  * But for systems with a net charge or a few charged molecules
 326  * the error can be significant when the field is high.
 327  * Solution: implement a self-consitent electric field into PME.
 328  */
 329 static void calc_f_el(FILE *fp, int  start, int homenr,
 330                       real charge[], rvec x[], rvec f[],
 331                       t_cosines Ex[], t_cosines Et[], double t)
 332 {
 333     rvec Ext;
 334     real t0;
 335     int  i, m;
 336
 337     for (m = 0; (m < DIM); m++)
 338     {
 339         if (Et[m].n > 0)
 340         {
 341             if (Et[m].n == 3)
 342             {
 343                 t0     = Et[m].a[1];
 344                 Ext[m] = cos(Et[m].a[0]*(t-t0))*exp(-sqr(t-t0)/(2.0*sqr(Et[m].a[2])));
 345             }
 346             else
 347             {
 348                 Ext[m] = cos(Et[m].a[0]*t);
 349             }
 350         }
 351         else
 352         {
 353             Ext[m] = 1.0;
 354         }
 355         if (Ex[m].n > 0)
 356         {
 357             /* Convert the field strength from V/nm to MD-units */
 358             Ext[m] *= Ex[m].a[0]*FIELDFAC;
 359             for (i = start; (i < start+homenr); i++)
 360             {
 361                 f[i][m] += charge[i]*Ext[m];
 362             }
 363         }
 364         else
 365         {
 366             Ext[m] = 0;
 367         }
 368     }
 369     if (fp != NULL)
 370     {
 371         fprintf(fp, "%10g  %10g  %10g  %10g #FIELD\n", t,
 372                 Ext[XX]/FIELDFAC, Ext[YY]/FIELDFAC, Ext[ZZ]/FIELDFAC);
 373     }
 374 }
 375
 376 static void calc_virial(FILE *fplog, int start, int homenr, rvec x[], rvec f[],
 377                         tensor vir_part, t_graph *graph, matrix box,
 378                         t_nrnb *nrnb, const t_forcerec *fr, int ePBC)
 379 {
 380     int    i, j;
 381     tensor virtest;
 382
 383     /* The short-range virial from surrounding boxes */
 384     clear_mat(vir_part);
 385     calc_vir(fplog, SHIFTS, fr->shift_vec, fr->fshift, vir_part, ePBC == epbcSCREW, box);
 386     inc_nrnb(nrnb, eNR_VIRIAL, SHIFTS);
 387
 388     /* Calculate partial virial, for local atoms only, based on short range.
 389      * Total virial is computed in global_stat, called from do_md
 390      */
 391     f_calc_vir(fplog, start, start+homenr, x, f, vir_part, graph, box);
 392     inc_nrnb(nrnb, eNR_VIRIAL, homenr);
 393
 394     /* Add position restraint contribution */
 395     for (i = 0; i < DIM; i++)
 396     {
 397         vir_part[i][i] += fr->vir_diag_posres[i];
 398     }
 399
 400     /* Add wall contribution */
 401     for (i = 0; i < DIM; i++)
 402     {
 403         vir_part[i][ZZ] += fr->vir_wall_z[i];
 404     }
 405
 406     if (debug)
 407     {
 408         pr_rvecs(debug, 0, "vir_part", vir_part, DIM);
 409     }
 410 }
 411
 412 static void posres_wrapper(FILE *fplog,
 413                            int flags,
 414                            gmx_bool bSepDVDL,
 415                            t_inputrec *ir,
 416                            t_nrnb *nrnb,
 417                            gmx_localtop_t *top,
 418                            matrix box, rvec x[],
 419                            rvec f[],
 420                            gmx_enerdata_t *enerd,
 421                            real *lambda,
 422                            t_forcerec *fr)
 423 {
 424     t_pbc pbc;
 425     real  v, dvdl;
 426     int   i;
 427
 428     /* Position restraints always require full pbc */
 429     set_pbc(&pbc, ir->ePBC, box);
 430     dvdl = 0;
 431     v    = posres(top->idef.il[F_POSRES].nr, top->idef.il[F_POSRES].iatoms,
 432                   top->idef.iparams_posres,
 433                   (const rvec*)x, fr->f_novirsum, fr->vir_diag_posres,
 434                   ir->ePBC == epbcNONE ? NULL : &pbc,
 435                   lambda[efptRESTRAINT], &dvdl,
 436                   fr->rc_scaling, fr->ePBC, fr->posres_com, fr->posres_comB);
 437     if (bSepDVDL)
 438     {
 439         fprintf(fplog, sepdvdlformat,
 440                 interaction_function[F_POSRES].longname, v, dvdl);
 441     }
 442     enerd->term[F_POSRES] += v;
 443     /* If just the force constant changes, the FEP term is linear,
 444      * but if k changes, it is not.
 445      */
 446     enerd->dvdl_nonlin[efptRESTRAINT] += dvdl;
 447     inc_nrnb(nrnb, eNR_POSRES, top->idef.il[F_POSRES].nr/2);
 448
 449     if ((ir->fepvals->n_lambda > 0) && (flags & GMX_FORCE_DHDL))
 450     {
 451         for (i = 0; i < enerd->n_lambda; i++)
 452         {
 453             real dvdl_dum, lambda_dum;
 454
 455             lambda_dum = (i == 0 ? lambda[efptRESTRAINT] : ir->fepvals->all_lambda[efptRESTRAINT][i-1]);
 456             v          = posres(top->idef.il[F_POSRES].nr, top->idef.il[F_POSRES].iatoms,
 457                                 top->idef.iparams_posres,
 458                                 (const rvec*)x, NULL, NULL,
 459                                 ir->ePBC == epbcNONE ? NULL : &pbc, lambda_dum, &dvdl,
 460                                 fr->rc_scaling, fr->ePBC, fr->posres_com, fr->posres_comB);
 461             enerd->enerpart_lambda[i] += v;
 462         }
 463     }
 464 }
 465
 466 static void pull_potential_wrapper(FILE *fplog,
 467                                    gmx_bool bSepDVDL,
 468                                    t_commrec *cr,
 469                                    t_inputrec *ir,
 470                                    matrix box, rvec x[],
 471                                    rvec f[],
 472                                    tensor vir_force,
 473                                    t_mdatoms *mdatoms,
 474                                    gmx_enerdata_t *enerd,
 475                                    real *lambda,
 476                                    double t)
 477 {
 478     t_pbc  pbc;
 479     real   dvdl;
 480
 481     /* Calculate the center of mass forces, this requires communication,
 482      * which is why pull_potential is called close to other communication.
 483      * The virial contribution is calculated directly,
 484      * which is why we call pull_potential after calc_virial.
 485      */
 486     set_pbc(&pbc, ir->ePBC, box);
 487     dvdl                     = 0;
 488     enerd->term[F_COM_PULL] +=
 489         pull_potential(ir->ePull, ir->pull, mdatoms, &pbc,
 490                        cr, t, lambda[efptRESTRAINT], x, f, vir_force, &dvdl);
 491     if (bSepDVDL)
 492     {
 493         fprintf(fplog, sepdvdlformat, "Com pull", enerd->term[F_COM_PULL], dvdl);
 494     }
 495     enerd->dvdl_lin[efptRESTRAINT] += dvdl;
 496 }
 497
 498 static void pme_receive_force_ener(FILE           *fplog,
 499                                    gmx_bool        bSepDVDL,
 500                                    t_commrec      *cr,
 501                                    gmx_wallcycle_t wcycle,
 502                                    gmx_enerdata_t *enerd,
 503                                    t_forcerec     *fr)
 504 {
 505     real   e, v, dvdl;
 506     float  cycles_ppdpme, cycles_seppme;
 507
 508     cycles_ppdpme = wallcycle_stop(wcycle, ewcPPDURINGPME);
 509     dd_cycles_add(cr->dd, cycles_ppdpme, ddCyclPPduringPME);
 510
 511     /* In case of node-splitting, the PP nodes receive the long-range
 512      * forces, virial and energy from the PME nodes here.
 513      */
 514     wallcycle_start(wcycle, ewcPP_PMEWAITRECVF);
 515     dvdl = 0;
 516     gmx_pme_receive_f(cr, fr->f_novirsum, fr->vir_el_recip, &e, &dvdl,
 517                       &cycles_seppme);
 518     if (bSepDVDL)
 519     {
 520         fprintf(fplog, sepdvdlformat, "PME mesh", e, dvdl);
 521     }
 522     enerd->term[F_COUL_RECIP] += e;
 523     enerd->dvdl_lin[efptCOUL] += dvdl;
 524     if (wcycle)
 525     {
 526         dd_cycles_add(cr->dd, cycles_seppme, ddCyclPME);
 527     }
 528     wallcycle_stop(wcycle, ewcPP_PMEWAITRECVF);
 529 }
 530
 531 static void print_large_forces(FILE *fp, t_mdatoms *md, t_commrec *cr,
 532                                gmx_large_int_t step, real pforce, rvec *x, rvec *f)
 533 {
 534     int  i;
 535     real pf2, fn2;
 536     char buf[STEPSTRSIZE];
 537
 538     pf2 = sqr(pforce);
 539     for (i = md->start; i < md->start+md->homenr; i++)
 540     {
 541         fn2 = norm2(f[i]);
 542         /* We also catch NAN, if the compiler does not optimize this away. */
 543         if (fn2 >= pf2 || fn2 != fn2)
 544         {
 545             fprintf(fp, "step %s  atom %6d  x %8.3f %8.3f %8.3f  force %12.5e\n",
 546                     gmx_step_str(step, buf),
 547                     ddglatnr(cr->dd, i), x[i][XX], x[i][YY], x[i][ZZ], sqrt(fn2));
 548         }
 549     }
 550 }
 551
 552 static void post_process_forces(FILE *fplog,
 553                                 t_commrec *cr,
 554                                 gmx_large_int_t step,
 555                                 t_nrnb *nrnb, gmx_wallcycle_t wcycle,
 556                                 gmx_localtop_t *top,
 557                                 matrix box, rvec x[],
 558                                 rvec f[],
 559                                 tensor vir_force,
 560                                 t_mdatoms *mdatoms,
 561                                 t_graph *graph,
 562                                 t_forcerec *fr, gmx_vsite_t *vsite,
 563                                 int flags)
 564 {
 565     if (fr->bF_NoVirSum)
 566     {
 567         if (vsite)
 568         {
 569             /* Spread the mesh force on virtual sites to the other particles...
 570              * This is parallellized. MPI communication is performed
 571              * if the constructing atoms aren't local.
 572              */
 573             wallcycle_start(wcycle, ewcVSITESPREAD);
 574             spread_vsite_f(fplog, vsite, x, fr->f_novirsum, NULL,
 575                            (flags & GMX_FORCE_VIRIAL), fr->vir_el_recip,
 576                            nrnb,
 577                            &top->idef, fr->ePBC, fr->bMolPBC, graph, box, cr);
 578             wallcycle_stop(wcycle, ewcVSITESPREAD);
 579         }
 580         if (flags & GMX_FORCE_VIRIAL)
 581         {
 582             /* Now add the forces, this is local */
 583             if (fr->bDomDec)
 584             {
 585                 sum_forces(0, fr->f_novirsum_n, f, fr->f_novirsum);
 586             }
 587             else
 588             {
 589                 sum_forces(mdatoms->start, mdatoms->start+mdatoms->homenr,
 590                            f, fr->f_novirsum);
 591             }
 592             if (EEL_FULL(fr->eeltype))
 593             {
 594                 /* Add the mesh contribution to the virial */
 595                 m_add(vir_force, fr->vir_el_recip, vir_force);
 596             }
 597             if (debug)
 598             {
 599                 pr_rvecs(debug, 0, "vir_force", vir_force, DIM);
 600             }
 601         }
 602     }
 603
 604     if (fr->print_force >= 0)
 605     {
 606         print_large_forces(stderr, mdatoms, cr, step, fr->print_force, x, f);
 607     }
 608 }
 609
 610 static void do_nb_verlet(t_forcerec *fr,
 611                          interaction_const_t *ic,
 612                          gmx_enerdata_t *enerd,
 613                          int flags, int ilocality,
 614                          int clearF,
 615                          t_nrnb *nrnb,
 616                          gmx_wallcycle_t wcycle)
 617 {
 618     int                        nnbl, kernel_type, enr_nbnxn_kernel_ljc, enr_nbnxn_kernel_lj;
 619     char                      *env;
 620     nonbonded_verlet_group_t  *nbvg;
 621
 622     if (!(flags & GMX_FORCE_NONBONDED))
 623     {
 624         /* skip non-bonded calculation */
 625         return;
 626     }
 627
 628     nbvg = &fr->nbv->grp[ilocality];
 629
 630     /* CUDA kernel launch overhead is already timed separately */
 631     if (fr->cutoff_scheme != ecutsVERLET)
 632     {
 633         gmx_incons("Invalid cut-off scheme passed!");
 634     }
 635
 636     if (nbvg->kernel_type != nbnxnk8x8x8_CUDA)
 637     {
 638         wallcycle_sub_start(wcycle, ewcsNONBONDED);
 639     }
 640     switch (nbvg->kernel_type)
 641     {
 642         case nbnxnk4x4_PlainC:
 643             nbnxn_kernel_ref(&nbvg->nbl_lists,
 644                              nbvg->nbat, ic,
 645                              fr->shift_vec,
 646                              flags,
 647                              clearF,
 648                              fr->fshift[0],
 649                              enerd->grpp.ener[egCOULSR],
 650                              fr->bBHAM ?
