src/gromacs/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  * Copyright (c) 2013,2014, by the GROMACS development team, led by
   7  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   8  * and including many others, as listed in the AUTHORS file in the
   9  * top-level source directory and at http://www.gromacs.org.
  10  *
  11  * GROMACS is free software; you can redistribute it and/or
  12  * modify it under the terms of the GNU Lesser General Public License
  13  * as published by the Free Software Foundation; either version 2.1
  14  * of the License, or (at your option) any later version.
  15  *
  16  * GROMACS is distributed in the hope that it will be useful,
  17  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  18  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  19  * Lesser General Public License for more details.
  20  *
  21  * You should have received a copy of the GNU Lesser General Public
  22  * License along with GROMACS; if not, see
  23  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  24  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  25  *
  26  * If you want to redistribute modifications to GROMACS, please
  27  * consider that scientific software is very special. Version
  28  * control is crucial - bugs must be traceable. We will be happy to
  29  * consider code for inclusion in the official distribution, but
  30  * derived work must not be called official GROMACS. Details are found
  31  * in the README & COPYING files - if they are missing, get the
  32  * official version at http://www.gromacs.org.
  33  *
  34  * To help us fund GROMACS development, we humbly ask that you cite
  35  * the research papers on the package. Check out http://www.gromacs.org.
  36  */
  37 #ifdef HAVE_CONFIG_H
  38 #include <config.h>
  39 #endif
  40
  41 #include <stdio.h>
  42 #ifdef HAVE_SYS_TIME_H
  43 #include <sys/time.h>
  44 #endif
  45 #include <math.h>
  46 #include "typedefs.h"
  47 #include "string2.h"
  48 #include "smalloc.h"
  49 #include "names.h"
  50 #include "txtdump.h"
  51 #include "pbc.h"
  52 #include "chargegroup.h"
  53 #include "vec.h"
  54 #include "nrnb.h"
  55 #include "mshift.h"
  56 #include "mdrun.h"
  57 #include "sim_util.h"
  58 #include "update.h"
  59 #include "physics.h"
  60 #include "main.h"
  61 #include "mdatoms.h"
  62 #include "force.h"
  63 #include "bondf.h"
  64 #include "pme.h"
  65 #include "disre.h"
  66 #include "orires.h"
  67 #include "network.h"
  68 #include "calcmu.h"
  69 #include "constr.h"
  70 #include "xvgr.h"
  71 #include "copyrite.h"
  72 #include "gromacs/random/random.h"
  73 #include "domdec.h"
  74 #include "genborn.h"
  75 #include "nbnxn_atomdata.h"
  76 #include "nbnxn_search.h"
  77 #include "nbnxn_kernels/nbnxn_kernel_ref.h"
  78 #include "nbnxn_kernels/simd_4xn/nbnxn_kernel_simd_4xn.h"
  79 #include "nbnxn_kernels/simd_2xnn/nbnxn_kernel_simd_2xnn.h"
  80 #include "nbnxn_kernels/nbnxn_kernel_gpu_ref.h"
  81
  82 #include "gromacs/timing/wallcycle.h"
  83 #include "gromacs/timing/walltime_accounting.h"
  84 #include "gromacs/utility/gmxmpi.h"
  85 #include "gromacs/essentialdynamics/edsam.h"
  86 #include "gromacs/pulling/pull.h"
  87 #include "gromacs/pulling/pull_rotation.h"
  88
  89 #include "adress.h"
  90 #include "qmmm.h"
  91
  92 #include "nbnxn_cuda_data_mgmt.h"
  93 #include "nbnxn_cuda/nbnxn_cuda.h"
  94
  95 void print_time(FILE                     *out,
  96                 gmx_walltime_accounting_t walltime_accounting,
  97                 gmx_int64_t               step,
  98                 t_inputrec               *ir,
  99                 t_commrec gmx_unused     *cr)
 100 {
 101     time_t finish;
 102     char   timebuf[STRLEN];
 103     double dt, elapsed_seconds, time_per_step;
 104     char   buf[48];
 105
 106 #ifndef GMX_THREAD_MPI
 107     if (!PAR(cr))
 108 #endif
 109     {
 110         fprintf(out, "\r");
 111     }
 112     fprintf(out, "step %s", gmx_step_str(step, buf));
 113     if ((step >= ir->nstlist))
 114     {
 115         double seconds_since_epoch = gmx_gettime();
 116         elapsed_seconds = seconds_since_epoch - walltime_accounting_get_start_time_stamp(walltime_accounting);
 117         time_per_step   = elapsed_seconds/(step - ir->init_step + 1);
 118         dt              = (ir->nsteps + ir->init_step - step) * time_per_step;
 119
 120         if (ir->nsteps >= 0)
 121         {
 122             if (dt >= 300)
 123             {
 124                 finish = (time_t) (seconds_since_epoch + dt);
 125                 gmx_ctime_r(&finish, timebuf, STRLEN);
 126                 sprintf(buf, "%s", timebuf);
 127                 buf[strlen(buf)-1] = '\0';
 128                 fprintf(out, ", will finish %s", buf);
 129             }
 130             else
 131             {
 132                 fprintf(out, ", remaining wall clock time: %5d s          ", (int)dt);
 133             }
 134         }
 135         else
 136         {
 137             fprintf(out, " performance: %.1f ns/day    ",
 138                     ir->delta_t/1000*24*60*60/time_per_step);
 139         }
 140     }
 141 #ifndef GMX_THREAD_MPI
 142     if (PAR(cr))
 143     {
 144         fprintf(out, "\n");
 145     }
 146 #endif
 147
 148     fflush(out);
 149 }
 150
 151 void print_date_and_time(FILE *fplog, int nodeid, const char *title,
 152                          const gmx_walltime_accounting_t walltime_accounting)
 153 {
 154     int    i;
 155     char   timebuf[STRLEN];
 156     char   time_string[STRLEN];
 157     time_t tmptime;
 158
 159     if (fplog)
 160     {
 161         if (walltime_accounting != NULL)
 162         {
 163             tmptime = (time_t) walltime_accounting_get_start_time_stamp(walltime_accounting);
 164             gmx_ctime_r(&tmptime, timebuf, STRLEN);
 165         }
 166         else
 167         {
 168             tmptime = (time_t) gmx_gettime();
 169             gmx_ctime_r(&tmptime, timebuf, STRLEN);
 170         }
 171         for (i = 0; timebuf[i] >= ' '; i++)
 172         {
 173             time_string[i] = timebuf[i];
 174         }
 175         time_string[i] = '\0';
 176
 177         fprintf(fplog, "%s on node %d %s\n", title, nodeid, time_string);
 178     }
 179 }
 180
 181 void print_start(FILE *fplog, t_commrec *cr,
 182                  gmx_walltime_accounting_t walltime_accounting,
 183                  const char *name)
 184 {
 185     char buf[STRLEN];
 186
 187     sprintf(buf, "Started %s", name);
 188     print_date_and_time(fplog, cr->nodeid, buf, walltime_accounting);
 189 }
 190
 191 static void sum_forces(int start, int end, rvec f[], rvec flr[])
 192 {
 193     int i;
 194
 195     if (gmx_debug_at)
 196     {
 197         pr_rvecs(debug, 0, "fsr", f+start, end-start);
 198         pr_rvecs(debug, 0, "flr", flr+start, end-start);
 199     }
 200     for (i = start; (i < end); i++)
 201     {
 202         rvec_inc(f[i], flr[i]);
 203     }
 204 }
 205
 206 /*
 207  * calc_f_el calculates forces due to an electric field.
 208  *
 209  * force is kJ mol^-1 nm^-1 = e * kJ mol^-1 nm^-1 / e
 210  *
 211  * Et[] contains the parameters for the time dependent
 212  * part of the field (not yet used).
 213  * Ex[] contains the parameters for
 214  * the spatial dependent part of the field. You can have cool periodic
 215  * fields in principle, but only a constant field is supported
 216  * now.
 217  * The function should return the energy due to the electric field
 218  * (if any) but for now returns 0.
 219  *
 220  * WARNING:
 221  * There can be problems with the virial.
 222  * Since the field is not self-consistent this is unavoidable.
 223  * For neutral molecules the virial is correct within this approximation.
 224  * For neutral systems with many charged molecules the error is small.
 225  * But for systems with a net charge or a few charged molecules
 226  * the error can be significant when the field is high.
 227  * Solution: implement a self-consitent electric field into PME.
 228  */
 229 static void calc_f_el(FILE *fp, int  start, int homenr,
 230                       real charge[], rvec f[],
 231                       t_cosines Ex[], t_cosines Et[], double t)
 232 {
 233     rvec Ext;
 234     real t0;
 235     int  i, m;
 236
 237     for (m = 0; (m < DIM); m++)
 238     {
 239         if (Et[m].n > 0)
 240         {
 241             if (Et[m].n == 3)
 242             {
 243                 t0     = Et[m].a[1];
 244                 Ext[m] = cos(Et[m].a[0]*(t-t0))*exp(-sqr(t-t0)/(2.0*sqr(Et[m].a[2])));
 245             }
 246             else
 247             {
 248                 Ext[m] = cos(Et[m].a[0]*t);
 249             }
 250         }
 251         else
 252         {
 253             Ext[m] = 1.0;
 254         }
 255         if (Ex[m].n > 0)
 256         {
 257             /* Convert the field strength from V/nm to MD-units */
 258             Ext[m] *= Ex[m].a[0]*FIELDFAC;
 259             for (i = start; (i < start+homenr); i++)
 260             {
 261                 f[i][m] += charge[i]*Ext[m];
 262             }
 263         }
 264         else
 265         {
 266             Ext[m] = 0;
 267         }
 268     }
 269     if (fp != NULL)
 270     {
 271         fprintf(fp, "%10g  %10g  %10g  %10g #FIELD\n", t,
 272                 Ext[XX]/FIELDFAC, Ext[YY]/FIELDFAC, Ext[ZZ]/FIELDFAC);
 273     }
 274 }
 275
 276 static void calc_virial(int start, int homenr, rvec x[], rvec f[],
 277                         tensor vir_part, t_graph *graph, matrix box,
 278                         t_nrnb *nrnb, const t_forcerec *fr, int ePBC)
 279 {
 280     int    i, j;
 281     tensor virtest;
 282
 283     /* The short-range virial from surrounding boxes */
 284     clear_mat(vir_part);
 285     calc_vir(SHIFTS, fr->shift_vec, fr->fshift, vir_part, ePBC == epbcSCREW, box);
 286     inc_nrnb(nrnb, eNR_VIRIAL, SHIFTS);
 287
 288     /* Calculate partial virial, for local atoms only, based on short range.
 289      * Total virial is computed in global_stat, called from do_md
 290      */
 291     f_calc_vir(start, start+homenr, x, f, vir_part, graph, box);
 292     inc_nrnb(nrnb, eNR_VIRIAL, homenr);
 293
 294     /* Add position restraint contribution */
 295     for (i = 0; i < DIM; i++)
 296     {
 297         vir_part[i][i] += fr->vir_diag_posres[i];
 298     }
 299
 300     /* Add wall contribution */
 301     for (i = 0; i < DIM; i++)
 302     {
 303         vir_part[i][ZZ] += fr->vir_wall_z[i];
 304     }
 305
 306     if (debug)
 307     {
 308         pr_rvecs(debug, 0, "vir_part", vir_part, DIM);
 309     }
 310 }
 311
 312 static void posres_wrapper(FILE *fplog,
 313                            int flags,
 314                            gmx_bool bSepDVDL,
 315                            t_inputrec *ir,
 316                            t_nrnb *nrnb,
 317                            gmx_localtop_t *top,
 318                            matrix box, rvec x[],
 319                            gmx_enerdata_t *enerd,
 320                            real *lambda,
 321                            t_forcerec *fr)
 322 {
 323     t_pbc pbc;
 324     real  v, dvdl;
 325     int   i;
 326
 327     /* Position restraints always require full pbc */
 328     set_pbc(&pbc, ir->ePBC, box);
 329     dvdl = 0;
 330     v    = posres(top->idef.il[F_POSRES].nr, top->idef.il[F_POSRES].iatoms,
 331                   top->idef.iparams_posres,
 332                   (const rvec*)x, fr->f_novirsum, fr->vir_diag_posres,
 333                   ir->ePBC == epbcNONE ? NULL : &pbc,
 334                   lambda[efptRESTRAINT], &dvdl,
 335                   fr->rc_scaling, fr->ePBC, fr->posres_com, fr->posres_comB);
 336     if (bSepDVDL)
 337     {
 338         gmx_print_sepdvdl(fplog, interaction_function[F_POSRES].longname, v, dvdl);
 339     }
 340     enerd->term[F_POSRES] += v;
 341     /* If just the force constant changes, the FEP term is linear,
 342      * but if k changes, it is not.