 651                              enerd->grpp.ener[egBHAMSR] :
 652                              enerd->grpp.ener[egLJSR]);
 653             break;
 654
 655         case nbnxnk4xN_SIMD_4xN:
 656             nbnxn_kernel_simd_4xn(&nbvg->nbl_lists,
 657                                   nbvg->nbat, ic,
 658                                   nbvg->ewald_excl,
 659                                   fr->shift_vec,
 660                                   flags,
 661                                   clearF,
 662                                   fr->fshift[0],
 663                                   enerd->grpp.ener[egCOULSR],
 664                                   fr->bBHAM ?
 665                                   enerd->grpp.ener[egBHAMSR] :
 666                                   enerd->grpp.ener[egLJSR]);
 667             break;
 668         case nbnxnk4xN_SIMD_2xNN:
 669             nbnxn_kernel_simd_2xnn(&nbvg->nbl_lists,
 670                                    nbvg->nbat, ic,
 671                                    nbvg->ewald_excl,
 672                                    fr->shift_vec,
 673                                    flags,
 674                                    clearF,
 675                                    fr->fshift[0],
 676                                    enerd->grpp.ener[egCOULSR],
 677                                    fr->bBHAM ?
 678                                    enerd->grpp.ener[egBHAMSR] :
 679                                    enerd->grpp.ener[egLJSR]);
 680             break;
 681
 682         case nbnxnk8x8x8_CUDA:
 683             nbnxn_cuda_launch_kernel(fr->nbv->cu_nbv, nbvg->nbat, flags, ilocality);
 684             break;
 685
 686         case nbnxnk8x8x8_PlainC:
 687             nbnxn_kernel_gpu_ref(nbvg->nbl_lists.nbl[0],
 688                                  nbvg->nbat, ic,
 689                                  fr->shift_vec,
 690                                  flags,
 691                                  clearF,
 692                                  nbvg->nbat->out[0].f,
 693                                  fr->fshift[0],
 694                                  enerd->grpp.ener[egCOULSR],
 695                                  fr->bBHAM ?
 696                                  enerd->grpp.ener[egBHAMSR] :
 697                                  enerd->grpp.ener[egLJSR]);
 698             break;
 699
 700         default:
 701             gmx_incons("Invalid nonbonded kernel type passed!");
 702
 703     }
 704     if (nbvg->kernel_type != nbnxnk8x8x8_CUDA)
 705     {
 706         wallcycle_sub_stop(wcycle, ewcsNONBONDED);
 707     }
 708
 709     if (EEL_RF(ic->eeltype) || ic->eeltype == eelCUT)
 710     {
 711         enr_nbnxn_kernel_ljc = eNR_NBNXN_LJ_RF;
 712     }
 713     else if (nbvg->ewald_excl == ewaldexclTable)
 714     {
 715         enr_nbnxn_kernel_ljc = eNR_NBNXN_LJ_TAB;
 716     }
 717     else
 718     {
 719         enr_nbnxn_kernel_ljc = eNR_NBNXN_LJ_EWALD;
 720     }
 721     enr_nbnxn_kernel_lj = eNR_NBNXN_LJ;
 722     if (flags & GMX_FORCE_ENERGY)
 723     {
 724         /* In eNR_??? the nbnxn F+E kernels are always the F kernel + 1 */
 725         enr_nbnxn_kernel_ljc += 1;
 726         enr_nbnxn_kernel_lj  += 1;
 727     }
 728
 729     inc_nrnb(nrnb, enr_nbnxn_kernel_ljc,
 730              nbvg->nbl_lists.natpair_ljq);
 731     inc_nrnb(nrnb, enr_nbnxn_kernel_lj,
 732              nbvg->nbl_lists.natpair_lj);
 733     inc_nrnb(nrnb, enr_nbnxn_kernel_ljc-eNR_NBNXN_LJ_RF+eNR_NBNXN_RF,
 734              nbvg->nbl_lists.natpair_q);
 735 }
 736
 737 void do_force_cutsVERLET(FILE *fplog, t_commrec *cr,
 738                          t_inputrec *inputrec,
 739                          gmx_large_int_t step, t_nrnb *nrnb, gmx_wallcycle_t wcycle,
 740                          gmx_localtop_t *top,
 741                          gmx_mtop_t *mtop,
 742                          gmx_groups_t *groups,
 743                          matrix box, rvec x[], history_t *hist,
 744                          rvec f[],
 745                          tensor vir_force,
 746                          t_mdatoms *mdatoms,
 747                          gmx_enerdata_t *enerd, t_fcdata *fcd,
 748                          real *lambda, t_graph *graph,
 749                          t_forcerec *fr, interaction_const_t *ic,
 750                          gmx_vsite_t *vsite, rvec mu_tot,
 751                          double t, FILE *field, gmx_edsam_t ed,
 752                          gmx_bool bBornRadii,
 753                          int flags)
 754 {
 755     int                 cg0, cg1, i, j;
 756     int                 start, homenr;
 757     int                 nb_kernel_type;
 758     double              mu[2*DIM];
 759     gmx_bool            bSepDVDL, bStateChanged, bNS, bFillGrid, bCalcCGCM, bBS;
 760     gmx_bool            bDoLongRange, bDoForces, bSepLRF, bUseGPU, bUseOrEmulGPU;
 761     gmx_bool            bDiffKernels = FALSE;
 762     matrix              boxs;
 763     rvec                vzero, box_diag;
 764     real                e, v, dvdl;
 765     float               cycles_pme, cycles_force;
 766     nonbonded_verlet_t *nbv;
 767
 768     cycles_force   = 0;
 769     nbv            = fr->nbv;
 770     nb_kernel_type = fr->nbv->grp[0].kernel_type;
 771
 772     start  = mdatoms->start;
 773     homenr = mdatoms->homenr;
 774
 775     bSepDVDL = (fr->bSepDVDL && do_per_step(step, inputrec->nstlog));
 776
 777     clear_mat(vir_force);
 778
 779     cg0 = 0;
 780     if (DOMAINDECOMP(cr))
 781     {
 782         cg1 = cr->dd->ncg_tot;
 783     }
 784     else
 785     {
 786         cg1 = top->cgs.nr;
 787     }
 788     if (fr->n_tpi > 0)
 789     {
 790         cg1--;
 791     }
 792
 793     bStateChanged = (flags & GMX_FORCE_STATECHANGED);
 794     bNS           = (flags & GMX_FORCE_NS) && (fr->bAllvsAll == FALSE);
 795     bFillGrid     = (bNS && bStateChanged);
 796     bCalcCGCM     = (bFillGrid && !DOMAINDECOMP(cr));
 797     bDoLongRange  = (fr->bTwinRange && bNS && (flags & GMX_FORCE_DO_LR));
 798     bDoForces     = (flags & GMX_FORCE_FORCES);
 799     bSepLRF       = (bDoLongRange && bDoForces && (flags & GMX_FORCE_SEPLRF));
 800     bUseGPU       = fr->nbv->bUseGPU;
 801     bUseOrEmulGPU = bUseGPU || (nbv->grp[0].kernel_type == nbnxnk8x8x8_PlainC);
 802
 803     if (bStateChanged)
 804     {
 805         update_forcerec(fplog, fr, box);
 806
 807         if (NEED_MUTOT(*inputrec))
 808         {
 809             /* Calculate total (local) dipole moment in a temporary common array.
 810              * This makes it possible to sum them over nodes faster.
 811              */
 812             calc_mu(start, homenr,
 813                     x, mdatoms->chargeA, mdatoms->chargeB, mdatoms->nChargePerturbed,
 814                     mu, mu+DIM);
 815         }
 816     }
 817
 818     if (fr->ePBC != epbcNONE)
 819     {
 820         /* Compute shift vectors every step,
 821          * because of pressure coupling or box deformation!
 822          */
 823         if ((flags & GMX_FORCE_DYNAMICBOX) && bStateChanged)
 824         {
 825             calc_shifts(box, fr->shift_vec);
 826         }
 827
 828         if (bCalcCGCM)
 829         {
 830             put_atoms_in_box_omp(fr->ePBC, box, homenr, x);
 831             inc_nrnb(nrnb, eNR_SHIFTX, homenr);
 832         }
 833         else if (EI_ENERGY_MINIMIZATION(inputrec->eI) && graph)
 834         {
 835             unshift_self(graph, box, x);
 836         }
 837     }
 838
 839     nbnxn_atomdata_copy_shiftvec(flags & GMX_FORCE_DYNAMICBOX,
 840                                  fr->shift_vec, nbv->grp[0].nbat);
 841
 842 #ifdef GMX_MPI
 843     if (!(cr->duty & DUTY_PME))
 844     {
 845         /* Send particle coordinates to the pme nodes.
 846          * Since this is only implemented for domain decomposition
 847          * and domain decomposition does not use the graph,
 848          * we do not need to worry about shifting.