 343      */
 344     enerd->dvdl_nonlin[efptRESTRAINT] += dvdl;
 345     inc_nrnb(nrnb, eNR_POSRES, top->idef.il[F_POSRES].nr/2);
 346
 347     if ((ir->fepvals->n_lambda > 0) && (flags & GMX_FORCE_DHDL))
 348     {
 349         for (i = 0; i < enerd->n_lambda; i++)
 350         {
 351             real dvdl_dum, lambda_dum;
 352
 353             lambda_dum = (i == 0 ? lambda[efptRESTRAINT] : ir->fepvals->all_lambda[efptRESTRAINT][i-1]);
 354             v          = posres(top->idef.il[F_POSRES].nr, top->idef.il[F_POSRES].iatoms,
 355                                 top->idef.iparams_posres,
 356                                 (const rvec*)x, NULL, NULL,
 357                                 ir->ePBC == epbcNONE ? NULL : &pbc, lambda_dum, &dvdl,
 358                                 fr->rc_scaling, fr->ePBC, fr->posres_com, fr->posres_comB);
 359             enerd->enerpart_lambda[i] += v;
 360         }
 361     }
 362 }
 363
 364 static void fbposres_wrapper(t_inputrec *ir,
 365                              t_nrnb *nrnb,
 366                              gmx_localtop_t *top,
 367                              matrix box, rvec x[],
 368                              gmx_enerdata_t *enerd,
 369                              t_forcerec *fr)
 370 {
 371     t_pbc pbc;
 372     real  v;
 373
 374     /* Flat-bottomed position restraints always require full pbc */
 375     set_pbc(&pbc, ir->ePBC, box);
 376     v = fbposres(top->idef.il[F_FBPOSRES].nr, top->idef.il[F_FBPOSRES].iatoms,
 377                  top->idef.iparams_fbposres,
 378                  (const rvec*)x, fr->f_novirsum, fr->vir_diag_posres,
 379                  ir->ePBC == epbcNONE ? NULL : &pbc,
 380                  fr->rc_scaling, fr->ePBC, fr->posres_com);
 381     enerd->term[F_FBPOSRES] += v;
 382     inc_nrnb(nrnb, eNR_FBPOSRES, top->idef.il[F_FBPOSRES].nr/2);
 383 }
 384
 385 static void pull_potential_wrapper(FILE *fplog,
 386                                    gmx_bool bSepDVDL,
 387                                    t_commrec *cr,
 388                                    t_inputrec *ir,
 389                                    matrix box, rvec x[],
 390                                    rvec f[],
 391                                    tensor vir_force,
 392                                    t_mdatoms *mdatoms,
 393                                    gmx_enerdata_t *enerd,
 394                                    real *lambda,
 395                                    double t)
 396 {
 397     t_pbc  pbc;
 398     real   dvdl;
 399
 400     /* Calculate the center of mass forces, this requires communication,
 401      * which is why pull_potential is called close to other communication.
 402      * The virial contribution is calculated directly,
 403      * which is why we call pull_potential after calc_virial.
 404      */
 405     set_pbc(&pbc, ir->ePBC, box);
 406     dvdl                     = 0;
 407     enerd->term[F_COM_PULL] +=
 408         pull_potential(ir->ePull, ir->pull, mdatoms, &pbc,
 409                        cr, t, lambda[efptRESTRAINT], x, f, vir_force, &dvdl);
 410     if (bSepDVDL)
 411     {
 412         gmx_print_sepdvdl(fplog, "Com pull", enerd->term[F_COM_PULL], dvdl);
 413     }
 414     enerd->dvdl_lin[efptRESTRAINT] += dvdl;
 415 }
 416
 417 static void pme_receive_force_ener(FILE           *fplog,
 418                                    gmx_bool        bSepDVDL,
 419                                    t_commrec      *cr,
 420                                    gmx_wallcycle_t wcycle,
 421                                    gmx_enerdata_t *enerd,
 422                                    t_forcerec     *fr)
 423 {
 424     real   e_q, e_lj, v, dvdl_q, dvdl_lj;
 425     float  cycles_ppdpme, cycles_seppme;
 426
 427     cycles_ppdpme = wallcycle_stop(wcycle, ewcPPDURINGPME);
 428     dd_cycles_add(cr->dd, cycles_ppdpme, ddCyclPPduringPME);
 429
 430     /* In case of node-splitting, the PP nodes receive the long-range
 431      * forces, virial and energy from the PME nodes here.
 432      */
 433     wallcycle_start(wcycle, ewcPP_PMEWAITRECVF);
 434     dvdl_q  = 0;
 435     dvdl_lj = 0;
 436     gmx_pme_receive_f(cr, fr->f_novirsum, fr->vir_el_recip, &e_q,
 437                       fr->vir_lj_recip, &e_lj, &dvdl_q, &dvdl_lj,
 438                       &cycles_seppme);
 439     if (bSepDVDL)
 440     {
 441         gmx_print_sepdvdl(fplog, "Electrostatic PME mesh", e_q, dvdl_q);
 442         gmx_print_sepdvdl(fplog, "Lennard-Jones PME mesh", e_lj, dvdl_lj);
 443     }
 444     enerd->term[F_COUL_RECIP] += e_q;
 445     enerd->term[F_LJ_RECIP]   += e_lj;
 446     enerd->dvdl_lin[efptCOUL] += dvdl_q;
 447     enerd->dvdl_lin[efptVDW]  += dvdl_lj;
 448
 449     if (wcycle)
 450     {
 451         dd_cycles_add(cr->dd, cycles_seppme, ddCyclPME);
 452     }
 453     wallcycle_stop(wcycle, ewcPP_PMEWAITRECVF);
 454 }
 455
 456 static void print_large_forces(FILE *fp, t_mdatoms *md, t_commrec *cr,
 457                                gmx_int64_t step, real pforce, rvec *x, rvec *f)
 458 {
 459     int  i;
 460     real pf2, fn2;
 461     char buf[STEPSTRSIZE];
 462
 463     pf2 = sqr(pforce);
 464     for (i = 0; i < md->homenr; i++)
 465     {
 466         fn2 = norm2(f[i]);
 467         /* We also catch NAN, if the compiler does not optimize this away. */
 468         if (fn2 >= pf2 || fn2 != fn2)
 469         {
 470             fprintf(fp, "step %s  atom %6d  x %8.3f %8.3f %8.3f  force %12.5e\n",
 471                     gmx_step_str(step, buf),
 472                     ddglatnr(cr->dd, i), x[i][XX], x[i][YY], x[i][ZZ], sqrt(fn2));
 473         }
 474     }
 475 }
 476
 477 static void post_process_forces(t_commrec *cr,
 478                                 gmx_int64_t step,
 479                                 t_nrnb *nrnb, gmx_wallcycle_t wcycle,
 480                                 gmx_localtop_t *top,
 481                                 matrix box, rvec x[],
 482                                 rvec f[],
 483                                 tensor vir_force,
 484                                 t_mdatoms *mdatoms,
 485                                 t_graph *graph,
 486                                 t_forcerec *fr, gmx_vsite_t *vsite,
 487                                 int flags)
 488 {
 489     if (fr->bF_NoVirSum)
 490     {
 491         if (vsite)
 492         {
 493             /* Spread the mesh force on virtual sites to the other particles...
 494              * This is parallellized. MPI communication is performed
 495              * if the constructing atoms aren't local.
 496              */
 497             wallcycle_start(wcycle, ewcVSITESPREAD);
 498             spread_vsite_f(vsite, x, fr->f_novirsum, NULL,
 499                            (flags & GMX_FORCE_VIRIAL), fr->vir_el_recip,
 500                            nrnb,
 501                            &top->idef, fr->ePBC, fr->bMolPBC, graph, box, cr);
 502             wallcycle_stop(wcycle, ewcVSITESPREAD);
 503         }
 504         if (flags & GMX_FORCE_VIRIAL)
 505         {
 506             /* Now add the forces, this is local */
 507             if (fr->bDomDec)
 508             {
 509                 sum_forces(0, fr->f_novirsum_n, f, fr->f_novirsum);
 510             }
 511             else
 512             {
 513                 sum_forces(0, mdatoms->homenr,
 514                            f, fr->f_novirsum);
 515             }
 516             if (EEL_FULL(fr->eeltype))
 517             {
 518                 /* Add the mesh contribution to the virial */
 519                 m_add(vir_force, fr->vir_el_recip, vir_force);
 520             }
 521             if (EVDW_PME(fr->vdwtype))
 522             {
 523                 /* Add the mesh contribution to the virial */
 524                 m_add(vir_force, fr->vir_lj_recip, vir_force);
 525             }
 526             if (debug)
 527             {
 528                 pr_rvecs(debug, 0, "vir_force", vir_force, DIM);
 529             }
 530         }
 531     }
 532
 533     if (fr->print_force >= 0)
 534     {
 535         print_large_forces(stderr, mdatoms, cr, step, fr->print_force, x, f);
 536     }
 537 }
 538
 539 static void do_nb_verlet(t_forcerec *fr,
 540                          interaction_const_t *ic,
 541                          gmx_enerdata_t *enerd,
 542                          int flags, int ilocality,
 543                          int clearF,
 544                          t_nrnb *nrnb,
 545                          gmx_wallcycle_t wcycle)
 546 {
 547     int                        nnbl, kernel_type, enr_nbnxn_kernel_ljc, enr_nbnxn_kernel_lj;
 548     char                      *env;
 549     nonbonded_verlet_group_t  *nbvg;
 550     gmx_bool                   bCUDA;
 551
 552     if (!(flags & GMX_FORCE_NONBONDED))
 553     {
 554         /* skip non-bonded calculation */
 555         return;
 556     }
 557
 558     nbvg = &fr->nbv->grp[ilocality];
 559
 560     /* CUDA kernel launch overhead is already timed separately */
 561     if (fr->cutoff_scheme != ecutsVERLET)
 562     {
 563         gmx_incons("Invalid cut-off scheme passed!");
 564     }
 565
 566     bCUDA = (nbvg->kernel_type == nbnxnk8x8x8_CUDA);
 567
 568     if (!bCUDA)
 569     {
 570         wallcycle_sub_start(wcycle, ewcsNONBONDED);
 571     }
 572     switch (nbvg->kernel_type)
 573     {
 574         case nbnxnk4x4_PlainC:
 575             nbnxn_kernel_ref(&nbvg->nbl_lists,
 576                              nbvg->nbat, ic,
 577                              fr->shift_vec,
 578                              flags,
 579                              clearF,
 580                              fr->fshift[0],
 581                              enerd->grpp.ener[egCOULSR],
 582                              fr->bBHAM ?
 583                              enerd->grpp.ener[egBHAMSR] :
 584                              enerd->grpp.ener[egLJSR]);
 585             break;
 586
 587         case nbnxnk4xN_SIMD_4xN:
 588             nbnxn_kernel_simd_4xn(&nbvg->nbl_lists,
 589                                   nbvg->nbat, ic,
 590                                   nbvg->ewald_excl,
 591                                   fr->shift_vec,
 592                                   flags,
 593                                   clearF,
 594                                   fr->fshift[0],
 595                                   enerd->grpp.ener[egCOULSR],
 596                                   fr->bBHAM ?
 597                                   enerd->grpp.ener[egBHAMSR] :
 598                                   enerd->grpp.ener[egLJSR]);
 599             break;
 600         case nbnxnk4xN_SIMD_2xNN:
 601             nbnxn_kernel_simd_2xnn(&nbvg->nbl_lists,
 602                                    nbvg->nbat, ic,
 603                                    nbvg->ewald_excl,
 604                                    fr->shift_vec,
 605                                    flags,
 606                                    clearF,
 607                                    fr->fshift[0],
 608                                    enerd->grpp.ener[egCOULSR],
 609                                    fr->bBHAM ?
 610                                    enerd->grpp.ener[egBHAMSR] :
 611                                    enerd->grpp.ener[egLJSR]);
 612             break;
 613
 614         case nbnxnk8x8x8_CUDA:
 615             nbnxn_cuda_launch_kernel(fr->nbv->cu_nbv, nbvg->nbat, flags, ilocality);
 616             break;
 617
 618         case nbnxnk8x8x8_PlainC:
 619             nbnxn_kernel_gpu_ref(nbvg->nbl_lists.nbl[0],
 620                                  nbvg->nbat, ic,
 621                                  fr->shift_vec,
 622                                  flags,
 623                                  clearF,
 624                                  nbvg->nbat->out[0].f,
 625                                  fr->fshift[0],
 626                                  enerd->grpp.ener[egCOULSR],
 627                                  fr->bBHAM ?
 628                                  enerd->grpp.ener[egBHAMSR] :
 629                                  enerd->grpp.ener[egLJSR]);
 630             break;
 631
 632         default:
 633             gmx_incons("Invalid nonbonded kernel type passed!");
 634
 635     }
 636     if (!bCUDA)
 637     {
 638         wallcycle_sub_stop(wcycle, ewcsNONBONDED);
 639     }
 640
 641     if (EEL_RF(ic->eeltype) || ic->eeltype == eelCUT)
 642     {
 643         enr_nbnxn_kernel_ljc = eNR_NBNXN_LJ_RF;
 644     }
 645     else if ((!bCUDA && nbvg->ewald_excl == ewaldexclAnalytical) ||
 646              (bCUDA && nbnxn_cuda_is_kernel_ewald_analytical(fr->nbv->cu_nbv)))
 647     {
 648         enr_nbnxn_kernel_ljc = eNR_NBNXN_LJ_EWALD;
 649     }
 650     else
 651     {
 652         enr_nbnxn_kernel_ljc = eNR_NBNXN_LJ_TAB;
 653     }
 654     enr_nbnxn_kernel_lj = eNR_NBNXN_LJ;
 655     if (flags & GMX_FORCE_ENERGY)
 656     {
 657         /* In eNR_??? the nbnxn F+E kernels are always the F kernel + 1 */
 658         enr_nbnxn_kernel_ljc += 1;
 659         enr_nbnxn_kernel_lj  += 1;
 660     }
 661
 662     inc_nrnb(nrnb, enr_nbnxn_kernel_ljc,
 663              nbvg->nbl_lists.natpair_ljq);
 664     inc_nrnb(nrnb, enr_nbnxn_kernel_lj,
 665              nbvg->nbl_lists.natpair_lj);
 666     /* The Coulomb-only kernels are offset -eNR_NBNXN_LJ_RF+eNR_NBNXN_RF */
 667     inc_nrnb(nrnb, enr_nbnxn_kernel_ljc-eNR_NBNXN_LJ_RF+eNR_NBNXN_RF,
 668              nbvg->nbl_lists.natpair_q);
 669
 670     if (ic->vdw_modifier == eintmodFORCESWITCH)
 671     {
 672         /* We add up the switch cost separately */
 673         inc_nrnb(nrnb, eNR_NBNXN_ADD_LJ_FSW+((flags & GMX_FORCE_ENERGY) ? 1 : 0),
 674                  nbvg->nbl_lists.natpair_ljq + nbvg->nbl_lists.natpair_lj);
 675     }
 676     if (ic->vdw_modifier == eintmodPOTSWITCH)
 677     {
 678         /* We add up the switch cost separately */
 679         inc_nrnb(nrnb, eNR_NBNXN_ADD_LJ_PSW+((flags & GMX_FORCE_ENERGY) ? 1 : 0),
 680                  nbvg->nbl_lists.natpair_ljq + nbvg->nbl_lists.natpair_lj);
 681     }
 682     if (ic->vdwtype == evdwPME)
 683     {
 684         /* We add up the LJ Ewald cost separately */
 685         inc_nrnb(nrnb, eNR_NBNXN_ADD_LJ_EWALD+((flags & GMX_FORCE_ENERGY) ? 1 : 0),
 686                  nbvg->nbl_lists.natpair_ljq + nbvg->nbl_lists.natpair_lj);
 687     }
 688 }
 689
 690 void do_force_cutsVERLET(FILE *fplog, t_commrec *cr,
 691                          t_inputrec *inputrec,
 692                          gmx_int64_t step, t_nrnb *nrnb, gmx_wallcycle_t wcycle,
 693                          gmx_localtop_t *top,
 694                          gmx_groups_t gmx_unused *groups,
 695                          matrix box, rvec x[], history_t *hist,
 696                          rvec f[],
 697                          tensor vir_force,
 698                          t_mdatoms *mdatoms,
 699                          gmx_enerdata_t *enerd, t_fcdata *fcd,
 700                          real *lambda, t_graph *graph,
 701                          t_forcerec *fr, interaction_const_t *ic,
 702                          gmx_vsite_t *vsite, rvec mu_tot,
 703                          double t, FILE *field, gmx_edsam_t ed,
 704                          gmx_bool bBornRadii,
 705                          int flags)
 706 {
 707     int                 cg0, cg1, i, j;
 708     int                 start, homenr;
 709     int                 nb_kernel_type;
 710     double              mu[2*DIM];
 711     gmx_bool            bSepDVDL, bStateChanged, bNS, bFillGrid, bCalcCGCM, bBS;
 712     gmx_bool            bDoLongRange, bDoForces, bSepLRF, bUseGPU, bUseOrEmulGPU;
 713     gmx_bool            bDiffKernels = FALSE;
 714     matrix              boxs;
 715     rvec                vzero, box_diag;
 716     real                e, v, dvdl;
 717     float               cycles_pme, cycles_force, cycles_wait_gpu;
 718     nonbonded_verlet_t *nbv;
 719
 720     cycles_force    = 0;
 721     cycles_wait_gpu = 0;
 722     nbv             = fr->nbv;
 723     nb_kernel_type  = fr->nbv->grp[0].kernel_type;
 724
 725     start  = 0;
 726     homenr = mdatoms->homenr;
 727
 728     bSepDVDL = (fr->bSepDVDL && do_per_step(step, inputrec->nstlog));
 729
 730     clear_mat(vir_force);
 731
 732     cg0 = 0;
 733     if (DOMAINDECOMP(cr))
 734     {
 735         cg1 = cr->dd->ncg_tot;
 736     }
 737     else
 738     {
 739         cg1 = top->cgs.nr;
 740     }
 741     if (fr->n_tpi > 0)
 742     {
 743         cg1--;
 744     }
 745
 746     bStateChanged = (flags & GMX_FORCE_STATECHANGED);
 747     bNS           = (flags & GMX_FORCE_NS) && (fr->bAllvsAll == FALSE);
 748     bFillGrid     = (bNS && bStateChanged);
 749     bCalcCGCM     = (bFillGrid && !DOMAINDECOMP(cr));
 750     bDoLongRange  = (fr->bTwinRange && bNS && (flags & GMX_FORCE_DO_LR));
 751     bDoForces     = (flags & GMX_FORCE_FORCES);
 752     bSepLRF       = (bDoLongRange && bDoForces && (flags & GMX_FORCE_SEPLRF));
 753     bUseGPU       = fr->nbv->bUseGPU;
 754     bUseOrEmulGPU = bUseGPU || (nbv->grp[0].kernel_type == nbnxnk8x8x8_PlainC);
 755
 756     if (bStateChanged)
 757     {
 758         update_forcerec(fr, box);
 759
 760         if (NEED_MUTOT(*inputrec))
 761         {
 762             /* Calculate total (local) dipole moment in a temporary common array.
 763              * This makes it possible to sum them over nodes faster.
 764              */
 765             calc_mu(start, homenr,
 766                     x, mdatoms->chargeA, mdatoms->chargeB, mdatoms->nChargePerturbed,
 767                     mu, mu+DIM);
 768         }
 769     }
 770
 771     if (fr->ePBC != epbcNONE)
 772     {
 773         /* Compute shift vectors every step,
 774          * because of pressure coupling or box deformation!
 775          */
 776         if ((flags & GMX_FORCE_DYNAMICBOX) && bStateChanged)
 777         {
 778             calc_shifts(box, fr->shift_vec);
 779         }
 780
 781         if (bCalcCGCM)
 782         {
 783             put_atoms_in_box_omp(fr->ePBC, box, homenr, x);
 784             inc_nrnb(nrnb, eNR_SHIFTX, homenr);
 785         }
 786         else if (EI_ENERGY_MINIMIZATION(inputrec->eI) && graph)
 787         {
 788             unshift_self(graph, box, x);
 789         }
 790     }
 791
 792     nbnxn_atomdata_copy_shiftvec(flags & GMX_FORCE_DYNAMICBOX,
 793                                  fr->shift_vec, nbv->grp[0].nbat);
 794
 795 #ifdef GMX_MPI
 796     if (!(cr->duty & DUTY_PME))
 797     {
 798         /* Send particle coordinates to the pme nodes.
 799          * Since this is only implemented for domain decomposition
 800          * and domain decomposition does not use the graph,
 801          * we do not need to worry about shifting.
 802          */
 803
 804         int pme_flags = 0;
 805
 806         wallcycle_start(wcycle, ewcPP_PMESENDX);
 807
 808         bBS = (inputrec->nwall == 2);
 809         if (bBS)
 810         {
 811             copy_mat(box, boxs);
 812             svmul(inputrec->wall_ewald_zfac, boxs[ZZ], boxs[ZZ]);
 813         }
 814
 815         if (EEL_PME(fr->eeltype))
 816         {
 817             pme_flags |= GMX_PME_DO_COULOMB;
 818         }
 819
 820         if (EVDW_PME(fr->vdwtype))
 821         {
 822             pme_flags |= GMX_PME_DO_LJ;
 823             if (fr->ljpme_combination_rule == eljpmeLB)
 824             {
 825                 pme_flags |= GMX_PME_LJ_LB;
 826             }
 827         }
 828
 829         gmx_pme_send_coordinates(cr, bBS ? boxs : box, x,
 830                                  mdatoms->nChargePerturbed, mdatoms->nTypePerturbed, lambda[efptCOUL], lambda[efptVDW],
 831                                  (flags & (GMX_FORCE_VIRIAL | GMX_FORCE_ENERGY)),
 832                                  pme_flags, step);
 833
 834         wallcycle_stop(wcycle, ewcPP_PMESENDX);
 835     }
 836 #endif /* GMX_MPI */
 837
 838     /* do gridding for pair search */
 839     if (bNS)
 840     {
 841         if (graph && bStateChanged)
 842         {
 843             /* Calculate intramolecular shift vectors to make molecules whole */
 844             mk_mshift(fplog, graph, fr->ePBC, box, x);
 845         }
 846
 847         clear_rvec(vzero);
 848         box_diag[XX] = box[XX][XX];
 849         box_diag[YY] = box[YY][YY];
 850         box_diag[ZZ] = box[ZZ][ZZ];
 851
 852         wallcycle_start(wcycle, ewcNS);
 853         if (!fr->bDomDec)
 854         {
 855             wallcycle_sub_start(wcycle, ewcsNBS_GRID_LOCAL);
 856             nbnxn_put_on_grid(nbv->nbs, fr->ePBC, box,
 857                               0, vzero, box_diag,
 858                               0, mdatoms->homenr, -1, fr->cginfo, x,
 859                               0, NULL,
 860                               nbv->grp[eintLocal].kernel_type,
 861                               nbv->grp[eintLocal].nbat);
 862             wallcycle_sub_stop(wcycle, ewcsNBS_GRID_LOCAL);
 863         }
 864         else
 865         {
 866             wallcycle_sub_start(wcycle, ewcsNBS_GRID_NONLOCAL);
 867             nbnxn_put_on_grid_nonlocal(nbv->nbs, domdec_zones(cr->dd),
 868                                        fr->cginfo, x,
 869                                        nbv->grp[eintNonlocal].kernel_type,
 870                                        nbv->grp[eintNonlocal].nbat);
 871             wallcycle_sub_stop(wcycle, ewcsNBS_GRID_NONLOCAL);
 872         }
 873
 874         if (nbv->ngrp == 1 ||
 875             nbv->grp[eintNonlocal].nbat == nbv->grp[eintLocal].nbat)
 876         {
 877             nbnxn_atomdata_set(nbv->grp[eintLocal].nbat, eatAll,
 878                                nbv->nbs, mdatoms, fr->cginfo);
 879         }
 880         else
 881         {
 882             nbnxn_atomdata_set(nbv->grp[eintLocal].nbat, eatLocal,
 883                                nbv->nbs, mdatoms, fr->cginfo);
 884             nbnxn_atomdata_set(nbv->grp[eintNonlocal].nbat, eatAll,
 885                                nbv->nbs, mdatoms, fr->cginfo);
 886         }
 887         wallcycle_stop(wcycle, ewcNS);
 888     }
 889
 890     /* initialize the GPU atom data and copy shift vector */
 891     if (bUseGPU)
 892     {
 893         if (bNS)
 894         {
 895             wallcycle_start_nocount(wcycle, ewcLAUNCH_GPU_NB);
 896             nbnxn_cuda_init_atomdata(nbv->cu_nbv, nbv->grp[eintLocal].nbat);
 897             wallcycle_stop(wcycle, ewcLAUNCH_GPU_NB);
 898         }
 899
 900         wallcycle_start_nocount(wcycle, ewcLAUNCH_GPU_NB);
 901         nbnxn_cuda_upload_shiftvec(nbv->cu_nbv, nbv->grp[eintLocal].nbat);
 902         wallcycle_stop(wcycle, ewcLAUNCH_GPU_NB);
 903     }
 904
 905     /* do local pair search */
 906     if (bNS)
 907     {
 908         wallcycle_start_nocount(wcycle, ewcNS);
 909         wallcycle_sub_start(wcycle, ewcsNBS_SEARCH_LOCAL);
 910         nbnxn_make_pairlist(nbv->nbs, nbv->grp[eintLocal].nbat,
 911                             &top->excls,
 912                             ic->rlist,
 913                             nbv->min_ci_balanced,
 914                             &nbv->grp[eintLocal].nbl_lists,
 915                             eintLocal,
 916                             nbv->grp[eintLocal].kernel_type,
 917                             nrnb);
 918         wallcycle_sub_stop(wcycle, ewcsNBS_SEARCH_LOCAL);
 919
 920         if (bUseGPU)
 921         {
 922             /* initialize local pair-list on the GPU */
 923             nbnxn_cuda_init_pairlist(nbv->cu_nbv,
 924                                      nbv->grp[eintLocal].nbl_lists.nbl[0],
 925                                      eintLocal);
 926         }
 927         wallcycle_stop(wcycle, ewcNS);
 928     }
 929     else
 930     {
 931         wallcycle_start(wcycle, ewcNB_XF_BUF_OPS);
 932         wallcycle_sub_start(wcycle, ewcsNB_X_BUF_OPS);
 933         nbnxn_atomdata_copy_x_to_nbat_x(nbv->nbs, eatLocal, FALSE, x,
 934                                         nbv->grp[eintLocal].nbat);
 935         wallcycle_sub_stop(wcycle, ewcsNB_X_BUF_OPS);
 936         wallcycle_stop(wcycle, ewcNB_XF_BUF_OPS);
 937     }
 938
 939     if (bUseGPU)
 940     {
 941         wallcycle_start(wcycle, ewcLAUNCH_GPU_NB);
 942         /* launch local nonbonded F on GPU */
 943         do_nb_verlet(fr, ic, enerd, flags, eintLocal, enbvClearFNo,
 944                      nrnb, wcycle);
 945         wallcycle_stop(wcycle, ewcLAUNCH_GPU_NB);
 946     }
 947
 948     /* Communicate coordinates and sum dipole if necessary +
 949        do non-local pair search */
 950     if (DOMAINDECOMP(cr))
 951     {
 952         bDiffKernels = (nbv->grp[eintNonlocal].kernel_type !=
 953                         nbv->grp[eintLocal].kernel_type);
 954
 955         if (bDiffKernels)
 956         {
 957             /* With GPU+CPU non-bonded calculations we need to copy
 958              * the local coordinates to the non-local nbat struct
 959              * (in CPU format) as the non-local kernel call also
 960              * calculates the local - non-local interactions.