 849          */
 850
 851         wallcycle_start(wcycle, ewcPP_PMESENDX);
 852         GMX_MPE_LOG(ev_send_coordinates_start);
 853
 854         bBS = (inputrec->nwall == 2);
 855         if (bBS)
 856         {
 857             copy_mat(box, boxs);
 858             svmul(inputrec->wall_ewald_zfac, boxs[ZZ], boxs[ZZ]);
 859         }
 860
 861         gmx_pme_send_x(cr, bBS ? boxs : box, x,
 862                        mdatoms->nChargePerturbed, lambda[efptCOUL],
 863                        (flags & (GMX_FORCE_VIRIAL | GMX_FORCE_ENERGY)), step);
 864
 865         GMX_MPE_LOG(ev_send_coordinates_finish);
 866         wallcycle_stop(wcycle, ewcPP_PMESENDX);
 867     }
 868 #endif /* GMX_MPI */
 869
 870     /* do gridding for pair search */
 871     if (bNS)
 872     {
 873         if (graph && bStateChanged)
 874         {
 875             /* Calculate intramolecular shift vectors to make molecules whole */
 876             mk_mshift(fplog, graph, fr->ePBC, box, x);
 877         }
 878
 879         clear_rvec(vzero);
 880         box_diag[XX] = box[XX][XX];
 881         box_diag[YY] = box[YY][YY];
 882         box_diag[ZZ] = box[ZZ][ZZ];
 883
 884         wallcycle_start(wcycle, ewcNS);
 885         if (!fr->bDomDec)
 886         {
 887             wallcycle_sub_start(wcycle, ewcsNBS_GRID_LOCAL);
 888             nbnxn_put_on_grid(nbv->nbs, fr->ePBC, box,
 889                               0, vzero, box_diag,
 890                               0, mdatoms->homenr, -1, fr->cginfo, x,
 891                               0, NULL,
 892                               nbv->grp[eintLocal].kernel_type,
 893                               nbv->grp[eintLocal].nbat);
 894             wallcycle_sub_stop(wcycle, ewcsNBS_GRID_LOCAL);
 895         }
 896         else
 897         {
 898             wallcycle_sub_start(wcycle, ewcsNBS_GRID_NONLOCAL);
 899             nbnxn_put_on_grid_nonlocal(nbv->nbs, domdec_zones(cr->dd),
 900                                        fr->cginfo, x,
 901                                        nbv->grp[eintNonlocal].kernel_type,
 902                                        nbv->grp[eintNonlocal].nbat);
 903             wallcycle_sub_stop(wcycle, ewcsNBS_GRID_NONLOCAL);
 904         }
 905
 906         if (nbv->ngrp == 1 ||
 907             nbv->grp[eintNonlocal].nbat == nbv->grp[eintLocal].nbat)
 908         {
 909             nbnxn_atomdata_set(nbv->grp[eintLocal].nbat, eatAll,
 910                                nbv->nbs, mdatoms, fr->cginfo);
 911         }
 912         else
 913         {
 914             nbnxn_atomdata_set(nbv->grp[eintLocal].nbat, eatLocal,
 915                                nbv->nbs, mdatoms, fr->cginfo);
 916             nbnxn_atomdata_set(nbv->grp[eintNonlocal].nbat, eatAll,
 917                                nbv->nbs, mdatoms, fr->cginfo);
 918         }
 919         wallcycle_stop(wcycle, ewcNS);
 920     }
 921
 922     /* initialize the GPU atom data and copy shift vector */
 923     if (bUseGPU)
 924     {
 925         if (bNS)
 926         {
 927             wallcycle_start_nocount(wcycle, ewcLAUNCH_GPU_NB);
 928             nbnxn_cuda_init_atomdata(nbv->cu_nbv, nbv->grp[eintLocal].nbat);
 929             wallcycle_stop(wcycle, ewcLAUNCH_GPU_NB);
 930         }
 931
 932         wallcycle_start_nocount(wcycle, ewcLAUNCH_GPU_NB);
 933         nbnxn_cuda_upload_shiftvec(nbv->cu_nbv, nbv->grp[eintLocal].nbat);
 934         wallcycle_stop(wcycle, ewcLAUNCH_GPU_NB);
 935     }
 936
 937     /* do local pair search */
 938     if (bNS)
 939     {
 940         wallcycle_start_nocount(wcycle, ewcNS);
 941         wallcycle_sub_start(wcycle, ewcsNBS_SEARCH_LOCAL);
 942         nbnxn_make_pairlist(nbv->nbs, nbv->grp[eintLocal].nbat,
 943                             &top->excls,
 944                             ic->rlist,
 945                             nbv->min_ci_balanced,
 946                             &nbv->grp[eintLocal].nbl_lists,
 947                             eintLocal,
 948                             nbv->grp[eintLocal].kernel_type,
 949                             nrnb);
 950         wallcycle_sub_stop(wcycle, ewcsNBS_SEARCH_LOCAL);
 951
 952         if (bUseGPU)
 953         {
 954             /* initialize local pair-list on the GPU */
 955             nbnxn_cuda_init_pairlist(nbv->cu_nbv,
 956                                      nbv->grp[eintLocal].nbl_lists.nbl[0],
 957                                      eintLocal);
 958         }
 959         wallcycle_stop(wcycle, ewcNS);
 960     }
 961     else
 962     {
 963         wallcycle_start(wcycle, ewcNB_XF_BUF_OPS);
 964         wallcycle_sub_start(wcycle, ewcsNB_X_BUF_OPS);
 965         nbnxn_atomdata_copy_x_to_nbat_x(nbv->nbs, eatLocal, FALSE, x,
 966                                         nbv->grp[eintLocal].nbat);
 967         wallcycle_sub_stop(wcycle, ewcsNB_X_BUF_OPS);
 968         wallcycle_stop(wcycle, ewcNB_XF_BUF_OPS);
 969     }
 970
 971     if (bUseGPU)
 972     {
 973         wallcycle_start(wcycle, ewcLAUNCH_GPU_NB);
 974         /* launch local nonbonded F on GPU */
 975         do_nb_verlet(fr, ic, enerd, flags, eintLocal, enbvClearFNo,
 976                      nrnb, wcycle);
 977         wallcycle_stop(wcycle, ewcLAUNCH_GPU_NB);
 978     }
 979
 980     /* Communicate coordinates and sum dipole if necessary +
 981        do non-local pair search */
 982     if (DOMAINDECOMP(cr))
 983     {
 984         bDiffKernels = (nbv->grp[eintNonlocal].kernel_type !=
 985                         nbv->grp[eintLocal].kernel_type);
 986
 987         if (bDiffKernels)
 988         {
 989             /* With GPU+CPU non-bonded calculations we need to copy
 990              * the local coordinates to the non-local nbat struct
 991              * (in CPU format) as the non-local kernel call also
 992              * calculates the local - non-local interactions.
 993              */
 994             wallcycle_start(wcycle, ewcNB_XF_BUF_OPS);
 995             wallcycle_sub_start(wcycle, ewcsNB_X_BUF_OPS);
 996             nbnxn_atomdata_copy_x_to_nbat_x(nbv->nbs, eatLocal, TRUE, x,
 997                                             nbv->grp[eintNonlocal].nbat);
 998             wallcycle_sub_stop(wcycle, ewcsNB_X_BUF_OPS);
 999             wallcycle_stop(wcycle, ewcNB_XF_BUF_OPS);
1000         }
1001
1002         if (bNS)
1003         {
1004             wallcycle_start_nocount(wcycle, ewcNS);
1005             wallcycle_sub_start(wcycle, ewcsNBS_SEARCH_NONLOCAL);
1006
1007             if (bDiffKernels)
1008             {
1009                 nbnxn_grid_add_simple(nbv->nbs, nbv->grp[eintNonlocal].nbat);
1010             }
1011
1012             nbnxn_make_pairlist(nbv->nbs, nbv->grp[eintNonlocal].nbat,
1013                                 &top->excls,
1014                                 ic->rlist,
1015                                 nbv->min_ci_balanced,
1016                                 &nbv->grp[eintNonlocal].nbl_lists,
1017                                 eintNonlocal,
1018                                 nbv->grp[eintNonlocal].kernel_type,
1019                                 nrnb);
1020
1021             wallcycle_sub_stop(wcycle, ewcsNBS_SEARCH_NONLOCAL);
1022
1023             if (nbv->grp[eintNonlocal].kernel_type == nbnxnk8x8x8_CUDA)
1024             {
1025                 /* initialize non-local pair-list on the GPU */
1026                 nbnxn_cuda_init_pairlist(nbv->cu_nbv,
1027                                          nbv->grp[eintNonlocal].nbl_lists.nbl[0],
1028                                          eintNonlocal);
1029             }
1030             wallcycle_stop(wcycle, ewcNS);
1031         }
1032         else
1033         {
1034             wallcycle_start(wcycle, ewcMOVEX);
1035             dd_move_x(cr->dd, box, x);
1036
1037             /* When we don't need the total dipole we sum it in global_stat */
1038             if (bStateChanged && NEED_MUTOT(*inputrec))
1039             {
1040                 gmx_sumd(2*DIM, mu, cr);
1041             }
1042             wallcycle_stop(wcycle, ewcMOVEX);
1043
1044             wallcycle_start(wcycle, ewcNB_XF_BUF_OPS);
1045             wallcycle_sub_start(wcycle, ewcsNB_X_BUF_OPS);
1046             nbnxn_atomdata_copy_x_to_nbat_x(nbv->nbs, eatNonlocal, FALSE, x,
1047                                             nbv->grp[eintNonlocal].nbat);
1048             wallcycle_sub_stop(wcycle, ewcsNB_X_BUF_OPS);
1049             cycles_force += wallcycle_stop(wcycle, ewcNB_XF_BUF_OPS);
1050         }
1051
1052         if (bUseGPU && !bDiffKernels)
1053         {
1054             wallcycle_start(wcycle, ewcLAUNCH_GPU_NB);
1055             /* launch non-local nonbonded F on GPU */
1056             do_nb_verlet(fr, ic, enerd, flags, eintNonlocal, enbvClearFNo,
1057                          nrnb, wcycle);
1058             cycles_force += wallcycle_stop(wcycle, ewcLAUNCH_GPU_NB);
1059         }
1060     }
1061
1062     if (bUseGPU)
1063     {
1064         /* launch D2H copy-back F */
1065         wallcycle_start_nocount(wcycle, ewcLAUNCH_GPU_NB);
1066         if (DOMAINDECOMP(cr) && !bDiffKernels)
1067         {
1068             nbnxn_cuda_launch_cpyback(nbv->cu_nbv, nbv->grp[eintNonlocal].nbat,
1069                                       flags, eatNonlocal);
1070         }
1071         nbnxn_cuda_launch_cpyback(nbv->cu_nbv, nbv->grp[eintLocal].nbat,
1072                                   flags, eatLocal);
1073         cycles_force += wallcycle_stop(wcycle, ewcLAUNCH_GPU_NB);
1074     }
1075
1076     if (bStateChanged && NEED_MUTOT(*inputrec))
1077     {
1078         if (PAR(cr))
1079         {
1080             gmx_sumd(2*DIM, mu, cr);
1081         }
1082
1083         for (i = 0; i < 2; i++)
1084         {
1085             for (j = 0; j < DIM; j++)
1086             {
1087                 fr->mu_tot[i][j] = mu[i*DIM + j];
1088             }
1089         }
1090     }
1091     if (fr->efep == efepNO)
1092     {
1093         copy_rvec(fr->mu_tot[0], mu_tot);
1094     }
1095     else
1096     {
1097         for (j = 0; j < DIM; j++)
1098         {
1099             mu_tot[j] =
1100                 (1.0 - lambda[efptCOUL])*fr->mu_tot[0][j] +
1101                 lambda[efptCOUL]*fr->mu_tot[1][j];
1102         }
1103     }
1104
1105     /* Reset energies */
1106     reset_enerdata(&(inputrec->opts), fr, bNS, enerd, MASTER(cr));
1107     clear_rvecs(SHIFTS, fr->fshift);
1108
1109     if (DOMAINDECOMP(cr))
1110     {
1111         if (!(cr->duty & DUTY_PME))
1112         {
1113             wallcycle_start(wcycle, ewcPPDURINGPME);
1114             dd_force_flop_start(cr->dd, nrnb);
1115         }
1116     }
1117
1118     if (inputrec->bRot)
1119     {
1120         /* Enforced rotation has its own cycle counter that starts after the collective
1121          * coordinates have been communicated. It is added to ddCyclF to allow
1122          * for proper load-balancing */
1123         wallcycle_start(wcycle, ewcROT);
1124         do_rotation(cr, inputrec, box, x, t, step, wcycle, bNS);
1125         wallcycle_stop(wcycle, ewcROT);
1126     }
1127
1128     /* Start the force cycle counter.
1129      * This counter is stopped in do_forcelow_level.
1130      * No parallel communication should occur while this counter is running,
1131      * since that will interfere with the dynamic load balancing.
1132      */
1133     wallcycle_start(wcycle, ewcFORCE);
1134     if (bDoForces)
1135     {
1136         /* Reset forces for which the virial is calculated separately:
1137          * PME/Ewald forces if necessary */
1138         if (fr->bF_NoVirSum)
1139         {
1140             if (flags & GMX_FORCE_VIRIAL)
1141             {
1142                 fr->f_novirsum = fr->f_novirsum_alloc;
1143                 GMX_BARRIER(cr->mpi_comm_mygroup);
1144                 if (fr->bDomDec)
1145                 {
1146                     clear_rvecs(fr->f_novirsum_n, fr->f_novirsum);
1147                 }
1148                 else
1149                 {
1150                     clear_rvecs(homenr, fr->f_novirsum+start);
1151                 }
1152                 GMX_BARRIER(cr->mpi_comm_mygroup);
1153             }
1154             else
1155             {
1156                 /* We are not calculating the pressure so we do not need
1157                  * a separate array for forces that do not contribute
1158                  * to the pressure.
1159                  */
1160                 fr->f_novirsum = f;
1161             }
1162         }
1163
1164         /* Clear the short- and long-range forces */
1165         clear_rvecs(fr->natoms_force_constr, f);
1166         if (bSepLRF && do_per_step(step, inputrec->nstcalclr))
1167         {
1168             clear_rvecs(fr->natoms_force_constr, fr->f_twin);
1169         }
1170
1171         clear_rvec(fr->vir_diag_posres);
1172
1173         GMX_BARRIER(cr->mpi_comm_mygroup);
1174     }
1175
1176     if (inputrec->ePull == epullCONSTRAINT)
1177     {
1178         clear_pull_forces(inputrec->pull);
1179     }
1180
1181     /* We calculate the non-bonded forces, when done on the CPU, here.
1182      * We do this before calling do_force_lowlevel, as in there bondeds
1183      * forces are calculated before PME, which does communication.
1184      * With this order, non-bonded and bonded force calculation imbalance
1185      * can be balanced out by the domain decomposition load balancing.
1186      */
1187
1188     if (!bUseOrEmulGPU)
1189     {
1190         /* Maybe we should move this into do_force_lowlevel */
1191         do_nb_verlet(fr, ic, enerd, flags, eintLocal, enbvClearFYes,
1192                      nrnb, wcycle);
1193     }
1194
1195     if (!bUseOrEmulGPU || bDiffKernels)
1196     {
1197         int aloc;
1198
1199         if (DOMAINDECOMP(cr))
1200         {
1201             do_nb_verlet(fr, ic, enerd, flags, eintNonlocal,
1202                          bDiffKernels ? enbvClearFYes : enbvClearFNo,
1203                          nrnb, wcycle);
1204         }
1205
1206         if (!bUseOrEmulGPU)
1207         {
1208             aloc = eintLocal;
1209         }
1210         else
1211         {
1212             aloc = eintNonlocal;
1213         }
1214
1215         /* Add all the non-bonded force to the normal force array.
1216          * This can be split into a local a non-local part when overlapping
1217          * communication with calculation with domain decomposition.
1218          */
1219         cycles_force += wallcycle_stop(wcycle, ewcFORCE);
1220         wallcycle_start(wcycle, ewcNB_XF_BUF_OPS);
1221         wallcycle_sub_start(wcycle, ewcsNB_F_BUF_OPS);
1222         nbnxn_atomdata_add_nbat_f_to_f(nbv->nbs, eatAll, nbv->grp[aloc].nbat, f);
1223         wallcycle_sub_stop(wcycle, ewcsNB_F_BUF_OPS);
1224         cycles_force += wallcycle_stop(wcycle, ewcNB_XF_BUF_OPS);
1225         wallcycle_start_nocount(wcycle, ewcFORCE);
1226
1227         /* if there are multiple fshift output buffers reduce them */
1228         if ((flags & GMX_FORCE_VIRIAL) &&
1229             nbv->grp[aloc].nbl_lists.nnbl > 1)
1230         {
1231             nbnxn_atomdata_add_nbat_fshift_to_fshift(nbv->grp[aloc].nbat,
1232                                                      fr->fshift);
1233         }
1234     }
1235
1236     /* update QMMMrec, if necessary */
1237     if (fr->bQMMM)
1238     {
1239         update_QMMMrec(cr, fr, x, mdatoms, box, top);
1240     }
1241
1242     if ((flags & GMX_FORCE_BONDED) && top->idef.il[F_POSRES].nr > 0)
1243     {
1244         posres_wrapper(fplog, flags, bSepDVDL, inputrec, nrnb, top, box, x,
1245                        f, enerd, lambda, fr);
1246     }
1247
1248     /* Compute the bonded and non-bonded energies and optionally forces */
1249     do_force_lowlevel(fplog, step, fr, inputrec, &(top->idef),
1250                       cr, nrnb, wcycle, mdatoms, &(inputrec->opts),
1251                       x, hist, f, bSepLRF ? fr->f_twin : f, enerd, fcd, mtop, top, fr->born,
1252                       &(top->atomtypes), bBornRadii, box,
1253                       inputrec->fepvals, lambda, graph, &(top->excls), fr->mu_tot,
1254                       flags, &cycles_pme);
1255
1256     if (bSepLRF)
1257     {
1258         if (do_per_step(step, inputrec->nstcalclr))
1259         {
1260             /* Add the long range forces to the short range forces */
1261             for (i = 0; i < fr->natoms_force_constr; i++)
1262             {
1263                 rvec_add(fr->f_twin[i], f[i], f[i]);
1264             }
1265         }
1266     }
1267
1268     cycles_force += wallcycle_stop(wcycle, ewcFORCE);
1269     GMX_BARRIER(cr->mpi_comm_mygroup);
1270
1271     if (ed)
1272     {
1273         do_flood(cr, inputrec, x, f, ed, box, step, bNS);
1274     }
1275
1276     if (bUseOrEmulGPU && !bDiffKernels)
1277     {
1278         /* wait for non-local forces (or calculate in emulation mode) */
1279         if (DOMAINDECOMP(cr))
1280         {
1281             if (bUseGPU)
1282             {
1283                 wallcycle_start(wcycle, ewcWAIT_GPU_NB_NL);
1284                 nbnxn_cuda_wait_gpu(nbv->cu_nbv,
1285                                     nbv->grp[eintNonlocal].nbat,
1286                                     flags, eatNonlocal,
1287                                     enerd->grpp.ener[egLJSR], enerd->grpp.ener[egCOULSR],
1288                                     fr->fshift);
1289                 cycles_force += wallcycle_stop(wcycle, ewcWAIT_GPU_NB_NL);
1290             }
1291             else
1292             {
1293                 wallcycle_start_nocount(wcycle, ewcFORCE);
1294                 do_nb_verlet(fr, ic, enerd, flags, eintNonlocal, enbvClearFYes,
1295                              nrnb, wcycle);
1296                 cycles_force += wallcycle_stop(wcycle, ewcFORCE);
1297             }
1298             wallcycle_start(wcycle, ewcNB_XF_BUF_OPS);
1299             wallcycle_sub_start(wcycle, ewcsNB_F_BUF_OPS);
1300             /* skip the reduction if there was no non-local work to do */
1301             if (nbv->grp[eintLocal].nbl_lists.nbl[0]->nsci > 0)
1302             {
1303                 nbnxn_atomdata_add_nbat_f_to_f(nbv->nbs, eatNonlocal,
1304                                                nbv->grp[eintNonlocal].nbat, f);
1305             }
1306             wallcycle_sub_stop(wcycle, ewcsNB_F_BUF_OPS);
1307             cycles_force += wallcycle_stop(wcycle, ewcNB_XF_BUF_OPS);
1308         }
1309     }
1310
1311     if (bDoForces)
1312     {
1313         /* Communicate the forces */
1314         if (PAR(cr))
1315         {
1316             wallcycle_start(wcycle, ewcMOVEF);
1317             if (DOMAINDECOMP(cr))
1318             {
1319                 dd_move_f(cr->dd, f, fr->fshift);
1320                 /* Do we need to communicate the separate force array
1321                  * for terms that do not contribute to the single sum virial?
1322                  * Position restraints and electric fields do not introduce
1323                  * inter-cg forces, only full electrostatics methods do.
1324                  * When we do not calculate the virial, fr->f_novirsum = f,
1325                  * so we have already communicated these forces.
1326                  */
1327                 if (EEL_FULL(fr->eeltype) && cr->dd->n_intercg_excl &&
1328                     (flags & GMX_FORCE_VIRIAL))
1329                 {
1330                     dd_move_f(cr->dd, fr->f_novirsum, NULL);
1331                 }
1332                 if (bSepLRF)
1333                 {
1334                     /* We should not update the shift forces here,
1335                      * since f_twin is already included in f.
1336                      */
1337                     dd_move_f(cr->dd, fr->f_twin, NULL);
1338                 }
1339             }
1340             wallcycle_stop(wcycle, ewcMOVEF);
1341         }
1342     }
1343
1344     if (bUseOrEmulGPU)
1345     {
1346         /* wait for local forces (or calculate in emulation mode) */
1347         if (bUseGPU)
1348         {
1349             wallcycle_start(wcycle, ewcWAIT_GPU_NB_L);
1350             nbnxn_cuda_wait_gpu(nbv->cu_nbv,
1351                                 nbv->grp[eintLocal].nbat,
1352                                 flags, eatLocal,
1353                                 enerd->grpp.ener[egLJSR], enerd->grpp.ener[egCOULSR],
1354                                 fr->fshift);
1355             wallcycle_stop(wcycle, ewcWAIT_GPU_NB_L);
1356
1357             /* now clear the GPU outputs while we finish the step on the CPU */
1358
1359             wallcycle_start_nocount(wcycle, ewcLAUNCH_GPU_NB);
1360             nbnxn_cuda_clear_outputs(nbv->cu_nbv, flags);
1361             wallcycle_stop(wcycle, ewcLAUNCH_GPU_NB);
1362         }
1363         else
1364         {
1365             wallcycle_start_nocount(wcycle, ewcFORCE);
1366             do_nb_verlet(fr, ic, enerd, flags, eintLocal,
1367                          DOMAINDECOMP(cr) ? enbvClearFNo : enbvClearFYes,
1368                          nrnb, wcycle);
1369             wallcycle_stop(wcycle, ewcFORCE);
1370         }
1371         wallcycle_start(wcycle, ewcNB_XF_BUF_OPS);
1372         wallcycle_sub_start(wcycle, ewcsNB_F_BUF_OPS);
1373         if (nbv->grp[eintLocal].nbl_lists.nbl[0]->nsci > 0)
1374         {
1375             /* skip the reduction if there was no non-local work to do */
1376             nbnxn_atomdata_add_nbat_f_to_f(nbv->nbs, eatLocal,
1377                                            nbv->grp[eintLocal].nbat, f);
1378         }
1379         wallcycle_sub_stop(wcycle, ewcsNB_F_BUF_OPS);
1380         wallcycle_stop(wcycle, ewcNB_XF_BUF_OPS);
1381     }
1382
1383     if (DOMAINDECOMP(cr))
1384     {
1385         dd_force_flop_stop(cr->dd, nrnb);
1386         if (wcycle)
1387         {
1388             dd_cycles_add(cr->dd, cycles_force-cycles_pme, ddCyclF);
1389         }
1390     }
1391
1392     if (bDoForces)
1393     {
1394         if (IR_ELEC_FIELD(*inputrec))
1395         {
1396             /* Compute forces due to electric field */
1397             calc_f_el(MASTER(cr) ? field : NULL,
1398                       start, homenr, mdatoms->chargeA, x, fr->f_novirsum,
1399                       inputrec->ex, inputrec->et, t);
1400         }
1401
1402         /* If we have NoVirSum forces, but we do not calculate the virial,
1403          * we sum fr->f_novirum=f later.
1404          */
1405         if (vsite && !(fr->bF_NoVirSum && !(flags & GMX_FORCE_VIRIAL)))
1406         {
1407             wallcycle_start(wcycle, ewcVSITESPREAD);
1408             spread_vsite_f(fplog, vsite, x, f, fr->fshift, FALSE, NULL, nrnb,
1409                            &top->idef, fr->ePBC, fr->bMolPBC, graph, box, cr);
1410             wallcycle_stop(wcycle, ewcVSITESPREAD);
1411
1412             if (bSepLRF)
1413             {
1414                 wallcycle_start(wcycle, ewcVSITESPREAD);
1415                 spread_vsite_f(fplog, vsite, x, fr->f_twin, NULL, FALSE, NULL,
1416                                nrnb,
1417                                &top->idef, fr->ePBC, fr->bMolPBC, graph, box, cr);
1418                 wallcycle_stop(wcycle, ewcVSITESPREAD);
1419             }
1420         }
1421
1422         if (flags & GMX_FORCE_VIRIAL)
1423         {
1424             /* Calculation of the virial must be done after vsites! */
1425             calc_virial(fplog, mdatoms->start, mdatoms->homenr, x, f,
1426                         vir_force, graph, box, nrnb, fr, inputrec->ePBC);
1427         }
1428     }
1429
1430     if (inputrec->ePull == epullUMBRELLA || inputrec->ePull == epullCONST_F)
1431     {
1432         pull_potential_wrapper(fplog, bSepDVDL, cr, inputrec, box, x,
1433                                f, vir_force, mdatoms, enerd, lambda, t);
1434     }
1435
1436     /* Add the forces from enforced rotation potentials (if any) */
1437     if (inputrec->bRot)
1438     {
1439         wallcycle_start(wcycle, ewcROTadd);
1440         enerd->term[F_COM_PULL] += add_rot_forces(inputrec->rot, f, cr, step, t);
1441         wallcycle_stop(wcycle, ewcROTadd);
1442     }
1443
1444     if (PAR(cr) && !(cr->duty & DUTY_PME))
1445     {
1446         /* In case of node-splitting, the PP nodes receive the long-range
1447          * forces, virial and energy from the PME nodes here.