 961              */
 962             wallcycle_start(wcycle, ewcNB_XF_BUF_OPS);
 963             wallcycle_sub_start(wcycle, ewcsNB_X_BUF_OPS);
 964             nbnxn_atomdata_copy_x_to_nbat_x(nbv->nbs, eatLocal, TRUE, x,
 965                                             nbv->grp[eintNonlocal].nbat);
 966             wallcycle_sub_stop(wcycle, ewcsNB_X_BUF_OPS);
 967             wallcycle_stop(wcycle, ewcNB_XF_BUF_OPS);
 968         }
 969
 970         if (bNS)
 971         {
 972             wallcycle_start_nocount(wcycle, ewcNS);
 973             wallcycle_sub_start(wcycle, ewcsNBS_SEARCH_NONLOCAL);
 974
 975             if (bDiffKernels)
 976             {
 977                 nbnxn_grid_add_simple(nbv->nbs, nbv->grp[eintNonlocal].nbat);
 978             }
 979
 980             nbnxn_make_pairlist(nbv->nbs, nbv->grp[eintNonlocal].nbat,
 981                                 &top->excls,
 982                                 ic->rlist,
 983                                 nbv->min_ci_balanced,
 984                                 &nbv->grp[eintNonlocal].nbl_lists,
 985                                 eintNonlocal,
 986                                 nbv->grp[eintNonlocal].kernel_type,
 987                                 nrnb);
 988
 989             wallcycle_sub_stop(wcycle, ewcsNBS_SEARCH_NONLOCAL);
 990
 991             if (nbv->grp[eintNonlocal].kernel_type == nbnxnk8x8x8_CUDA)
 992             {
 993                 /* initialize non-local pair-list on the GPU */
 994                 nbnxn_cuda_init_pairlist(nbv->cu_nbv,
 995                                          nbv->grp[eintNonlocal].nbl_lists.nbl[0],
 996                                          eintNonlocal);
 997             }
 998             wallcycle_stop(wcycle, ewcNS);
 999         }
1000         else
1001         {
1002             wallcycle_start(wcycle, ewcMOVEX);
1003             dd_move_x(cr->dd, box, x);
1004
1005             /* When we don't need the total dipole we sum it in global_stat */
1006             if (bStateChanged && NEED_MUTOT(*inputrec))
1007             {
1008                 gmx_sumd(2*DIM, mu, cr);
1009             }
1010             wallcycle_stop(wcycle, ewcMOVEX);
1011
1012             wallcycle_start(wcycle, ewcNB_XF_BUF_OPS);
1013             wallcycle_sub_start(wcycle, ewcsNB_X_BUF_OPS);
1014             nbnxn_atomdata_copy_x_to_nbat_x(nbv->nbs, eatNonlocal, FALSE, x,
1015                                             nbv->grp[eintNonlocal].nbat);
1016             wallcycle_sub_stop(wcycle, ewcsNB_X_BUF_OPS);
1017             cycles_force += wallcycle_stop(wcycle, ewcNB_XF_BUF_OPS);
1018         }
1019
1020         if (bUseGPU && !bDiffKernels)
1021         {
1022             wallcycle_start(wcycle, ewcLAUNCH_GPU_NB);
1023             /* launch non-local nonbonded F on GPU */
1024             do_nb_verlet(fr, ic, enerd, flags, eintNonlocal, enbvClearFNo,
1025                          nrnb, wcycle);
1026             cycles_force += wallcycle_stop(wcycle, ewcLAUNCH_GPU_NB);
1027         }
1028     }
1029
1030     if (bUseGPU)
1031     {
1032         /* launch D2H copy-back F */
1033         wallcycle_start_nocount(wcycle, ewcLAUNCH_GPU_NB);
1034         if (DOMAINDECOMP(cr) && !bDiffKernels)
1035         {
1036             nbnxn_cuda_launch_cpyback(nbv->cu_nbv, nbv->grp[eintNonlocal].nbat,
1037                                       flags, eatNonlocal);
1038         }
1039         nbnxn_cuda_launch_cpyback(nbv->cu_nbv, nbv->grp[eintLocal].nbat,
1040                                   flags, eatLocal);
1041         cycles_force += wallcycle_stop(wcycle, ewcLAUNCH_GPU_NB);
1042     }
1043
1044     if (bStateChanged && NEED_MUTOT(*inputrec))
1045     {
1046         if (PAR(cr))
1047         {
1048             gmx_sumd(2*DIM, mu, cr);
1049         }
1050
1051         for (i = 0; i < 2; i++)
1052         {
1053             for (j = 0; j < DIM; j++)
1054             {
1055                 fr->mu_tot[i][j] = mu[i*DIM + j];
1056             }
1057         }
1058     }
1059     if (fr->efep == efepNO)
1060     {
1061         copy_rvec(fr->mu_tot[0], mu_tot);
1062     }
1063     else
1064     {
1065         for (j = 0; j < DIM; j++)
1066         {
1067             mu_tot[j] =
1068                 (1.0 - lambda[efptCOUL])*fr->mu_tot[0][j] +
1069                 lambda[efptCOUL]*fr->mu_tot[1][j];
1070         }
1071     }
1072
1073     /* Reset energies */
1074     reset_enerdata(fr, bNS, enerd, MASTER(cr));
1075     clear_rvecs(SHIFTS, fr->fshift);
1076
1077     if (DOMAINDECOMP(cr) && !(cr->duty & DUTY_PME))
1078     {
1079         wallcycle_start(wcycle, ewcPPDURINGPME);
1080         dd_force_flop_start(cr->dd, nrnb);
1081     }
1082
1083     if (inputrec->bRot)
1084     {
1085         /* Enforced rotation has its own cycle counter that starts after the collective
1086          * coordinates have been communicated. It is added to ddCyclF to allow
1087          * for proper load-balancing */
1088         wallcycle_start(wcycle, ewcROT);
1089         do_rotation(cr, inputrec, box, x, t, step, wcycle, bNS);
1090         wallcycle_stop(wcycle, ewcROT);
1091     }
1092
1093     /* Start the force cycle counter.
1094      * This counter is stopped in do_forcelow_level.
1095      * No parallel communication should occur while this counter is running,
1096      * since that will interfere with the dynamic load balancing.
1097      */
1098     wallcycle_start(wcycle, ewcFORCE);
1099     if (bDoForces)
1100     {
1101         /* Reset forces for which the virial is calculated separately:
1102          * PME/Ewald forces if necessary */
1103         if (fr->bF_NoVirSum)
1104         {
1105             if (flags & GMX_FORCE_VIRIAL)
1106             {
1107                 fr->f_novirsum = fr->f_novirsum_alloc;
1108                 if (fr->bDomDec)
1109                 {
1110                     clear_rvecs(fr->f_novirsum_n, fr->f_novirsum);
1111                 }
1112                 else
1113                 {
1114                     clear_rvecs(homenr, fr->f_novirsum+start);
1115                 }
1116             }
1117             else
1118             {
1119                 /* We are not calculating the pressure so we do not need
1120                  * a separate array for forces that do not contribute
1121                  * to the pressure.
1122                  */
1123                 fr->f_novirsum = f;
1124             }
1125         }
1126
1127         /* Clear the short- and long-range forces */
1128         clear_rvecs(fr->natoms_force_constr, f);
1129         if (bSepLRF && do_per_step(step, inputrec->nstcalclr))
1130         {
1131             clear_rvecs(fr->natoms_force_constr, fr->f_twin);
1132         }
1133
1134         clear_rvec(fr->vir_diag_posres);
1135     }
1136
1137     if (inputrec->ePull == epullCONSTRAINT)
1138     {
1139         clear_pull_forces(inputrec->pull);
1140     }
1141
1142     /* We calculate the non-bonded forces, when done on the CPU, here.
1143      * We do this before calling do_force_lowlevel, as in there bondeds
1144      * forces are calculated before PME, which does communication.
1145      * With this order, non-bonded and bonded force calculation imbalance
1146      * can be balanced out by the domain decomposition load balancing.
1147      */
1148
1149     if (!bUseOrEmulGPU)
1150     {
1151         /* Maybe we should move this into do_force_lowlevel */
1152         do_nb_verlet(fr, ic, enerd, flags, eintLocal, enbvClearFYes,
1153                      nrnb, wcycle);
1154     }
1155
1156     if (!bUseOrEmulGPU || bDiffKernels)
1157     {
1158         int aloc;
1159
1160         if (DOMAINDECOMP(cr))
1161         {
1162             do_nb_verlet(fr, ic, enerd, flags, eintNonlocal,
1163                          bDiffKernels ? enbvClearFYes : enbvClearFNo,
1164                          nrnb, wcycle);
1165         }
1166
1167         if (!bUseOrEmulGPU)
1168         {
1169             aloc = eintLocal;
1170         }
1171         else
1172         {
1173             aloc = eintNonlocal;
1174         }
1175
1176         /* Add all the non-bonded force to the normal force array.
1177          * This can be split into a local a non-local part when overlapping
1178          * communication with calculation with domain decomposition.
1179          */
1180         cycles_force += wallcycle_stop(wcycle, ewcFORCE);
1181         wallcycle_start(wcycle, ewcNB_XF_BUF_OPS);
1182         wallcycle_sub_start(wcycle, ewcsNB_F_BUF_OPS);
1183         nbnxn_atomdata_add_nbat_f_to_f(nbv->nbs, eatAll, nbv->grp[aloc].nbat, f);
1184         wallcycle_sub_stop(wcycle, ewcsNB_F_BUF_OPS);
1185         cycles_force += wallcycle_stop(wcycle, ewcNB_XF_BUF_OPS);
1186         wallcycle_start_nocount(wcycle, ewcFORCE);
1187
1188         /* if there are multiple fshift output buffers reduce them */
1189         if ((flags & GMX_FORCE_VIRIAL) &&
1190             nbv->grp[aloc].nbl_lists.nnbl > 1)
1191         {
1192             nbnxn_atomdata_add_nbat_fshift_to_fshift(nbv->grp[aloc].nbat,
1193                                                      fr->fshift);
1194         }
1195     }
1196
1197     /* update QMMMrec, if necessary */
1198     if (fr->bQMMM)
1199     {
1200         update_QMMMrec(cr, fr, x, mdatoms, box, top);
1201     }
1202
1203     if ((flags & GMX_FORCE_BONDED) && top->idef.il[F_POSRES].nr > 0)
1204     {
1205         posres_wrapper(fplog, flags, bSepDVDL, inputrec, nrnb, top, box, x,
1206                        enerd, lambda, fr);
1207     }
1208
1209     if ((flags & GMX_FORCE_BONDED) && top->idef.il[F_FBPOSRES].nr > 0)
1210     {
1211         fbposres_wrapper(inputrec, nrnb, top, box, x, enerd, fr);
1212     }
1213
1214     /* Compute the bonded and non-bonded energies and optionally forces */
1215     do_force_lowlevel(fplog, step, fr, inputrec, &(top->idef),
1216                       cr, nrnb, wcycle, mdatoms,
1217                       x, hist, f, bSepLRF ? fr->f_twin : f, enerd, fcd, top, fr->born,
1218                       &(top->atomtypes), bBornRadii, box,
1219                       inputrec->fepvals, lambda, graph, &(top->excls), fr->mu_tot,
1220                       flags, &cycles_pme);
1221
1222     if (bSepLRF)
1223     {
1224         if (do_per_step(step, inputrec->nstcalclr))
1225         {
1226             /* Add the long range forces to the short range forces */
1227             for (i = 0; i < fr->natoms_force_constr; i++)
1228             {
1229                 rvec_add(fr->f_twin[i], f[i], f[i]);
1230             }
1231         }
1232     }
1233
1234     cycles_force += wallcycle_stop(wcycle, ewcFORCE);
1235
1236     if (ed)
1237     {
1238         do_flood(cr, inputrec, x, f, ed, box, step, bNS);
1239     }
1240
1241     if (bUseOrEmulGPU && !bDiffKernels)
1242     {
1243         /* wait for non-local forces (or calculate in emulation mode) */
1244         if (DOMAINDECOMP(cr))
1245         {
1246             if (bUseGPU)
1247             {
1248                 float cycles_tmp;
1249
1250                 wallcycle_start(wcycle, ewcWAIT_GPU_NB_NL);
1251                 nbnxn_cuda_wait_gpu(nbv->cu_nbv,
1252                                     nbv->grp[eintNonlocal].nbat,
1253                                     flags, eatNonlocal,
1254                                     enerd->grpp.ener[egLJSR], enerd->grpp.ener[egCOULSR],
1255                                     fr->fshift);
1256                 cycles_tmp       = wallcycle_stop(wcycle, ewcWAIT_GPU_NB_NL);
1257                 cycles_wait_gpu += cycles_tmp;
1258                 cycles_force    += cycles_tmp;
1259             }
1260             else
1261             {
1262                 wallcycle_start_nocount(wcycle, ewcFORCE);
1263                 do_nb_verlet(fr, ic, enerd, flags, eintNonlocal, enbvClearFYes,
1264                              nrnb, wcycle);
1265                 cycles_force += wallcycle_stop(wcycle, ewcFORCE);
1266             }
1267             wallcycle_start(wcycle, ewcNB_XF_BUF_OPS);
1268             wallcycle_sub_start(wcycle, ewcsNB_F_BUF_OPS);
1269             /* skip the reduction if there was no non-local work to do */
1270             if (nbv->grp[eintLocal].nbl_lists.nbl[0]->nsci > 0)
1271             {
1272                 nbnxn_atomdata_add_nbat_f_to_f(nbv->nbs, eatNonlocal,
1273                                                nbv->grp[eintNonlocal].nbat, f);
1274             }
1275             wallcycle_sub_stop(wcycle, ewcsNB_F_BUF_OPS);
1276             cycles_force += wallcycle_stop(wcycle, ewcNB_XF_BUF_OPS);
1277         }
1278     }
1279
1280     if (bDoForces && DOMAINDECOMP(cr))
1281     {
1282         /* Communicate the forces */
1283         wallcycle_start(wcycle, ewcMOVEF);
1284         dd_move_f(cr->dd, f, fr->fshift);
1285         /* Do we need to communicate the separate force array
1286          * for terms that do not contribute to the single sum virial?