1448          */
1449         pme_receive_force_ener(fplog, bSepDVDL, cr, wcycle, enerd, fr);
1450     }
1451
1452     if (bDoForces)
1453     {
1454         post_process_forces(fplog, cr, step, nrnb, wcycle,
1455                             top, box, x, f, vir_force, mdatoms, graph, fr, vsite,
1456                             flags);
1457     }
1458
1459     /* Sum the potential energy terms from group contributions */
1460     sum_epot(&(inputrec->opts), &(enerd->grpp), enerd->term);
1461 }
1462
1463 void do_force_cutsGROUP(FILE *fplog, t_commrec *cr,
1464                         t_inputrec *inputrec,
1465                         gmx_large_int_t step, t_nrnb *nrnb, gmx_wallcycle_t wcycle,
1466                         gmx_localtop_t *top,
1467                         gmx_mtop_t *mtop,
1468                         gmx_groups_t *groups,
1469                         matrix box, rvec x[], history_t *hist,
1470                         rvec f[],
1471                         tensor vir_force,
1472                         t_mdatoms *mdatoms,
1473                         gmx_enerdata_t *enerd, t_fcdata *fcd,
1474                         real *lambda, t_graph *graph,
1475                         t_forcerec *fr, gmx_vsite_t *vsite, rvec mu_tot,
1476                         double t, FILE *field, gmx_edsam_t ed,
1477                         gmx_bool bBornRadii,
1478                         int flags)
1479 {
1480     int        cg0, cg1, i, j;
1481     int        start, homenr;
1482     double     mu[2*DIM];
1483     gmx_bool   bSepDVDL, bStateChanged, bNS, bFillGrid, bCalcCGCM, bBS;
1484     gmx_bool   bDoLongRangeNS, bDoForces, bDoPotential, bSepLRF;
1485     gmx_bool   bDoAdressWF;
1486     matrix     boxs;
1487     rvec       vzero, box_diag;
1488     real       e, v, dvdlambda[efptNR];
1489     t_pbc      pbc;
1490     float      cycles_pme, cycles_force;
1491
1492     start  = mdatoms->start;
1493     homenr = mdatoms->homenr;
1494
1495     bSepDVDL = (fr->bSepDVDL && do_per_step(step, inputrec->nstlog));
1496
1497     clear_mat(vir_force);
1498
1499     if (PARTDECOMP(cr))
1500     {
1501         pd_cg_range(cr, &cg0, &cg1);
1502     }
1503     else
1504     {
1505         cg0 = 0;
1506         if (DOMAINDECOMP(cr))
1507         {
1508             cg1 = cr->dd->ncg_tot;
1509         }
1510         else
1511         {
1512             cg1 = top->cgs.nr;
1513         }
1514         if (fr->n_tpi > 0)
1515         {
1516             cg1--;
1517         }
1518     }
1519
1520     bStateChanged  = (flags & GMX_FORCE_STATECHANGED);
1521     bNS            = (flags & GMX_FORCE_NS) && (fr->bAllvsAll == FALSE);
1522     /* Should we update the long-range neighborlists at this step? */
1523     bDoLongRangeNS = fr->bTwinRange && bNS;
1524     /* Should we perform the long-range nonbonded evaluation inside the neighborsearching? */
1525     bFillGrid      = (bNS && bStateChanged);
1526     bCalcCGCM      = (bFillGrid && !DOMAINDECOMP(cr));
1527     bDoForces      = (flags & GMX_FORCE_FORCES);
1528     bDoPotential   = (flags & GMX_FORCE_ENERGY);
1529     bSepLRF        = ((inputrec->nstcalclr > 1) && bDoForces &&
1530                       (flags & GMX_FORCE_SEPLRF) && (flags & GMX_FORCE_DO_LR));
1531
1532     /* should probably move this to the forcerec since it doesn't change */
1533     bDoAdressWF   = ((fr->adress_type != eAdressOff));
1534
1535     if (bStateChanged)
1536     {
1537         update_forcerec(fplog, fr, box);
1538
1539         if (NEED_MUTOT(*inputrec))
1540         {
1541             /* Calculate total (local) dipole moment in a temporary common array.
1542              * This makes it possible to sum them over nodes faster.
1543              */
1544             calc_mu(start, homenr,
1545                     x, mdatoms->chargeA, mdatoms->chargeB, mdatoms->nChargePerturbed,
1546                     mu, mu+DIM);
1547         }
1548     }
1549
1550     if (fr->ePBC != epbcNONE)
1551     {
1552         /* Compute shift vectors every step,
1553          * because of pressure coupling or box deformation!
1554          */
1555         if ((flags & GMX_FORCE_DYNAMICBOX) && bStateChanged)
1556         {
1557             calc_shifts(box, fr->shift_vec);
1558         }
1559
1560         if (bCalcCGCM)
1561         {
1562             put_charge_groups_in_box(fplog, cg0, cg1, fr->ePBC, box,
1563                                      &(top->cgs), x, fr->cg_cm);
1564             inc_nrnb(nrnb, eNR_CGCM, homenr);
1565             inc_nrnb(nrnb, eNR_RESETX, cg1-cg0);
1566         }
1567         else if (EI_ENERGY_MINIMIZATION(inputrec->eI) && graph)
1568         {
1569             unshift_self(graph, box, x);
1570         }
1571     }
1572     else if (bCalcCGCM)
1573     {
1574         calc_cgcm(fplog, cg0, cg1, &(top->cgs), x, fr->cg_cm);
1575         inc_nrnb(nrnb, eNR_CGCM, homenr);
1576     }
1577
1578     if (bCalcCGCM)
1579     {
1580         if (PAR(cr))
1581         {
1582             move_cgcm(fplog, cr, fr->cg_cm);
1583         }
1584         if (gmx_debug_at)
1585         {
1586             pr_rvecs(debug, 0, "cgcm", fr->cg_cm, top->cgs.nr);
1587         }
1588     }
1589
1590 #ifdef GMX_MPI
1591     if (!(cr->duty & DUTY_PME))
1592     {
1593         /* Send particle coordinates to the pme nodes.
1594          * Since this is only implemented for domain decomposition
1595          * and domain decomposition does not use the graph,
1596          * we do not need to worry about shifting.
1597          */
1598
1599         wallcycle_start(wcycle, ewcPP_PMESENDX);
1600         GMX_MPE_LOG(ev_send_coordinates_start);
1601
1602         bBS = (inputrec->nwall == 2);
1603         if (bBS)
1604         {
1605             copy_mat(box, boxs);
1606             svmul(inputrec->wall_ewald_zfac, boxs[ZZ], boxs[ZZ]);
1607         }
1608
1609         gmx_pme_send_x(cr, bBS ? boxs : box, x,
1610                        mdatoms->nChargePerturbed, lambda[efptCOUL],
1611                        (flags & (GMX_FORCE_VIRIAL | GMX_FORCE_ENERGY)), step);
1612
1613         GMX_MPE_LOG(ev_send_coordinates_finish);
1614         wallcycle_stop(wcycle, ewcPP_PMESENDX);
1615     }
1616 #endif /* GMX_MPI */
1617
1618     /* Communicate coordinates and sum dipole if necessary */
1619     if (PAR(cr))
1620     {
1621         wallcycle_start(wcycle, ewcMOVEX);
1622         if (DOMAINDECOMP(cr))
1623         {
1624             dd_move_x(cr->dd, box, x);
1625         }
1626         else
1627         {
1628             move_x(fplog, cr, GMX_LEFT, GMX_RIGHT, x, nrnb);
1629         }
1630         wallcycle_stop(wcycle, ewcMOVEX);
1631     }
1632
1633     /* update adress weight beforehand */
1634     if (bStateChanged && bDoAdressWF)
1635     {
1636         /* need pbc for adress weight calculation with pbc_dx */
1637         set_pbc(&pbc, inputrec->ePBC, box);
1638         if (fr->adress_site == eAdressSITEcog)
1639         {
1640             update_adress_weights_cog(top->idef.iparams, top->idef.il, x, fr, mdatoms,
1641                                       inputrec->ePBC == epbcNONE ? NULL : &pbc);
1642         }
1643         else if (fr->adress_site == eAdressSITEcom)
1644         {
1645             update_adress_weights_com(fplog, cg0, cg1, &(top->cgs), x, fr, mdatoms,
1646                                       inputrec->ePBC == epbcNONE ? NULL : &pbc);
1647         }
1648         else if (fr->adress_site == eAdressSITEatomatom)
1649         {
1650             update_adress_weights_atom_per_atom(cg0, cg1, &(top->cgs), x, fr, mdatoms,
1651                                                 inputrec->ePBC == epbcNONE ? NULL : &pbc);
1652         }
1653         else
1654         {
1655             update_adress_weights_atom(cg0, cg1, &(top->cgs), x, fr, mdatoms,
1656                                        inputrec->ePBC == epbcNONE ? NULL : &pbc);
1657         }
1658     }
1659
1660     if (NEED_MUTOT(*inputrec))
1661     {
1662
1663         if (bStateChanged)
1664         {
1665             if (PAR(cr))
1666             {
1667                 gmx_sumd(2*DIM, mu, cr);
1668             }
1669             for (i = 0; i < 2; i++)
1670             {
1671                 for (j = 0; j < DIM; j++)
1672                 {
1673                     fr->mu_tot[i][j] = mu[i*DIM + j];
1674                 }
1675             }
1676         }
1677         if (fr->efep == efepNO)
1678         {
1679             copy_rvec(fr->mu_tot[0], mu_tot);
1680         }
1681         else
1682         {
1683             for (j = 0; j < DIM; j++)
1684             {
1685                 mu_tot[j] =
1686                     (1.0 - lambda[efptCOUL])*fr->mu_tot[0][j] + lambda[efptCOUL]*fr->mu_tot[1][j];
1687             }
1688         }
1689     }
1690
1691     /* Reset energies */
1692     reset_enerdata(&(inputrec->opts), fr, bNS, enerd, MASTER(cr));
1693     clear_rvecs(SHIFTS, fr->fshift);
1694
1695     if (bNS)
1696     {
1697         wallcycle_start(wcycle, ewcNS);
1698
1699         if (graph && bStateChanged)
1700         {
1701             /* Calculate intramolecular shift vectors to make molecules whole */
1702             mk_mshift(fplog, graph, fr->ePBC, box, x);
1703         }
1704
1705         /* Do the actual neighbour searching and if twin range electrostatics
1706          * also do the calculation of long range forces and energies.
1707          */
1708         for (i = 0; i < efptNR; i++)
1709         {
1710             dvdlambda[i] = 0;
1711         }
1712         ns(fplog, fr, x, box,
1713            groups, &(inputrec->opts), top, mdatoms,
1714            cr, nrnb, lambda, dvdlambda, &enerd->grpp, bFillGrid,
1715            bDoLongRangeNS);
1716         if (bSepDVDL)
1717         {
1718             fprintf(fplog, sepdvdlformat, "LR non-bonded", 0.0, dvdlambda);
1719         }
1720         enerd->dvdl_lin[efptVDW]  += dvdlambda[efptVDW];
1721         enerd->dvdl_lin[efptCOUL] += dvdlambda[efptCOUL];
1722
1723         wallcycle_stop(wcycle, ewcNS);
1724     }
1725
1726     if (inputrec->implicit_solvent && bNS)
1727     {
1728         make_gb_nblist(cr, inputrec->gb_algorithm, inputrec->rlist,
1729                        x, box, fr, &top->idef, graph, fr->born);
1730     }
1731
1732     if (DOMAINDECOMP(cr))
1733     {
1734         if (!(cr->duty & DUTY_PME))
1735         {
1736             wallcycle_start(wcycle, ewcPPDURINGPME);
1737             dd_force_flop_start(cr->dd, nrnb);
1738         }
1739     }
1740
1741     if (inputrec->bRot)
1742     {
1743         /* Enforced rotation has its own cycle counter that starts after the collective
1744          * coordinates have been communicated. It is added to ddCyclF to allow
1745          * for proper load-balancing */
1746         wallcycle_start(wcycle, ewcROT);
1747         do_rotation(cr, inputrec, box, x, t, step, wcycle, bNS);
1748         wallcycle_stop(wcycle, ewcROT);
1749     }
1750
1751     /* Start the force cycle counter.
1752      * This counter is stopped in do_forcelow_level.
1753      * No parallel communication should occur while this counter is running,
1754      * since that will interfere with the dynamic load balancing.
1755      */
1756     wallcycle_start(wcycle, ewcFORCE);
1757
1758     if (bDoForces)
1759     {
1760         /* Reset forces for which the virial is calculated separately:
1761          * PME/Ewald forces if necessary */
1762         if (fr->bF_NoVirSum)
1763         {
1764             if (flags & GMX_FORCE_VIRIAL)
1765             {
1766                 fr->f_novirsum = fr->f_novirsum_alloc;
1767                 GMX_BARRIER(cr->mpi_comm_mygroup);
1768                 if (fr->bDomDec)
1769                 {
1770                     clear_rvecs(fr->f_novirsum_n, fr->f_novirsum);
1771                 }
1772                 else
1773                 {
1774                     clear_rvecs(homenr, fr->f_novirsum+start);
1775                 }
1776                 GMX_BARRIER(cr->mpi_comm_mygroup);
1777             }
1778             else
1779             {
1780                 /* We are not calculating the pressure so we do not need
1781                  * a separate array for forces that do not contribute
1782                  * to the pressure.
1783                  */
1784                 fr->f_novirsum = f;
1785             }
1786         }
1787
1788         /* Clear the short- and long-range forces */
1789         clear_rvecs(fr->natoms_force_constr, f);
1790         if (bSepLRF && do_per_step(step, inputrec->nstcalclr))
1791         {
1792             clear_rvecs(fr->natoms_force_constr, fr->f_twin);
1793         }
1794
1795         clear_rvec(fr->vir_diag_posres);
1796
1797         GMX_BARRIER(cr->mpi_comm_mygroup);
1798     }
1799     if (inputrec->ePull == epullCONSTRAINT)
1800     {
1801         clear_pull_forces(inputrec->pull);
1802     }
1803
1804     /* update QMMMrec, if necessary */
1805     if (fr->bQMMM)
1806     {
1807         update_QMMMrec(cr, fr, x, mdatoms, box, top);
1808     }
1809
1810     if ((flags & GMX_FORCE_BONDED) && top->idef.il[F_POSRES].nr > 0)
1811     {
1812         posres_wrapper(fplog, flags, bSepDVDL, inputrec, nrnb, top, box, x,
1813                        f, enerd, lambda, fr);
1814     }
1815
1816     /* Compute the bonded and non-bonded energies and optionally forces */
1817     do_force_lowlevel(fplog, step, fr, inputrec, &(top->idef),
1818                       cr, nrnb, wcycle, mdatoms, &(inputrec->opts),
1819                       x, hist, f, bSepLRF ? fr->f_twin : f, enerd, fcd, mtop, top, fr->born,
1820                       &(top->atomtypes), bBornRadii, box,
1821                       inputrec->fepvals, lambda,
1822                       graph, &(top->excls), fr->mu_tot,
1823                       flags,
1824                       &cycles_pme);
1825
1826     if (bSepLRF)
1827     {
1828         if (do_per_step(step, inputrec->nstcalclr))
1829         {
1830             /* Add the long range forces to the short range forces */
1831             for (i = 0; i < fr->natoms_force_constr; i++)
1832             {
1833                 rvec_add(fr->f_twin[i], f[i], f[i]);
1834             }
1835         }
1836     }
1837
1838     cycles_force = wallcycle_stop(wcycle, ewcFORCE);
1839     GMX_BARRIER(cr->mpi_comm_mygroup);
1840
1841     if (ed)
1842     {
1843         do_flood(cr, inputrec, x, f, ed, box, step, bNS);
1844     }
1845
1846     if (DOMAINDECOMP(cr))
1847     {
1848         dd_force_flop_stop(cr->dd, nrnb);
1849         if (wcycle)
1850         {
1851             dd_cycles_add(cr->dd, cycles_force-cycles_pme, ddCyclF);
1852         }
1853     }
1854
1855     if (bDoForces)
1856     {
1857         if (IR_ELEC_FIELD(*inputrec))
1858         {
1859             /* Compute forces due to electric field */
1860             calc_f_el(MASTER(cr) ? field : NULL,
1861                       start, homenr, mdatoms->chargeA, x, fr->f_novirsum,
1862                       inputrec->ex, inputrec->et, t);
1863         }
1864
1865         if (bDoAdressWF && fr->adress_icor == eAdressICThermoForce)
1866         {
1867             /* Compute thermodynamic force in hybrid AdResS region */
1868             adress_thermo_force(start, homenr, &(top->cgs), x, fr->f_novirsum, fr, mdatoms,
1869                                 inputrec->ePBC == epbcNONE ? NULL : &pbc);
1870         }
1871
1872         /* Communicate the forces */
1873         if (PAR(cr))
1874         {
1875             wallcycle_start(wcycle, ewcMOVEF);
1876             if (DOMAINDECOMP(cr))
1877             {
1878                 dd_move_f(cr->dd, f, fr->fshift);
1879                 /* Do we need to communicate the separate force array
1880                  * for terms that do not contribute to the single sum virial?
1881                  * Position restraints and electric fields do not introduce
1882                  * inter-cg forces, only full electrostatics methods do.
1883                  * When we do not calculate the virial, fr->f_novirsum = f,
1884                  * so we have already communicated these forces.
1885                  */
1886                 if (EEL_FULL(fr->eeltype) && cr->dd->n_intercg_excl &&
1887                     (flags & GMX_FORCE_VIRIAL))
1888                 {
1889                     dd_move_f(cr->dd, fr->f_novirsum, NULL);
1890                 }
1891                 if (bSepLRF)
1892                 {
1893                     /* We should not update the shift forces here,
1894                      * since f_twin is already included in f.
1895                      */
1896                     dd_move_f(cr->dd, fr->f_twin, NULL);
1897                 }
1898             }
1899             else
1900             {
1901                 pd_move_f(cr, f, nrnb);
1902                 if (bSepLRF)
1903                 {
1904                     pd_move_f(cr, fr->f_twin, nrnb);
1905                 }
1906             }
1907             wallcycle_stop(wcycle, ewcMOVEF);
1908         }
1909
1910         /* If we have NoVirSum forces, but we do not calculate the virial,
1911          * we sum fr->f_novirum=f later.
1912          */
1913         if (vsite && !(fr->bF_NoVirSum && !(flags & GMX_FORCE_VIRIAL)))
1914         {
1915             wallcycle_start(wcycle, ewcVSITESPREAD);
1916             spread_vsite_f(fplog, vsite, x, f, fr->fshift, FALSE, NULL, nrnb,
1917                            &top->idef, fr->ePBC, fr->bMolPBC, graph, box, cr);
1918             wallcycle_stop(wcycle, ewcVSITESPREAD);
1919
1920             if (bSepLRF)
1921             {
1922                 wallcycle_start(wcycle, ewcVSITESPREAD);
1923                 spread_vsite_f(fplog, vsite, x, fr->f_twin, NULL, FALSE, NULL,
1924                                nrnb,
1925                                &top->idef, fr->ePBC, fr->bMolPBC, graph, box, cr);
1926                 wallcycle_stop(wcycle, ewcVSITESPREAD);
1927             }
1928         }
1929
1930         if (flags & GMX_FORCE_VIRIAL)
1931         {
1932             /* Calculation of the virial must be done after vsites! */
1933             calc_virial(fplog, mdatoms->start, mdatoms->homenr, x, f,
1934                         vir_force, graph, box, nrnb, fr, inputrec->ePBC);
1935         }
1936     }
1937
1938     if (inputrec->ePull == epullUMBRELLA || inputrec->ePull == epullCONST_F)
1939     {
1940         pull_potential_wrapper(fplog, bSepDVDL, cr, inputrec, box, x,
1941                                f, vir_force, mdatoms, enerd, lambda, t);
1942     }
1943
1944     /* Add the forces from enforced rotation potentials (if any) */
1945     if (inputrec->bRot)
1946     {
1947         wallcycle_start(wcycle, ewcROTadd);
1948         enerd->term[F_COM_PULL] += add_rot_forces(inputrec->rot, f, cr, step, t);
1949         wallcycle_stop(wcycle, ewcROTadd);
1950     }
1951
1952     if (PAR(cr) && !(cr->duty & DUTY_PME))
1953     {
1954         /* In case of node-splitting, the PP nodes receive the long-range
1955          * forces, virial and energy from the PME nodes here.
1956          */
1957         pme_receive_force_ener(fplog, bSepDVDL, cr, wcycle, enerd, fr);
1958     }
1959
1960     if (bDoForces)
1961     {
1962         post_process_forces(fplog, cr, step, nrnb, wcycle,
1963                             top, box, x, f, vir_force, mdatoms, graph, fr, vsite,
1964                             flags);
1965     }
1966
1967     /* Sum the potential energy terms from group contributions */
1968     sum_epot(&(inputrec->opts), &(enerd->grpp), enerd->term);
1969 }
1970
1971 void do_force(FILE *fplog, t_commrec *cr,
1972               t_inputrec *inputrec,
1973               gmx_large_int_t step, t_nrnb *nrnb, gmx_wallcycle_t wcycle,
1974               gmx_localtop_t *top,
1975               gmx_mtop_t *mtop,
1976               gmx_groups_t *groups,
1977               matrix box, rvec x[], history_t *hist,
1978               rvec f[],
1979               tensor vir_force,
1980               t_mdatoms *mdatoms,
1981               gmx_enerdata_t *enerd, t_fcdata *fcd,
1982               real *lambda, t_graph *graph,
1983               t_forcerec *fr,
1984               gmx_vsite_t *vsite, rvec mu_tot,
1985               double t, FILE *field, gmx_edsam_t ed,
1986               gmx_bool bBornRadii,
1987               int flags)
1988 {
1989     /* modify force flag if not doing nonbonded */
1990     if (!fr->bNonbonded)
1991     {
1992         flags &= ~GMX_FORCE_NONBONDED;
1993     }
1994
1995     switch (inputrec->cutoff_scheme)
1996     {
1997         case ecutsVERLET:
1998             do_force_cutsVERLET(fplog, cr, inputrec,
1999                                 step, nrnb, wcycle,
2000                                 top, mtop,
2001                                 groups,
2002                                 box, x, hist,
2003                                 f, vir_force,
2004                                 mdatoms,
2005                                 enerd, fcd,
2006                                 lambda, graph,
2007                                 fr, fr->ic,
2008                                 vsite, mu_tot,
2009                                 t, field, ed,
2010                                 bBornRadii,
2011                                 flags);
2012             break;
2013         case ecutsGROUP:
2014             do_force_cutsGROUP(fplog, cr, inputrec,
2015                                step, nrnb, wcycle,
2016                                top, mtop,
2017                                groups,
2018                                box, x, hist,
2019                                f, vir_force,
2020                                mdatoms,
2021                                enerd, fcd,
2022                                lambda, graph,
2023                                fr, vsite, mu_tot,
2024                                t, field, ed,
2025                                bBornRadii,
2026                                flags);
2027             break;
2028         default:
2029             gmx_incons("Invalid cut-off scheme passed!");
2030     }
2031 }
2032
2033
2034 void do_constrain_first(FILE *fplog, gmx_constr_t constr,
2035                         t_inputrec *ir, t_mdatoms *md,
2036                         t_state *state, rvec *f,
2037                         t_graph *graph, t_commrec *cr, t_nrnb *nrnb,
2038                         t_forcerec *fr, gmx_localtop_t *top, tensor shake_vir)
2039 {
2040     int             i, m, start, end;
2041     gmx_large_int_t step;
2042     real            dt = ir->delta_t;
2043     real            dvdl_dum;
2044     rvec           *savex;
2045
2046     snew(savex, state->natoms);
2047
2048     start = md->start;
2049     end   = md->homenr + start;
2050
2051     if (debug)
2052     {
2053         fprintf(debug, "vcm: start=%d, homenr=%d, end=%d\n",
2054                 start, md->homenr, end);
2055     }
2056     /* Do a first constrain to reset particles... */
2057     step = ir->init_step;
2058     if (fplog)
2059     {
2060         char buf[STEPSTRSIZE];
2061         fprintf(fplog, "\nConstraining the starting coordinates (step %s)\n",
2062                 gmx_step_str(step, buf));
2063     }
2064     dvdl_dum = 0;
2065
2066     /* constrain the current position */
2067     constrain(NULL, TRUE, FALSE, constr, &(top->idef),
2068               ir, NULL, cr, step, 0, md,
2069               state->x, state->x, NULL,
2070               fr->bMolPBC, state->box,
2071               state->lambda[efptBONDED], &dvdl_dum,
2072               NULL, NULL, nrnb, econqCoord,
2073               ir->epc == epcMTTK, state->veta, state->veta);
2074     if (EI_VV(ir->eI))
2075     {
2076         /* constrain the inital velocity, and save it */
2077         /* also may be useful if we need the ekin from the halfstep for velocity verlet */
2078         /* might not yet treat veta correctly */
2079         constrain(NULL, TRUE, FALSE, constr, &(top->idef),
2080                   ir, NULL, cr, step, 0, md,
2081                   state->x, state->v, state->v,
2082                   fr->bMolPBC, state->box,
2083                   state->lambda[efptBONDED], &dvdl_dum,
2084                   NULL, NULL, nrnb, econqVeloc,
2085                   ir->epc == epcMTTK, state->veta, state->veta);
2086     }
2087     /* constrain the inital velocities at t-dt/2 */
2088     if (EI_STATE_VELOCITY(ir->eI) && ir->eI != eiVV)
2089     {
2090         for (i = start; (i < end); i++)
2091         {
2092             for (m = 0; (m < DIM); m++)
2093             {
2094                 /* Reverse the velocity */
2095                 state->v[i][m] = -state->v[i][m];
2096                 /* Store the position at t-dt in buf */
2097                 savex[i][m] = state->x[i][m] + dt*state->v[i][m];
2098             }
2099         }
2100         /* Shake the positions at t=-dt with the positions at t=0
2101          * as reference coordinates.