1287          * Position restraints and electric fields do not introduce
1288          * inter-cg forces, only full electrostatics methods do.
1289          * When we do not calculate the virial, fr->f_novirsum = f,
1290          * so we have already communicated these forces.
1291          */
1292         if (EEL_FULL(fr->eeltype) && cr->dd->n_intercg_excl &&
1293             (flags & GMX_FORCE_VIRIAL))
1294         {
1295             dd_move_f(cr->dd, fr->f_novirsum, NULL);
1296         }
1297         if (bSepLRF)
1298         {
1299             /* We should not update the shift forces here,
1300              * since f_twin is already included in f.
1301              */
1302             dd_move_f(cr->dd, fr->f_twin, NULL);
1303         }
1304         wallcycle_stop(wcycle, ewcMOVEF);
1305     }
1306
1307     if (bUseOrEmulGPU)
1308     {
1309         /* wait for local forces (or calculate in emulation mode) */
1310         if (bUseGPU)
1311         {
1312             wallcycle_start(wcycle, ewcWAIT_GPU_NB_L);
1313             nbnxn_cuda_wait_gpu(nbv->cu_nbv,
1314                                 nbv->grp[eintLocal].nbat,
1315                                 flags, eatLocal,
1316                                 enerd->grpp.ener[egLJSR], enerd->grpp.ener[egCOULSR],
1317                                 fr->fshift);
1318             cycles_wait_gpu += wallcycle_stop(wcycle, ewcWAIT_GPU_NB_L);
1319
1320             /* now clear the GPU outputs while we finish the step on the CPU */
1321
1322             wallcycle_start_nocount(wcycle, ewcLAUNCH_GPU_NB);
1323             nbnxn_cuda_clear_outputs(nbv->cu_nbv, flags);
1324             wallcycle_stop(wcycle, ewcLAUNCH_GPU_NB);
1325         }
1326         else
1327         {
1328             wallcycle_start_nocount(wcycle, ewcFORCE);
1329             do_nb_verlet(fr, ic, enerd, flags, eintLocal,
1330                          DOMAINDECOMP(cr) ? enbvClearFNo : enbvClearFYes,
1331                          nrnb, wcycle);
1332             wallcycle_stop(wcycle, ewcFORCE);
1333         }
1334         wallcycle_start(wcycle, ewcNB_XF_BUF_OPS);
1335         wallcycle_sub_start(wcycle, ewcsNB_F_BUF_OPS);
1336         if (nbv->grp[eintLocal].nbl_lists.nbl[0]->nsci > 0)
1337         {
1338             /* skip the reduction if there was no non-local work to do */
1339             nbnxn_atomdata_add_nbat_f_to_f(nbv->nbs, eatLocal,
1340                                            nbv->grp[eintLocal].nbat, f);
1341         }
1342         wallcycle_sub_stop(wcycle, ewcsNB_F_BUF_OPS);
1343         wallcycle_stop(wcycle, ewcNB_XF_BUF_OPS);
1344     }
1345
1346     if (DOMAINDECOMP(cr))
1347     {
1348         dd_force_flop_stop(cr->dd, nrnb);
1349         if (wcycle)
1350         {
1351             dd_cycles_add(cr->dd, cycles_force-cycles_pme, ddCyclF);
1352             if (bUseGPU)
1353             {
1354                 dd_cycles_add(cr->dd, cycles_wait_gpu, ddCyclWaitGPU);
1355             }
1356         }
1357     }
1358
1359     if (bDoForces)
1360     {
1361         if (IR_ELEC_FIELD(*inputrec))
1362         {
1363             /* Compute forces due to electric field */
1364             calc_f_el(MASTER(cr) ? field : NULL,
1365                       start, homenr, mdatoms->chargeA, fr->f_novirsum,
1366                       inputrec->ex, inputrec->et, t);
1367         }
1368
1369         /* If we have NoVirSum forces, but we do not calculate the virial,
1370          * we sum fr->f_novirum=f later.
1371          */
1372         if (vsite && !(fr->bF_NoVirSum && !(flags & GMX_FORCE_VIRIAL)))
1373         {
1374             wallcycle_start(wcycle, ewcVSITESPREAD);
1375             spread_vsite_f(vsite, x, f, fr->fshift, FALSE, NULL, nrnb,
1376                            &top->idef, fr->ePBC, fr->bMolPBC, graph, box, cr);
1377             wallcycle_stop(wcycle, ewcVSITESPREAD);
1378
1379             if (bSepLRF)
1380             {
1381                 wallcycle_start(wcycle, ewcVSITESPREAD);
1382                 spread_vsite_f(vsite, x, fr->f_twin, NULL, FALSE, NULL,
1383                                nrnb,
1384                                &top->idef, fr->ePBC, fr->bMolPBC, graph, box, cr);
1385                 wallcycle_stop(wcycle, ewcVSITESPREAD);
1386             }
1387         }
1388
1389         if (flags & GMX_FORCE_VIRIAL)
1390         {
1391             /* Calculation of the virial must be done after vsites! */
1392             calc_virial(0, mdatoms->homenr, x, f,
1393                         vir_force, graph, box, nrnb, fr, inputrec->ePBC);
1394         }
1395     }
1396
1397     if (inputrec->ePull == epullUMBRELLA || inputrec->ePull == epullCONST_F)
1398     {
1399         pull_potential_wrapper(fplog, bSepDVDL, cr, inputrec, box, x,
1400                                f, vir_force, mdatoms, enerd, lambda, t);
1401     }
1402
1403     /* Add the forces from enforced rotation potentials (if any) */
1404     if (inputrec->bRot)
1405     {
1406         wallcycle_start(wcycle, ewcROTadd);
1407         enerd->term[F_COM_PULL] += add_rot_forces(inputrec->rot, f, cr, step, t);
1408         wallcycle_stop(wcycle, ewcROTadd);
1409     }
1410
1411     if (PAR(cr) && !(cr->duty & DUTY_PME))
1412     {
1413         /* In case of node-splitting, the PP nodes receive the long-range
1414          * forces, virial and energy from the PME nodes here.
1415          */
1416         pme_receive_force_ener(fplog, bSepDVDL, cr, wcycle, enerd, fr);
1417     }
1418
1419     if (bDoForces)
1420     {
1421         post_process_forces(cr, step, nrnb, wcycle,
1422                             top, box, x, f, vir_force, mdatoms, graph, fr, vsite,
1423                             flags);
1424     }
1425
1426     /* Sum the potential energy terms from group contributions */
1427     sum_epot(&(enerd->grpp), enerd->term);
1428 }
1429
1430 void do_force_cutsGROUP(FILE *fplog, t_commrec *cr,
1431                         t_inputrec *inputrec,
1432                         gmx_int64_t step, t_nrnb *nrnb, gmx_wallcycle_t wcycle,
1433                         gmx_localtop_t *top,
1434                         gmx_groups_t *groups,
1435                         matrix box, rvec x[], history_t *hist,
1436                         rvec f[],
1437                         tensor vir_force,
1438                         t_mdatoms *mdatoms,
1439                         gmx_enerdata_t *enerd, t_fcdata *fcd,
1440                         real *lambda, t_graph *graph,
1441                         t_forcerec *fr, gmx_vsite_t *vsite, rvec mu_tot,
1442                         double t, FILE *field, gmx_edsam_t ed,
1443                         gmx_bool bBornRadii,
1444                         int flags)
1445 {
1446     int        cg0, cg1, i, j;
1447     int        start, homenr;
1448     double     mu[2*DIM];
1449     gmx_bool   bSepDVDL, bStateChanged, bNS, bFillGrid, bCalcCGCM, bBS;
1450     gmx_bool   bDoLongRangeNS, bDoForces, bDoPotential, bSepLRF;
1451     gmx_bool   bDoAdressWF;
1452     matrix     boxs;
1453     rvec       vzero, box_diag;
1454     real       e, v, dvdlambda[efptNR];
1455     t_pbc      pbc;
1456     float      cycles_pme, cycles_force;
1457
1458     start  = 0;
1459     homenr = mdatoms->homenr;
1460
1461     bSepDVDL = (fr->bSepDVDL && do_per_step(step, inputrec->nstlog));
1462
1463     clear_mat(vir_force);
1464
1465     cg0 = 0;
1466     if (DOMAINDECOMP(cr))
1467     {
1468         cg1 = cr->dd->ncg_tot;
1469     }
1470     else
1471     {
1472         cg1 = top->cgs.nr;
1473     }
1474     if (fr->n_tpi > 0)
1475     {
1476         cg1--;
1477     }
1478
1479     bStateChanged  = (flags & GMX_FORCE_STATECHANGED);
1480     bNS            = (flags & GMX_FORCE_NS) && (fr->bAllvsAll == FALSE);
1481     /* Should we update the long-range neighborlists at this step? */
1482     bDoLongRangeNS = fr->bTwinRange && bNS;
1483     /* Should we perform the long-range nonbonded evaluation inside the neighborsearching? */
1484     bFillGrid      = (bNS && bStateChanged);
1485     bCalcCGCM      = (bFillGrid && !DOMAINDECOMP(cr));
1486     bDoForces      = (flags & GMX_FORCE_FORCES);
1487     bDoPotential   = (flags & GMX_FORCE_ENERGY);
1488     bSepLRF        = ((inputrec->nstcalclr > 1) && bDoForces &&
1489                       (flags & GMX_FORCE_SEPLRF) && (flags & GMX_FORCE_DO_LR));
1490
1491     /* should probably move this to the forcerec since it doesn't change */
1492     bDoAdressWF   = ((fr->adress_type != eAdressOff));
1493
1494     if (bStateChanged)
1495     {
1496         update_forcerec(fr, box);
1497
1498         if (NEED_MUTOT(*inputrec))
1499         {
1500             /* Calculate total (local) dipole moment in a temporary common array.
1501              * This makes it possible to sum them over nodes faster.
1502              */
1503             calc_mu(start, homenr,
1504                     x, mdatoms->chargeA, mdatoms->chargeB, mdatoms->nChargePerturbed,
1505                     mu, mu+DIM);
1506         }
1507     }
1508
1509     if (fr->ePBC != epbcNONE)
1510     {
1511         /* Compute shift vectors every step,
1512          * because of pressure coupling or box deformation!
1513          */
1514         if ((flags & GMX_FORCE_DYNAMICBOX) && bStateChanged)
1515         {
1516             calc_shifts(box, fr->shift_vec);
1517         }
1518
1519         if (bCalcCGCM)
1520         {
1521             put_charge_groups_in_box(fplog, cg0, cg1, fr->ePBC, box,
1522                                      &(top->cgs), x, fr->cg_cm);
1523             inc_nrnb(nrnb, eNR_CGCM, homenr);
1524             inc_nrnb(nrnb, eNR_RESETX, cg1-cg0);
1525         }
1526         else if (EI_ENERGY_MINIMIZATION(inputrec->eI) && graph)
1527         {
1528             unshift_self(graph, box, x);
1529         }
1530     }
1531     else if (bCalcCGCM)
1532     {
1533         calc_cgcm(fplog, cg0, cg1, &(top->cgs), x, fr->cg_cm);
1534         inc_nrnb(nrnb, eNR_CGCM, homenr);
1535     }
1536
1537     if (bCalcCGCM && gmx_debug_at)
1538     {
1539         pr_rvecs(debug, 0, "cgcm", fr->cg_cm, top->cgs.nr);
1540     }
1541
1542 #ifdef GMX_MPI
1543     if (!(cr->duty & DUTY_PME))
1544     {
1545         /* Send particle coordinates to the pme nodes.