2102          */
2103         if (fplog)
2104         {
2105             char buf[STEPSTRSIZE];
2106             fprintf(fplog, "\nConstraining the coordinates at t0-dt (step %s)\n",
2107                     gmx_step_str(step, buf));
2108         }
2109         dvdl_dum = 0;
2110         constrain(NULL, TRUE, FALSE, constr, &(top->idef),
2111                   ir, NULL, cr, step, -1, md,
2112                   state->x, savex, NULL,
2113                   fr->bMolPBC, state->box,
2114                   state->lambda[efptBONDED], &dvdl_dum,
2115                   state->v, NULL, nrnb, econqCoord,
2116                   ir->epc == epcMTTK, state->veta, state->veta);
2117
2118         for (i = start; i < end; i++)
2119         {
2120             for (m = 0; m < DIM; m++)
2121             {
2122                 /* Re-reverse the velocities */
2123                 state->v[i][m] = -state->v[i][m];
2124             }
2125         }
2126     }
2127     sfree(savex);
2128 }
2129
2130 void calc_enervirdiff(FILE *fplog, int eDispCorr, t_forcerec *fr)
2131 {
2132     double eners[2], virs[2], enersum, virsum, y0, f, g, h;
2133     double r0, r1, r, rc3, rc9, ea, eb, ec, pa, pb, pc, pd;
2134     double invscale, invscale2, invscale3;
2135     int    ri0, ri1, ri, i, offstart, offset;
2136     real   scale, *vdwtab, tabfactor, tmp;
2137
2138     fr->enershiftsix    = 0;
2139     fr->enershifttwelve = 0;
2140     fr->enerdiffsix     = 0;
2141     fr->enerdifftwelve  = 0;
2142     fr->virdiffsix      = 0;
2143     fr->virdifftwelve   = 0;
2144
2145     if (eDispCorr != edispcNO)
2146     {
2147         for (i = 0; i < 2; i++)
2148         {
2149             eners[i] = 0;
2150             virs[i]  = 0;
2151         }
2152         if ((fr->vdwtype == evdwSWITCH) || (fr->vdwtype == evdwSHIFT))
2153         {
2154             if (fr->rvdw_switch == 0)
2155             {
2156                 gmx_fatal(FARGS,
2157                           "With dispersion correction rvdw-switch can not be zero "
2158                           "for vdw-type = %s", evdw_names[fr->vdwtype]);
2159             }
2160
2161             scale  = fr->nblists[0].table_elec_vdw.scale;
2162             vdwtab = fr->nblists[0].table_vdw.data;
2163
2164             /* Round the cut-offs to exact table values for precision */
2165             ri0  = floor(fr->rvdw_switch*scale);
2166             ri1  = ceil(fr->rvdw*scale);
2167             r0   = ri0/scale;
2168             r1   = ri1/scale;
2169             rc3  = r0*r0*r0;
2170             rc9  = rc3*rc3*rc3;
2171
2172             if (fr->vdwtype == evdwSHIFT)
2173             {
2174                 /* Determine the constant energy shift below rvdw_switch.
2175                  * Table has a scale factor since we have scaled it down to compensate
2176                  * for scaling-up c6/c12 with the derivative factors to save flops in analytical kernels.
2177                  */
2178                 fr->enershiftsix    = (real)(-1.0/(rc3*rc3)) - 6.0*vdwtab[8*ri0];
2179                 fr->enershifttwelve = (real)( 1.0/(rc9*rc3)) - 12.0*vdwtab[8*ri0 + 4];
2180             }
2181             /* Add the constant part from 0 to rvdw_switch.
2182              * This integration from 0 to rvdw_switch overcounts the number
2183              * of interactions by 1, as it also counts the self interaction.
2184              * We will correct for this later.
2185              */
2186             eners[0] += 4.0*M_PI*fr->enershiftsix*rc3/3.0;
2187             eners[1] += 4.0*M_PI*fr->enershifttwelve*rc3/3.0;
2188
2189             invscale  = 1.0/(scale);
2190             invscale2 = invscale*invscale;
2191             invscale3 = invscale*invscale2;
2192
2193             /* following summation derived from cubic spline definition,
2194                Numerical Recipies in C, second edition, p. 113-116.  Exact
2195                for the cubic spline.  We first calculate the negative of
2196                the energy from rvdw to rvdw_switch, assuming that g(r)=1,
2197                and then add the more standard, abrupt cutoff correction to
2198                that result, yielding the long-range correction for a
2199                switched function.  We perform both the pressure and energy
2200                loops at the same time for simplicity, as the computational
2201                cost is low. */
2202
2203             for (i = 0; i < 2; i++)
2204             {
2205                 enersum = 0.0; virsum = 0.0;
2206                 if (i == 0)
2207                 {
2208                     offstart = 0;
2209                     /* Since the dispersion table has been scaled down a factor 6.0 and the repulsion
2210                      * a factor 12.0 to compensate for the c6/c12 parameters inside nbfp[] being scaled
2211                      * up (to save flops in kernels), we need to correct for this.
2212                      */
2213                     tabfactor = 6.0;
2214                 }
2215                 else
2216                 {
2217                     offstart  = 4;
2218                     tabfactor = 12.0;
2219                 }
2220                 for (ri = ri0; ri < ri1; ri++)
2221                 {
2222                     r  = ri*invscale;
2223                     ea = invscale3;
2224                     eb = 2.0*invscale2*r;
2225                     ec = invscale*r*r;
2226
2227                     pa = invscale3;
2228                     pb = 3.0*invscale2*r;
2229                     pc = 3.0*invscale*r*r;
2230                     pd = r*r*r;
2231
2232                     /* this "8" is from the packing in the vdwtab array - perhaps should be #define'ed? */
2233                     offset = 8*ri + offstart;
2234                     y0     = vdwtab[offset];
2235                     f      = vdwtab[offset+1];
2236                     g      = vdwtab[offset+2];
2237                     h      = vdwtab[offset+3];
2238
2239                     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);
2240                     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);
2241                 }
2242
2243                 enersum  *= 4.0*M_PI*tabfactor;
2244                 virsum   *= 4.0*M_PI*tabfactor;
2245                 eners[i] -= enersum;
2246                 virs[i]  -= virsum;
2247             }
2248
2249             /* now add the correction for rvdw_switch to infinity */
2250             eners[0] += -4.0*M_PI/(3.0*rc3);
2251             eners[1] +=  4.0*M_PI/(9.0*rc9);
2252             virs[0]  +=  8.0*M_PI/rc3;
2253             virs[1]  += -16.0*M_PI/(3.0*rc9);
2254         }
2255         else if ((fr->vdwtype == evdwCUT) || (fr->vdwtype == evdwUSER))
2256         {
2257             if (fr->vdwtype == evdwUSER && fplog)
2258             {
2259                 fprintf(fplog,
2260                         "WARNING: using dispersion correction with user tables\n");
2261             }
2262             rc3  = fr->rvdw*fr->rvdw*fr->rvdw;
2263             rc9  = rc3*rc3*rc3;
2264             /* Contribution beyond the cut-off */
2265             eners[0] += -4.0*M_PI/(3.0*rc3);
2266             eners[1] +=  4.0*M_PI/(9.0*rc9);
2267             if (fr->vdw_modifier == eintmodPOTSHIFT)
2268             {
2269                 /* Contribution within the cut-off */
2270                 eners[0] += -4.0*M_PI/(3.0*rc3);
2271                 eners[1] +=  4.0*M_PI/(3.0*rc9);
2272             }
2273             /* Contribution beyond the cut-off */
2274             virs[0]  +=  8.0*M_PI/rc3;
2275             virs[1]  += -16.0*M_PI/(3.0*rc9);
2276         }
2277         else
2278         {
2279             gmx_fatal(FARGS,
2280                       "Dispersion correction is not implemented for vdw-type = %s",
2281                       evdw_names[fr->vdwtype]);
2282         }
2283         fr->enerdiffsix    = eners[0];
2284         fr->enerdifftwelve = eners[1];
2285         /* The 0.5 is due to the Gromacs definition of the virial */
2286         fr->virdiffsix     = 0.5*virs[0];
2287         fr->virdifftwelve  = 0.5*virs[1];
2288     }
2289 }
2290
2291 void calc_dispcorr(FILE *fplog, t_inputrec *ir, t_forcerec *fr,
2292                    gmx_large_int_t step, int natoms,
2293                    matrix box, real lambda, tensor pres, tensor virial,
2294                    real *prescorr, real *enercorr, real *dvdlcorr)
2295 {
2296     gmx_bool bCorrAll, bCorrPres;
2297     real     dvdlambda, invvol, dens, ninter, avcsix, avctwelve, enerdiff, svir = 0, spres = 0;
2298     int      m;
2299
2300     *prescorr = 0;
2301     *enercorr = 0;
2302     *dvdlcorr = 0;
2303
2304     clear_mat(virial);
2305     clear_mat(pres);
2306
2307     if (ir->eDispCorr != edispcNO)
2308     {
2309         bCorrAll  = (ir->eDispCorr == edispcAllEner ||
2310                      ir->eDispCorr == edispcAllEnerPres);
2311         bCorrPres = (ir->eDispCorr == edispcEnerPres ||
2312                      ir->eDispCorr == edispcAllEnerPres);
2313
2314         invvol = 1/det(box);
2315         if (fr->n_tpi)
2316         {
2317             /* Only correct for the interactions with the inserted molecule */
2318             dens   = (natoms - fr->n_tpi)*invvol;
2319             ninter = fr->n_tpi;
2320         }
2321         else
2322         {
2323             dens   = natoms*invvol;
2324             ninter = 0.5*natoms;
2325         }
2326
2327         if (ir->efep == efepNO)
2328         {
2329             avcsix    = fr->avcsix[0];
2330             avctwelve = fr->avctwelve[0];
2331         }
2332         else
2333         {
2334             avcsix    = (1 - lambda)*fr->avcsix[0]    + lambda*fr->avcsix[1];
2335             avctwelve = (1 - lambda)*fr->avctwelve[0] + lambda*fr->avctwelve[1];
2336         }
2337
2338         enerdiff   = ninter*(dens*fr->enerdiffsix - fr->enershiftsix);
2339         *enercorr += avcsix*enerdiff;
2340         dvdlambda  = 0.0;
2341         if (ir->efep != efepNO)
2342         {
2343             dvdlambda += (fr->avcsix[1] - fr->avcsix[0])*enerdiff;
2344         }
2345         if (bCorrAll)
2346         {
2347             enerdiff   = ninter*(dens*fr->enerdifftwelve - fr->enershifttwelve);
2348             *enercorr += avctwelve*enerdiff;
2349             if (fr->efep != efepNO)
2350             {
2351                 dvdlambda += (fr->avctwelve[1] - fr->avctwelve[0])*enerdiff;
2352             }
2353         }
2354
2355         if (bCorrPres)
2356         {
2357             svir = ninter*dens*avcsix*fr->virdiffsix/3.0;
2358             if (ir->eDispCorr == edispcAllEnerPres)
2359             {
2360                 svir += ninter*dens*avctwelve*fr->virdifftwelve/3.0;
2361             }
2362             /* The factor 2 is because of the Gromacs virial definition */
2363             spres = -2.0*invvol*svir*PRESFAC;
2364
2365             for (m = 0; m < DIM; m++)
2366             {
2367                 virial[m][m] += svir;
2368                 pres[m][m]   += spres;
2369             }
2370             *prescorr += spres;
2371         }
2372
2373         /* Can't currently control when it prints, for now, just print when degugging */
2374         if (debug)
2375         {
2376             if (bCorrAll)
2377             {
2378                 fprintf(debug, "Long Range LJ corr.: <C6> %10.4e, <C12> %10.4e\n",
2379                         avcsix, avctwelve);
2380             }
2381             if (bCorrPres)
2382             {
2383                 fprintf(debug,
2384                         "Long Range LJ corr.: Epot %10g, Pres: %10g, Vir: %10g\n",
2385                         *enercorr, spres, svir);
2386             }
2387             else
2388             {
2389                 fprintf(debug, "Long Range LJ corr.: Epot %10g\n", *enercorr);
2390             }
2391         }
2392
2393         if (fr->bSepDVDL && do_per_step(step, ir->nstlog))
2394         {
2395             fprintf(fplog, sepdvdlformat, "Dispersion correction",
2396                     *enercorr, dvdlambda);
2397         }
2398         if (fr->efep != efepNO)
2399         {
2400             *dvdlcorr += dvdlambda;
2401         }
2402     }
2403 }
2404
2405 void do_pbc_first(FILE *fplog, matrix box, t_forcerec *fr,
2406                   t_graph *graph, rvec x[])
2407 {
2408     if (fplog)
2409     {
2410         fprintf(fplog, "Removing pbc first time\n");
2411     }
2412     calc_shifts(box, fr->shift_vec);
2413     if (graph)
2414     {
2415         mk_mshift(fplog, graph, fr->ePBC, box, x);
2416         if (gmx_debug_at)
2417         {
2418             p_graph(debug, "do_pbc_first 1", graph);
2419         }
2420         shift_self(graph, box, x);
2421         /* By doing an extra mk_mshift the molecules that are broken
2422          * because they were e.g. imported from another software
2423          * will be made whole again. Such are the healing powers
2424          * of GROMACS.