1546          * Since this is only implemented for domain decomposition
1547          * and domain decomposition does not use the graph,
1548          * we do not need to worry about shifting.
1549          */
1550
1551         int pme_flags = 0;
1552
1553         wallcycle_start(wcycle, ewcPP_PMESENDX);
1554
1555         bBS = (inputrec->nwall == 2);
1556         if (bBS)
1557         {
1558             copy_mat(box, boxs);
1559             svmul(inputrec->wall_ewald_zfac, boxs[ZZ], boxs[ZZ]);
1560         }
1561
1562         if (EEL_PME(fr->eeltype))
1563         {
1564             pme_flags |= GMX_PME_DO_COULOMB;
1565         }
1566
1567         if (EVDW_PME(fr->vdwtype))
1568         {
1569             pme_flags |= GMX_PME_DO_LJ;
1570             if (fr->ljpme_combination_rule == eljpmeLB)
1571             {
1572                 pme_flags |= GMX_PME_LJ_LB;
1573             }
1574         }
1575
1576         gmx_pme_send_coordinates(cr, bBS ? boxs : box, x,
1577                                  mdatoms->nChargePerturbed, mdatoms->nTypePerturbed, lambda[efptCOUL], lambda[efptVDW],
1578                                  (flags & (GMX_FORCE_VIRIAL | GMX_FORCE_ENERGY)),
1579                                  pme_flags, step);
1580
1581         wallcycle_stop(wcycle, ewcPP_PMESENDX);
1582     }
1583 #endif /* GMX_MPI */
1584
1585     /* Communicate coordinates and sum dipole if necessary */
1586     if (DOMAINDECOMP(cr))
1587     {
1588         wallcycle_start(wcycle, ewcMOVEX);
1589         dd_move_x(cr->dd, box, x);
1590         wallcycle_stop(wcycle, ewcMOVEX);
1591     }
1592
1593     /* update adress weight beforehand */
1594     if (bStateChanged && bDoAdressWF)
1595     {
1596         /* need pbc for adress weight calculation with pbc_dx */
1597         set_pbc(&pbc, inputrec->ePBC, box);
1598         if (fr->adress_site == eAdressSITEcog)
1599         {
1600             update_adress_weights_cog(top->idef.iparams, top->idef.il, x, fr, mdatoms,
1601                                       inputrec->ePBC == epbcNONE ? NULL : &pbc);
1602         }
1603         else if (fr->adress_site == eAdressSITEcom)
1604         {
1605             update_adress_weights_com(fplog, cg0, cg1, &(top->cgs), x, fr, mdatoms,
1606                                       inputrec->ePBC == epbcNONE ? NULL : &pbc);
1607         }
1608         else if (fr->adress_site == eAdressSITEatomatom)
1609         {
1610             update_adress_weights_atom_per_atom(cg0, cg1, &(top->cgs), x, fr, mdatoms,
1611                                                 inputrec->ePBC == epbcNONE ? NULL : &pbc);
1612         }
1613         else
1614         {
1615             update_adress_weights_atom(cg0, cg1, &(top->cgs), x, fr, mdatoms,
1616                                        inputrec->ePBC == epbcNONE ? NULL : &pbc);
1617         }
1618     }
1619
1620     if (NEED_MUTOT(*inputrec))
1621     {
1622
1623         if (bStateChanged)
1624         {
1625             if (PAR(cr))
1626             {
1627                 gmx_sumd(2*DIM, mu, cr);
1628             }
1629             for (i = 0; i < 2; i++)
1630             {
1631                 for (j = 0; j < DIM; j++)
1632                 {
1633                     fr->mu_tot[i][j] = mu[i*DIM + j];
1634                 }
1635             }
1636         }
1637         if (fr->efep == efepNO)
1638         {
1639             copy_rvec(fr->mu_tot[0], mu_tot);
1640         }
1641         else
1642         {
1643             for (j = 0; j < DIM; j++)
1644             {
1645                 mu_tot[j] =
1646                     (1.0 - lambda[efptCOUL])*fr->mu_tot[0][j] + lambda[efptCOUL]*fr->mu_tot[1][j];
1647             }
1648         }
1649     }
1650
1651     /* Reset energies */
1652     reset_enerdata(fr, bNS, enerd, MASTER(cr));
1653     clear_rvecs(SHIFTS, fr->fshift);
1654
1655     if (bNS)
1656     {
1657         wallcycle_start(wcycle, ewcNS);
1658
1659         if (graph && bStateChanged)
1660         {
1661             /* Calculate intramolecular shift vectors to make molecules whole */
1662             mk_mshift(fplog, graph, fr->ePBC, box, x);
1663         }
1664
1665         /* Do the actual neighbour searching */
1666         ns(fplog, fr, box,
1667            groups, top, mdatoms,
1668            cr, nrnb, bFillGrid,
1669            bDoLongRangeNS);
1670
1671         wallcycle_stop(wcycle, ewcNS);
1672     }
1673
1674     if (inputrec->implicit_solvent && bNS)
1675     {
1676         make_gb_nblist(cr, inputrec->gb_algorithm,
1677                        x, box, fr, &top->idef, graph, fr->born);
1678     }
1679
1680     if (DOMAINDECOMP(cr) && !(cr->duty & DUTY_PME))
1681     {
1682         wallcycle_start(wcycle, ewcPPDURINGPME);
1683         dd_force_flop_start(cr->dd, nrnb);
1684     }
1685
1686     if (inputrec->bRot)
1687     {
1688         /* Enforced rotation has its own cycle counter that starts after the collective
1689          * coordinates have been communicated. It is added to ddCyclF to allow
1690          * for proper load-balancing */
1691         wallcycle_start(wcycle, ewcROT);
1692         do_rotation(cr, inputrec, box, x, t, step, wcycle, bNS);
1693         wallcycle_stop(wcycle, ewcROT);
1694     }
1695
1696     /* Start the force cycle counter.
1697      * This counter is stopped in do_forcelow_level.
1698      * No parallel communication should occur while this counter is running,
1699      * since that will interfere with the dynamic load balancing.
1700      */
1701     wallcycle_start(wcycle, ewcFORCE);
1702
1703     if (bDoForces)
1704     {
1705         /* Reset forces for which the virial is calculated separately:
1706          * PME/Ewald forces if necessary */
1707         if (fr->bF_NoVirSum)
1708         {
1709             if (flags & GMX_FORCE_VIRIAL)
1710             {
1711                 fr->f_novirsum = fr->f_novirsum_alloc;
1712                 if (fr->bDomDec)
1713                 {
1714                     clear_rvecs(fr->f_novirsum_n, fr->f_novirsum);
1715                 }
1716                 else
1717                 {
1718                     clear_rvecs(homenr, fr->f_novirsum+start);
1719                 }
1720             }
1721             else
1722             {
1723                 /* We are not calculating the pressure so we do not need
1724                  * a separate array for forces that do not contribute
1725                  * to the pressure.
1726                  */
1727                 fr->f_novirsum = f;
1728             }
1729         }
1730
1731         /* Clear the short- and long-range forces */
1732         clear_rvecs(fr->natoms_force_constr, f);
1733         if (bSepLRF && do_per_step(step, inputrec->nstcalclr))
1734         {
1735             clear_rvecs(fr->natoms_force_constr, fr->f_twin);
1736         }
1737
1738         clear_rvec(fr->vir_diag_posres);
1739     }
1740     if (inputrec->ePull == epullCONSTRAINT)
1741     {
1742         clear_pull_forces(inputrec->pull);
1743     }
1744
1745     /* update QMMMrec, if necessary */
1746     if (fr->bQMMM)
1747     {
1748         update_QMMMrec(cr, fr, x, mdatoms, box, top);
1749     }
1750
1751     if ((flags & GMX_FORCE_BONDED) && top->idef.il[F_POSRES].nr > 0)
1752     {
1753         posres_wrapper(fplog, flags, bSepDVDL, inputrec, nrnb, top, box, x,
1754                        enerd, lambda, fr);
1755     }
1756
1757     if ((flags & GMX_FORCE_BONDED) && top->idef.il[F_FBPOSRES].nr > 0)
1758     {
1759         fbposres_wrapper(inputrec, nrnb, top, box, x, enerd, fr);
1760     }
1761
1762     /* Compute the bonded and non-bonded energies and optionally forces */
1763     do_force_lowlevel(fplog, step, fr, inputrec, &(top->idef),
1764                       cr, nrnb, wcycle, mdatoms,
1765                       x, hist, f, bSepLRF ? fr->f_twin : f, enerd, fcd, top, fr->born,
1766                       &(top->atomtypes), bBornRadii, box,
1767                       inputrec->fepvals, lambda,
1768                       graph, &(top->excls), fr->mu_tot,
1769                       flags,
1770                       &cycles_pme);
1771
1772     if (bSepLRF)
1773     {
1774         if (do_per_step(step, inputrec->nstcalclr))
1775         {
1776             /* Add the long range forces to the short range forces */
1777             for (i = 0; i < fr->natoms_force_constr; i++)
1778             {
1779                 rvec_add(fr->f_twin[i], f[i], f[i]);
1780             }
1781         }
1782     }
1783
1784     cycles_force = wallcycle_stop(wcycle, ewcFORCE);
1785
1786     if (ed)
1787     {
1788         do_flood(cr, inputrec, x, f, ed, box, step, bNS);
1789     }
1790
1791     if (DOMAINDECOMP(cr))
1792     {
1793         dd_force_flop_stop(cr->dd, nrnb);
1794         if (wcycle)
1795         {
1796             dd_cycles_add(cr->dd, cycles_force-cycles_pme, ddCyclF);
1797         }
1798     }
1799
1800     if (bDoForces)
1801     {
1802         if (IR_ELEC_FIELD(*inputrec))
1803         {
1804             /* Compute forces due to electric field */
1805             calc_f_el(MASTER(cr) ? field : NULL,
1806                       start, homenr, mdatoms->chargeA, fr->f_novirsum,
1807                       inputrec->ex, inputrec->et, t);
1808         }
1809
1810         if (bDoAdressWF && fr->adress_icor == eAdressICThermoForce)
1811         {
1812             /* Compute thermodynamic force in hybrid AdResS region */
1813             adress_thermo_force(start, homenr, &(top->cgs), x, fr->f_novirsum, fr, mdatoms,
1814                                 inputrec->ePBC == epbcNONE ? NULL : &pbc);
1815         }
1816
1817         /* Communicate the forces */
1818         if (DOMAINDECOMP(cr))
1819         {
1820             wallcycle_start(wcycle, ewcMOVEF);
1821             dd_move_f(cr->dd, f, fr->fshift);
1822             /* Do we need to communicate the separate force array
1823              * for terms that do not contribute to the single sum virial?
1824              * Position restraints and electric fields do not introduce
1825              * inter-cg forces, only full electrostatics methods do.
1826              * When we do not calculate the virial, fr->f_novirsum = f,
1827              * so we have already communicated these forces.
1828              */
1829             if (EEL_FULL(fr->eeltype) && cr->dd->n_intercg_excl &&
1830                 (flags & GMX_FORCE_VIRIAL))
1831             {
1832                 dd_move_f(cr->dd, fr->f_novirsum, NULL);
1833             }
1834             if (bSepLRF)
1835             {
1836                 /* We should not update the shift forces here,
1837                  * since f_twin is already included in f.
1838                  */
1839                 dd_move_f(cr->dd, fr->f_twin, NULL);
1840             }
1841             wallcycle_stop(wcycle, ewcMOVEF);
1842         }
1843
1844         /* If we have NoVirSum forces, but we do not calculate the virial,
1845          * we sum fr->f_novirum=f later.
1846          */
1847         if (vsite && !(fr->bF_NoVirSum && !(flags & GMX_FORCE_VIRIAL)))
1848         {
1849             wallcycle_start(wcycle, ewcVSITESPREAD);
1850             spread_vsite_f(vsite, x, f, fr->fshift, FALSE, NULL, nrnb,
1851                            &top->idef, fr->ePBC, fr->bMolPBC, graph, box, cr);
1852             wallcycle_stop(wcycle, ewcVSITESPREAD);
1853
1854             if (bSepLRF)
1855             {
1856                 wallcycle_start(wcycle, ewcVSITESPREAD);
1857                 spread_vsite_f(vsite, x, fr->f_twin, NULL, FALSE, NULL,
1858                                nrnb,
1859                                &top->idef, fr->ePBC, fr->bMolPBC, graph, box, cr);
1860                 wallcycle_stop(wcycle, ewcVSITESPREAD);
1861             }
1862         }
1863
1864         if (flags & GMX_FORCE_VIRIAL)
1865         {
1866             /* Calculation of the virial must be done after vsites! */
1867             calc_virial(0, mdatoms->homenr, x, f,
1868                         vir_force, graph, box, nrnb, fr, inputrec->ePBC);
1869         }
1870     }
1871
1872     if (inputrec->ePull == epullUMBRELLA || inputrec->ePull == epullCONST_F)
1873     {
1874         pull_potential_wrapper(fplog, bSepDVDL, cr, inputrec, box, x,
1875                                f, vir_force, mdatoms, enerd, lambda, t);
1876     }
1877
1878     /* Add the forces from enforced rotation potentials (if any) */
1879     if (inputrec->bRot)
1880     {
1881         wallcycle_start(wcycle, ewcROTadd);
1882         enerd->term[F_COM_PULL] += add_rot_forces(inputrec->rot, f, cr, step, t);
1883         wallcycle_stop(wcycle, ewcROTadd);
1884     }
1885
1886     if (PAR(cr) && !(cr->duty & DUTY_PME))
1887     {
1888         /* In case of node-splitting, the PP nodes receive the long-range
1889          * forces, virial and energy from the PME nodes here.
1890          */
1891         pme_receive_force_ener(fplog, bSepDVDL, cr, wcycle, enerd, fr);
1892     }
1893
1894     if (bDoForces)
1895     {
1896         post_process_forces(cr, step, nrnb, wcycle,
1897                             top, box, x, f, vir_force, mdatoms, graph, fr, vsite,
1898                             flags);
1899     }
1900
1901     /* Sum the potential energy terms from group contributions */
1902     sum_epot(&(enerd->grpp), enerd->term);
1903 }
1904
1905 void do_force(FILE *fplog, t_commrec *cr,
1906               t_inputrec *inputrec,
1907               gmx_int64_t step, t_nrnb *nrnb, gmx_wallcycle_t wcycle,
1908               gmx_localtop_t *top,
1909               gmx_groups_t *groups,
1910               matrix box, rvec x[], history_t *hist,
1911               rvec f[],
1912               tensor vir_force,
1913               t_mdatoms *mdatoms,
1914               gmx_enerdata_t *enerd, t_fcdata *fcd,
1915               real *lambda, t_graph *graph,
1916               t_forcerec *fr,
1917               gmx_vsite_t *vsite, rvec mu_tot,
1918               double t, FILE *field, gmx_edsam_t ed,
1919               gmx_bool bBornRadii,
1920               int flags)
1921 {
1922     /* modify force flag if not doing nonbonded */
1923     if (!fr->bNonbonded)
1924     {
1925         flags &= ~GMX_FORCE_NONBONDED;
1926     }
1927
1928     switch (inputrec->cutoff_scheme)
1929     {
1930         case ecutsVERLET:
1931             do_force_cutsVERLET(fplog, cr, inputrec,
1932                                 step, nrnb, wcycle,
1933                                 top,
1934                                 groups,
1935                                 box, x, hist,
1936                                 f, vir_force,
1937                                 mdatoms,
1938                                 enerd, fcd,
1939                                 lambda, graph,
1940                                 fr, fr->ic,
1941                                 vsite, mu_tot,
1942                                 t, field, ed,
1943                                 bBornRadii,
1944                                 flags);
1945             break;
1946         case ecutsGROUP:
1947             do_force_cutsGROUP(fplog, cr, inputrec,
1948                                step, nrnb, wcycle,
1949                                top,
1950                                groups,
1951                                box, x, hist,
1952                                f, vir_force,
1953                                mdatoms,
1954                                enerd, fcd,
1955                                lambda, graph,
1956                                fr, vsite, mu_tot,
1957                                t, field, ed,
1958                                bBornRadii,
1959                                flags);
1960             break;
1961         default:
1962             gmx_incons("Invalid cut-off scheme passed!");
1963     }
1964 }
1965
1966
1967 void do_constrain_first(FILE *fplog, gmx_constr_t constr,
1968                         t_inputrec *ir, t_mdatoms *md,
1969                         t_state *state, t_commrec *cr, t_nrnb *nrnb,
1970                         t_forcerec *fr, gmx_localtop_t *top)
1971 {
1972     int             i, m, start, end;
1973     gmx_int64_t     step;
1974     real            dt = ir->delta_t;
1975     real            dvdl_dum;
1976     rvec           *savex;
1977
1978     snew(savex, state->natoms);
1979
1980     start = 0;
1981     end   = md->homenr;
1982
1983     if (debug)
1984     {
1985         fprintf(debug, "vcm: start=%d, homenr=%d, end=%d\n",
1986                 start, md->homenr, end);
1987     }
1988     /* Do a first constrain to reset particles... */
1989     step = ir->init_step;
1990     if (fplog)
1991     {
1992         char buf[STEPSTRSIZE];
1993         fprintf(fplog, "\nConstraining the starting coordinates (step %s)\n",
1994                 gmx_step_str(step, buf));
1995     }
1996     dvdl_dum = 0;
1997
1998     /* constrain the current position */
1999     constrain(NULL, TRUE, FALSE, constr, &(top->idef),
2000               ir, NULL, cr, step, 0, md,
2001               state->x, state->x, NULL,
2002               fr->bMolPBC, state->box,
2003               state->lambda[efptBONDED], &dvdl_dum,
2004               NULL, NULL, nrnb, econqCoord,
2005               ir->epc == epcMTTK, state->veta, state->veta);
2006     if (EI_VV(ir->eI))
2007     {
2008         /* constrain the inital velocity, and save it */
2009         /* also may be useful if we need the ekin from the halfstep for velocity verlet */
2010         /* might not yet treat veta correctly */
2011         constrain(NULL, TRUE, FALSE, constr, &(top->idef),
2012                   ir, NULL, cr, step, 0, md,
2013                   state->x, state->v, state->v,
2014                   fr->bMolPBC, state->box,
2015                   state->lambda[efptBONDED], &dvdl_dum,
2016                   NULL, NULL, nrnb, econqVeloc,
2017                   ir->epc == epcMTTK, state->veta, state->veta);
2018     }
2019     /* constrain the inital velocities at t-dt/2 */
2020     if (EI_STATE_VELOCITY(ir->eI) && ir->eI != eiVV)
2021     {
2022         for (i = start; (i < end); i++)
2023         {
2024             for (m = 0; (m < DIM); m++)
2025             {
2026                 /* Reverse the velocity */
2027                 state->v[i][m] = -state->v[i][m];
2028                 /* Store the position at t-dt in buf */
2029                 savex[i][m] = state->x[i][m] + dt*state->v[i][m];
2030             }
2031         }
2032         /* Shake the positions at t=-dt with the positions at t=0
2033          * as reference coordinates.
2034          */
2035         if (fplog)
2036         {
2037             char buf[STEPSTRSIZE];
2038             fprintf(fplog, "\nConstraining the coordinates at t0-dt (step %s)\n",
2039                     gmx_step_str(step, buf));
2040         }
2041         dvdl_dum = 0;
2042         constrain(NULL, TRUE, FALSE, constr, &(top->idef),
2043                   ir, NULL, cr, step, -1, md,
2044                   state->x, savex, NULL,
2045                   fr->bMolPBC, state->box,
2046                   state->lambda[efptBONDED], &dvdl_dum,
2047                   state->v, NULL, nrnb, econqCoord,
2048                   ir->epc == epcMTTK, state->veta, state->veta);
2049
2050         for (i = start; i < end; i++)
2051         {
2052             for (m = 0; m < DIM; m++)
2053             {
2054                 /* Re-reverse the velocities */
2055                 state->v[i][m] = -state->v[i][m];
2056             }
2057         }
2058     }
2059     sfree(savex);
2060 }
2061
2062
2063 static void
2064 integrate_table(real vdwtab[], real scale, int offstart, int rstart, int rend,
2065                 double *enerout, double *virout)
2066 {
2067     double enersum, virsum;
2068     double invscale, invscale2, invscale3;
2069     double r, ea, eb, ec, pa, pb, pc, pd;
2070     double y0, f, g, h;
2071     int    ri, offset, tabfactor;
2072
2073     invscale  = 1.0/scale;
2074     invscale2 = invscale*invscale;
2075     invscale3 = invscale*invscale2;
2076
2077     /* Following summation derived from cubic spline definition,
2078      * Numerical Recipies in C, second edition, p. 113-116.  Exact for
2079      * the cubic spline.  We first calculate the negative of the
2080      * energy from rvdw to rvdw_switch, assuming that g(r)=1, and then
2081      * add the more standard, abrupt cutoff correction to that result,
2082      * yielding the long-range correction for a switched function.  We
2083      * perform both the pressure and energy loops at the same time for
2084      * simplicity, as the computational cost is low. */
2085
2086     if (offstart == 0)
2087     {
2088         /* Since the dispersion table has been scaled down a factor
2089          * 6.0 and the repulsion a factor 12.0 to compensate for the
2090          * c6/c12 parameters inside nbfp[] being scaled up (to save
2091          * flops in kernels), we need to correct for this.
2092          */
2093         tabfactor = 6.0;
2094     }
2095     else
2096     {
2097         tabfactor = 12.0;
2098     }
2099
2100     enersum = 0.0;
2101     virsum  = 0.0;
2102     for (ri = rstart; ri < rend; ++ri)
2103     {
2104         r  = ri*invscale;
2105         ea = invscale3;
2106         eb = 2.0*invscale2*r;
2107         ec = invscale*r*r;
2108
2109         pa = invscale3;
2110         pb = 3.0*invscale2*r;
2111         pc = 3.0*invscale*r*r;
2112         pd = r*r*r;
2113
2114         /* this "8" is from the packing in the vdwtab array - perhaps
2115            should be defined? */
2116
2117         offset = 8*ri + offstart;
2118         y0     = vdwtab[offset];
2119         f      = vdwtab[offset+1];
2120         g      = vdwtab[offset+2];
2121         h      = vdwtab[offset+3];
2122
2123         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);
2124         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);
2125     }
2126     *enerout = 4.0*M_PI*enersum*tabfactor;
2127     *virout  = 4.0*M_PI*virsum*tabfactor;
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             fr->vdw_modifier == eintmodPOTSWITCH ||
2154             fr->vdw_modifier == eintmodFORCESWITCH)
2155         {
2156             if (fr->rvdw_switch == 0)
2157             {
2158                 gmx_fatal(FARGS,
2159                           "With dispersion correction rvdw-switch can not be zero "
2160                           "for vdw-type = %s", evdw_names[fr->vdwtype]);
2161             }
2162
2163             scale  = fr->nblists[0].table_elec_vdw.scale;
2164             vdwtab = fr->nblists[0].table_vdw.data;
2165
2166             /* Round the cut-offs to exact table values for precision */
2167             ri0  = floor(fr->rvdw_switch*scale);
2168             ri1  = ceil(fr->rvdw*scale);
2169             r0   = ri0/scale;
2170             r1   = ri1/scale;
2171             rc3  = r0*r0*r0;
2172             rc9  = rc3*rc3*rc3;
2173
2174             if (fr->vdwtype == evdwSHIFT ||
2175                 fr->vdw_modifier == eintmodFORCESWITCH)
2176             {
2177                 /* Determine the constant energy shift below rvdw_switch.
2178                  * Table has a scale factor since we have scaled it down to compensate
2179                  * for scaling-up c6/c12 with the derivative factors to save flops in analytical kernels.
2180                  */
2181                 fr->enershiftsix    = (real)(-1.0/(rc3*rc3)) - 6.0*vdwtab[8*ri0];
2182                 fr->enershifttwelve = (real)( 1.0/(rc9*rc3)) - 12.0*vdwtab[8*ri0 + 4];
2183             }
2184             /* Add the constant part from 0 to rvdw_switch.
2185              * This integration from 0 to rvdw_switch overcounts the number
2186              * of interactions by 1, as it also counts the self interaction.
2187              * We will correct for this later.
2188              */
2189             eners[0] += 4.0*M_PI*fr->enershiftsix*rc3/3.0;
2190             eners[1] += 4.0*M_PI*fr->enershifttwelve*rc3/3.0;
2191             for (i = 0; i < 2; i++)
2192             {
2193                 enersum = 0;
2194                 virsum  = 0;
2195                 integrate_table(vdwtab, scale, (i == 0 ? 0 : 4), ri0, ri1, &enersum, &virsum);
2196                 eners[i] -= enersum;
2197                 virs[i]  -= virsum;
2198             }
2199
2200             /* now add the correction for rvdw_switch to infinity */
2201             eners[0] += -4.0*M_PI/(3.0*rc3);
2202             eners[1] +=  4.0*M_PI/(9.0*rc9);
2203             virs[0]  +=  8.0*M_PI/rc3;
2204             virs[1]  += -16.0*M_PI/(3.0*rc9);
2205         }
2206         else if (fr->vdwtype == evdwCUT ||
2207                  EVDW_PME(fr->vdwtype) ||
2208                  fr->vdwtype == evdwUSER)
2209         {
2210             if (fr->vdwtype == evdwUSER && fplog)
2211             {
2212                 fprintf(fplog,
2213                         "WARNING: using dispersion correction with user tables\n");
2214             }
2215
2216             /* Note that with LJ-PME, the dispersion correction is multiplied
2217              * by the difference between the actual C6 and the value of C6
2218              * that would produce the combination rule.
2219              * This means the normal energy and virial difference formulas
2220              * can be used here.