2425          */
2426         mk_mshift(fplog, graph, fr->ePBC, box, x);
2427         if (gmx_debug_at)
2428         {
2429             p_graph(debug, "do_pbc_first 2", graph);
2430         }
2431     }
2432     if (fplog)
2433     {
2434         fprintf(fplog, "Done rmpbc\n");
2435     }
2436 }
2437
2438 static void low_do_pbc_mtop(FILE *fplog, int ePBC, matrix box,
2439                             gmx_mtop_t *mtop, rvec x[],
2440                             gmx_bool bFirst)
2441 {
2442     t_graph        *graph;
2443     int             mb, as, mol;
2444     gmx_molblock_t *molb;
2445
2446     if (bFirst && fplog)
2447     {
2448         fprintf(fplog, "Removing pbc first time\n");
2449     }
2450
2451     snew(graph, 1);
2452     as = 0;
2453     for (mb = 0; mb < mtop->nmolblock; mb++)
2454     {
2455         molb = &mtop->molblock[mb];
2456         if (molb->natoms_mol == 1 ||
2457             (!bFirst && mtop->moltype[molb->type].cgs.nr == 1))
2458         {
2459             /* Just one atom or charge group in the molecule, no PBC required */
2460             as += molb->nmol*molb->natoms_mol;
2461         }
2462         else
2463         {
2464             /* Pass NULL iso fplog to avoid graph prints for each molecule type */
2465             mk_graph_ilist(NULL, mtop->moltype[molb->type].ilist,
2466                            0, molb->natoms_mol, FALSE, FALSE, graph);
2467
2468             for (mol = 0; mol < molb->nmol; mol++)
2469             {
2470                 mk_mshift(fplog, graph, ePBC, box, x+as);
2471
2472                 shift_self(graph, box, x+as);
2473                 /* The molecule is whole now.
2474                  * We don't need the second mk_mshift call as in do_pbc_first,
2475                  * since we no longer need this graph.
2476                  */
2477
2478                 as += molb->natoms_mol;
2479             }
2480             done_graph(graph);
2481         }
2482     }
2483     sfree(graph);
2484 }
2485
2486 void do_pbc_first_mtop(FILE *fplog, int ePBC, matrix box,
2487                        gmx_mtop_t *mtop, rvec x[])
2488 {
2489     low_do_pbc_mtop(fplog, ePBC, box, mtop, x, TRUE);
2490 }
2491
2492 void do_pbc_mtop(FILE *fplog, int ePBC, matrix box,
2493                  gmx_mtop_t *mtop, rvec x[])
2494 {
2495     low_do_pbc_mtop(fplog, ePBC, box, mtop, x, FALSE);
2496 }
2497
2498 void finish_run(FILE *fplog, t_commrec *cr, const char *confout,
2499                 t_inputrec *inputrec,
2500                 t_nrnb nrnb[], gmx_wallcycle_t wcycle,
2501                 gmx_runtime_t *runtime,
2502                 wallclock_gpu_t *gputimes,
2503                 int omp_nth_pp,
2504                 gmx_bool bWriteStat)
2505 {
2506     int     i, j;
2507     t_nrnb *nrnb_tot = NULL;
2508     real    delta_t;
2509     double  nbfs, mflop;
2510
2511     wallcycle_sum(cr, wcycle);
2512
2513     if (cr->nnodes > 1)
2514     {
2515         snew(nrnb_tot, 1);
2516 #ifdef GMX_MPI
2517         MPI_Allreduce(nrnb->n, nrnb_tot->n, eNRNB, MPI_DOUBLE, MPI_SUM,
2518                       cr->mpi_comm_mysim);
2519 #endif
2520     }
2521     else
2522     {
2523         nrnb_tot = nrnb;
2524     }
2525
2526 #if defined(GMX_MPI) && !defined(GMX_THREAD_MPI)
2527     if (cr->nnodes > 1)
2528     {
2529         /* reduce nodetime over all MPI processes in the current simulation */
2530         double sum;
2531         MPI_Allreduce(&runtime->proctime, &sum, 1, MPI_DOUBLE, MPI_SUM,
2532                       cr->mpi_comm_mysim);
2533         runtime->proctime = sum;
2534     }
2535 #endif
2536
2537     if (SIMMASTER(cr))
2538     {
2539         print_flop(fplog, nrnb_tot, &nbfs, &mflop);
2540     }
2541     if (cr->nnodes > 1)
2542     {
2543         sfree(nrnb_tot);
2544     }
2545
2546     if ((cr->duty & DUTY_PP) && DOMAINDECOMP(cr))
2547     {
2548         print_dd_statistics(cr, inputrec, fplog);
2549     }
2550
2551 #ifdef GMX_MPI
2552     if (PARTDECOMP(cr))
2553     {
2554         if (MASTER(cr))
2555         {
2556             t_nrnb     *nrnb_all;
2557             int         s;
2558             MPI_Status  stat;
2559
2560             snew(nrnb_all, cr->nnodes);
2561             nrnb_all[0] = *nrnb;
2562             for (s = 1; s < cr->nnodes; s++)
2563             {
2564                 MPI_Recv(nrnb_all[s].n, eNRNB, MPI_DOUBLE, s, 0,
2565                          cr->mpi_comm_mysim, &stat);
2566             }
2567             pr_load(fplog, cr, nrnb_all);
2568             sfree(nrnb_all);
2569         }
2570         else
2571         {
2572             MPI_Send(nrnb->n, eNRNB, MPI_DOUBLE, MASTERRANK(cr), 0,
2573                      cr->mpi_comm_mysim);
2574         }
2575     }
2576 #endif
2577
2578     if (SIMMASTER(cr))
2579     {
2580         wallcycle_print(fplog, cr->nnodes, cr->npmenodes, runtime->realtime,
2581                         wcycle, gputimes);
2582
2583         if (EI_DYNAMICS(inputrec->eI))
2584         {
2585             delta_t = inputrec->delta_t;
2586         }
2587         else
2588         {
2589             delta_t = 0;
2590         }
2591
2592         if (fplog)
2593         {
2594             print_perf(fplog, runtime->proctime, runtime->realtime,
2595                        cr->nnodes-cr->npmenodes,
2596                        runtime->nsteps_done, delta_t, nbfs, mflop,
2597                        omp_nth_pp);
2598         }
2599         if (bWriteStat)
2600         {
2601             print_perf(stderr, runtime->proctime, runtime->realtime,
2602                        cr->nnodes-cr->npmenodes,
2603                        runtime->nsteps_done, delta_t, nbfs, mflop,
2604                        omp_nth_pp);
2605         }
2606     }
2607 }
2608
2609 extern void initialize_lambdas(FILE *fplog, t_inputrec *ir, int *fep_state, real *lambda, double *lam0)
2610 {
2611     /* this function works, but could probably use a logic rewrite to keep all the different
2612        types of efep straight. */
2613
2614     int       i;
2615     t_lambda *fep = ir->fepvals;
2616
2617     if ((ir->efep == efepNO) && (ir->bSimTemp == FALSE))
2618     {
2619         for (i = 0; i < efptNR; i++)
2620         {
2621             lambda[i] = 0.0;
2622             if (lam0)
2623             {
2624                 lam0[i] = 0.0;
2625             }
2626         }
2627         return;
2628     }
2629     else
2630     {
2631         *fep_state = fep->init_fep_state; /* this might overwrite the checkpoint
2632                                              if checkpoint is set -- a kludge is in for now
2633                                              to prevent this.*/
2634         for (i = 0; i < efptNR; i++)
2635         {
2636             /* overwrite lambda state with init_lambda for now for backwards compatibility */
2637             if (fep->init_lambda >= 0) /* if it's -1, it was never initializd */
2638             {
2639                 lambda[i] = fep->init_lambda;
2640                 if (lam0)
2641                 {
2642                     lam0[i] = lambda[i];
2643                 }
2644             }
2645             else
2646             {
2647                 lambda[i] = fep->all_lambda[i][*fep_state];
2648                 if (lam0)
2649                 {
2650                     lam0[i] = lambda[i];
2651                 }
2652             }
2653         }
2654         if (ir->bSimTemp)
2655         {
2656             /* need to rescale control temperatures to match current state */
2657             for (i = 0; i < ir->opts.ngtc; i++)
2658             {
2659                 if (ir->opts.ref_t[i] > 0)
2660                 {
2661                     ir->opts.ref_t[i] = ir->simtempvals->temperatures[*fep_state];
2662                 }
2663             }
2664         }
2665     }
2666
2667     /* Send to the log the information on the current lambdas */
2668     if (fplog != NULL)
2669     {
2670         fprintf(fplog, "Initial vector of lambda components:[ ");
2671         for (i = 0; i < efptNR; i++)
2672         {
2673             fprintf(fplog, "%10.4f ", lambda[i]);
2674         }
2675         fprintf(fplog, "]\n");
2676     }
2677     return;
2678 }
2679
2680
2681 void init_md(FILE *fplog,
2682              t_commrec *cr, t_inputrec *ir, const output_env_t oenv,
2683              double *t, double *t0,
2684              real *lambda, int *fep_state, double *lam0,
2685              t_nrnb *nrnb, gmx_mtop_t *mtop,
2686              gmx_update_t *upd,
2687              int nfile, const t_filenm fnm[],
2688              gmx_mdoutf_t **outf, t_mdebin **mdebin,
2689              tensor force_vir, tensor shake_vir, rvec mu_tot,
2690              gmx_bool *bSimAnn, t_vcm **vcm, t_state *state, unsigned long Flags)
2691 {
2692     int  i, j, n;
2693     real tmpt, mod;
2694
2695     /* Initial values */
2696     *t = *t0       = ir->init_t;
2697
2698     *bSimAnn = FALSE;
2699     for (i = 0; i < ir->opts.ngtc; i++)
2700     {
2701         /* set bSimAnn if any group is being annealed */
2702         if (ir->opts.annealing[i] != eannNO)
2703         {
2704             *bSimAnn = TRUE;
2705         }
2706     }
2707     if (*bSimAnn)
2708     {
2709         update_annealing_target_temp(&(ir->opts), ir->init_t);
2710     }
2711
2712     /* Initialize lambda variables */
2713     initialize_lambdas(fplog, ir, fep_state, lambda, lam0);
2714
2715     if (upd)
2716     {
2717         *upd = init_update(fplog, ir);
2718     }
2719
2720
2721     if (vcm != NULL)
2722     {
2723         *vcm = init_vcm(fplog, &mtop->groups, ir);
2724     }
2725
2726     if (EI_DYNAMICS(ir->eI) && !(Flags & MD_APPENDFILES))
2727     {
2728         if (ir->etc == etcBERENDSEN)
2729         {
2730             please_cite(fplog, "Berendsen84a");
2731         }
2732         if (ir->etc == etcVRESCALE)
2733         {
2734             please_cite(fplog, "Bussi2007a");
2735         }
2736     }
2737
2738     init_nrnb(nrnb);
2739
2740     if (nfile != -1)
2741     {
2742         *outf = init_mdoutf(nfile, fnm, Flags, cr, ir, oenv);
2743
2744         *mdebin = init_mdebin((Flags & MD_APPENDFILES) ? NULL : (*outf)->fp_ene,
2745                               mtop, ir, (*outf)->fp_dhdl);
2746     }
2747
2748     if (ir->bAdress)
2749     {
2750         please_cite(fplog, "Fritsch12");
2751         please_cite(fplog, "Junghans10");
2752     }
2753     /* Initiate variables */
2754     clear_mat(force_vir);
2755     clear_mat(shake_vir);
2756     clear_rvec(mu_tot);
2757
2758     debug_gmx();
2759 }