2221              */
2222
2223             rc3  = fr->rvdw*fr->rvdw*fr->rvdw;
2224             rc9  = rc3*rc3*rc3;
2225             /* Contribution beyond the cut-off */
2226             eners[0] += -4.0*M_PI/(3.0*rc3);
2227             eners[1] +=  4.0*M_PI/(9.0*rc9);
2228             if (fr->vdw_modifier == eintmodPOTSHIFT)
2229             {
2230                 /* Contribution within the cut-off */
2231                 eners[0] += -4.0*M_PI/(3.0*rc3);
2232                 eners[1] +=  4.0*M_PI/(3.0*rc9);
2233             }
2234             /* Contribution beyond the cut-off */
2235             virs[0]  +=  8.0*M_PI/rc3;
2236             virs[1]  += -16.0*M_PI/(3.0*rc9);
2237         }
2238         else
2239         {
2240             gmx_fatal(FARGS,
2241                       "Dispersion correction is not implemented for vdw-type = %s",
2242                       evdw_names[fr->vdwtype]);
2243         }
2244
2245         /* TODO: remove this code once we have group LJ-PME kernels
2246          * that calculate the exact, full LJ param C6/r^6 within the cut-off,
2247          * as the current nbnxn kernels do.
2248          */
2249         if (fr->vdwtype == evdwPME && fr->cutoff_scheme == ecutsGROUP)
2250         {
2251             /* Calculate self-interaction coefficient (assuming that
2252              * the reciprocal-space contribution is constant in the
2253              * region that contributes to the self-interaction).
2254              */
2255             fr->enershiftsix = pow(fr->ewaldcoeff_lj, 6) / 6.0;
2256
2257             eners[0] += -pow(sqrt(M_PI)*fr->ewaldcoeff_lj, 3)/3.0;
2258             virs[0]  +=  pow(sqrt(M_PI)*fr->ewaldcoeff_lj, 3);
2259         }
2260
2261         fr->enerdiffsix    = eners[0];
2262         fr->enerdifftwelve = eners[1];
2263         /* The 0.5 is due to the Gromacs definition of the virial */
2264         fr->virdiffsix     = 0.5*virs[0];
2265         fr->virdifftwelve  = 0.5*virs[1];
2266     }
2267 }
2268
2269 void calc_dispcorr(FILE *fplog, t_inputrec *ir, t_forcerec *fr,
2270                    gmx_int64_t step, int natoms,
2271                    matrix box, real lambda, tensor pres, tensor virial,
2272                    real *prescorr, real *enercorr, real *dvdlcorr)
2273 {
2274     gmx_bool bCorrAll, bCorrPres;
2275     real     dvdlambda, invvol, dens, ninter, avcsix, avctwelve, enerdiff, svir = 0, spres = 0;
2276     int      m;
2277
2278     *prescorr = 0;
2279     *enercorr = 0;
2280     *dvdlcorr = 0;
2281
2282     clear_mat(virial);
2283     clear_mat(pres);
2284
2285     if (ir->eDispCorr != edispcNO)
2286     {
2287         bCorrAll  = (ir->eDispCorr == edispcAllEner ||
2288                      ir->eDispCorr == edispcAllEnerPres);
2289         bCorrPres = (ir->eDispCorr == edispcEnerPres ||
2290                      ir->eDispCorr == edispcAllEnerPres);
2291
2292         invvol = 1/det(box);
2293         if (fr->n_tpi)
2294         {
2295             /* Only correct for the interactions with the inserted molecule */
2296             dens   = (natoms - fr->n_tpi)*invvol;
2297             ninter = fr->n_tpi;
2298         }
2299         else
2300         {
2301             dens   = natoms*invvol;
2302             ninter = 0.5*natoms;
2303         }
2304
2305         if (ir->efep == efepNO)
2306         {
2307             avcsix    = fr->avcsix[0];
2308             avctwelve = fr->avctwelve[0];
2309         }
2310         else
2311         {
2312             avcsix    = (1 - lambda)*fr->avcsix[0]    + lambda*fr->avcsix[1];
2313             avctwelve = (1 - lambda)*fr->avctwelve[0] + lambda*fr->avctwelve[1];
2314         }
2315
2316         enerdiff   = ninter*(dens*fr->enerdiffsix - fr->enershiftsix);
2317         *enercorr += avcsix*enerdiff;
2318         dvdlambda  = 0.0;
2319         if (ir->efep != efepNO)
2320         {
2321             dvdlambda += (fr->avcsix[1] - fr->avcsix[0])*enerdiff;
2322         }
2323         if (bCorrAll)
2324         {
2325             enerdiff   = ninter*(dens*fr->enerdifftwelve - fr->enershifttwelve);
2326             *enercorr += avctwelve*enerdiff;
2327             if (fr->efep != efepNO)
2328             {
2329                 dvdlambda += (fr->avctwelve[1] - fr->avctwelve[0])*enerdiff;
2330             }
2331         }
2332
2333         if (bCorrPres)
2334         {
2335             svir = ninter*dens*avcsix*fr->virdiffsix/3.0;
2336             if (ir->eDispCorr == edispcAllEnerPres)
2337             {
2338                 svir += ninter*dens*avctwelve*fr->virdifftwelve/3.0;
2339             }
2340             /* The factor 2 is because of the Gromacs virial definition */
2341             spres = -2.0*invvol*svir*PRESFAC;
2342
2343             for (m = 0; m < DIM; m++)
2344             {
2345                 virial[m][m] += svir;
2346                 pres[m][m]   += spres;
2347             }
2348             *prescorr += spres;
2349         }
2350
2351         /* Can't currently control when it prints, for now, just print when degugging */
2352         if (debug)
2353         {
2354             if (bCorrAll)
2355             {
2356                 fprintf(debug, "Long Range LJ corr.: <C6> %10.4e, <C12> %10.4e\n",
2357                         avcsix, avctwelve);
2358             }
2359             if (bCorrPres)
2360             {
2361                 fprintf(debug,
2362                         "Long Range LJ corr.: Epot %10g, Pres: %10g, Vir: %10g\n",
2363                         *enercorr, spres, svir);
2364             }
2365             else
2366             {
2367                 fprintf(debug, "Long Range LJ corr.: Epot %10g\n", *enercorr);
2368             }
2369         }
2370
2371         if (fr->bSepDVDL && do_per_step(step, ir->nstlog))
2372         {
2373             gmx_print_sepdvdl(fplog, "Dispersion correction", *enercorr, dvdlambda);
2374         }
2375         if (fr->efep != efepNO)
2376         {
2377             *dvdlcorr += dvdlambda;
2378         }
2379     }
2380 }
2381
2382 void do_pbc_first(FILE *fplog, matrix box, t_forcerec *fr,
2383                   t_graph *graph, rvec x[])
2384 {
2385     if (fplog)
2386     {
2387         fprintf(fplog, "Removing pbc first time\n");
2388     }
2389     calc_shifts(box, fr->shift_vec);
2390     if (graph)
2391     {
2392         mk_mshift(fplog, graph, fr->ePBC, box, x);
2393         if (gmx_debug_at)
2394         {
2395             p_graph(debug, "do_pbc_first 1", graph);
2396         }
2397         shift_self(graph, box, x);
2398         /* By doing an extra mk_mshift the molecules that are broken
2399          * because they were e.g. imported from another software
2400          * will be made whole again. Such are the healing powers
2401          * of GROMACS.
2402          */
2403         mk_mshift(fplog, graph, fr->ePBC, box, x);
2404         if (gmx_debug_at)
2405         {
2406             p_graph(debug, "do_pbc_first 2", graph);
2407         }
2408     }
2409     if (fplog)
2410     {
2411         fprintf(fplog, "Done rmpbc\n");
2412     }
2413 }
2414
2415 static void low_do_pbc_mtop(FILE *fplog, int ePBC, matrix box,
2416                             gmx_mtop_t *mtop, rvec x[],
2417                             gmx_bool bFirst)
2418 {
2419     t_graph        *graph;
2420     int             mb, as, mol;
2421     gmx_molblock_t *molb;
2422
2423     if (bFirst && fplog)
2424     {
2425         fprintf(fplog, "Removing pbc first time\n");
2426     }
2427
2428     snew(graph, 1);
2429     as = 0;
2430     for (mb = 0; mb < mtop->nmolblock; mb++)
2431     {
2432         molb = &mtop->molblock[mb];
2433         if (molb->natoms_mol == 1 ||
2434             (!bFirst && mtop->moltype[molb->type].cgs.nr == 1))
2435         {
2436             /* Just one atom or charge group in the molecule, no PBC required */
2437             as += molb->nmol*molb->natoms_mol;
2438         }
2439         else
2440         {
2441             /* Pass NULL iso fplog to avoid graph prints for each molecule type */
2442             mk_graph_ilist(NULL, mtop->moltype[molb->type].ilist,
2443                            0, molb->natoms_mol, FALSE, FALSE, graph);
2444
2445             for (mol = 0; mol < molb->nmol; mol++)
2446             {
2447                 mk_mshift(fplog, graph, ePBC, box, x+as);
2448
2449                 shift_self(graph, box, x+as);
2450                 /* The molecule is whole now.
2451                  * We don't need the second mk_mshift call as in do_pbc_first,
2452                  * since we no longer need this graph.
2453                  */
2454
2455                 as += molb->natoms_mol;
2456             }
2457             done_graph(graph);
2458         }
2459     }
2460     sfree(graph);
2461 }
2462
2463 void do_pbc_first_mtop(FILE *fplog, int ePBC, matrix box,
2464                        gmx_mtop_t *mtop, rvec x[])
2465 {
2466     low_do_pbc_mtop(fplog, ePBC, box, mtop, x, TRUE);
2467 }
2468
2469 void do_pbc_mtop(FILE *fplog, int ePBC, matrix box,
2470                  gmx_mtop_t *mtop, rvec x[])
2471 {
2472     low_do_pbc_mtop(fplog, ePBC, box, mtop, x, FALSE);
2473 }
2474
2475 void finish_run(FILE *fplog, t_commrec *cr,
2476                 t_inputrec *inputrec,
2477                 t_nrnb nrnb[], gmx_wallcycle_t wcycle,
2478                 gmx_walltime_accounting_t walltime_accounting,
2479                 wallclock_gpu_t *gputimes,
2480                 gmx_bool bWriteStat)
2481 {
2482     int     i, j;
2483     t_nrnb *nrnb_tot = NULL;
2484     real    delta_t;
2485     double  nbfs, mflop;
2486     double  elapsed_time,
2487             elapsed_time_over_all_ranks,
2488             elapsed_time_over_all_threads,
2489             elapsed_time_over_all_threads_over_all_ranks;
2490     wallcycle_sum(cr, wcycle);
2491
2492     if (cr->nnodes > 1)
2493     {
2494         snew(nrnb_tot, 1);
2495 #ifdef GMX_MPI
2496         MPI_Allreduce(nrnb->n, nrnb_tot->n, eNRNB, MPI_DOUBLE, MPI_SUM,
2497                       cr->mpi_comm_mysim);
2498 #endif
2499     }
2500     else
2501     {
2502         nrnb_tot = nrnb;
2503     }
2504
2505     elapsed_time                                 = walltime_accounting_get_elapsed_time(walltime_accounting);
2506     elapsed_time_over_all_ranks                  = elapsed_time;
2507     elapsed_time_over_all_threads                = walltime_accounting_get_elapsed_time_over_all_threads(walltime_accounting);
2508     elapsed_time_over_all_threads_over_all_ranks = elapsed_time_over_all_threads;
2509 #ifdef GMX_MPI
2510     if (cr->nnodes > 1)
2511     {
2512         /* reduce elapsed_time over all MPI ranks in the current simulation */
2513         MPI_Allreduce(&elapsed_time,
2514                       &elapsed_time_over_all_ranks,
2515                       1, MPI_DOUBLE, MPI_SUM,
2516                       cr->mpi_comm_mysim);
2517         elapsed_time_over_all_ranks /= cr->nnodes;
2518         /* Reduce elapsed_time_over_all_threads over all MPI ranks in the
2519          * current simulation. */
2520         MPI_Allreduce(&elapsed_time_over_all_threads,
2521                       &elapsed_time_over_all_threads_over_all_ranks,
2522                       1, MPI_DOUBLE, MPI_SUM,
2523                       cr->mpi_comm_mysim);
2524     }
2525 #endif
2526
2527     if (SIMMASTER(cr))
2528     {
2529         print_flop(fplog, nrnb_tot, &nbfs, &mflop);
2530     }
2531     if (cr->nnodes > 1)
2532     {
2533         sfree(nrnb_tot);
2534     }
2535
2536     if ((cr->duty & DUTY_PP) && DOMAINDECOMP(cr))
2537     {
2538         print_dd_statistics(cr, inputrec, fplog);
2539     }
2540
2541     if (SIMMASTER(cr))
2542     {
2543         wallcycle_print(fplog, cr->nnodes, cr->npmenodes,
2544                         elapsed_time_over_all_ranks,
2545                         wcycle, gputimes);
2546
2547         if (EI_DYNAMICS(inputrec->eI))
2548         {
2549             delta_t = inputrec->delta_t;
2550         }
2551         else
2552         {
2553             delta_t = 0;
2554         }
2555
2556         if (fplog)
2557         {
2558             print_perf(fplog, elapsed_time_over_all_threads_over_all_ranks,
2559                        elapsed_time_over_all_ranks,
2560                        walltime_accounting_get_nsteps_done(walltime_accounting),
2561                        delta_t, nbfs, mflop);
2562         }
2563         if (bWriteStat)
2564         {
2565             print_perf(stderr, elapsed_time_over_all_threads_over_all_ranks,
2566                        elapsed_time_over_all_ranks,
2567                        walltime_accounting_get_nsteps_done(walltime_accounting),
2568                        delta_t, nbfs, mflop);
2569         }
2570     }
2571 }
2572
2573 extern void initialize_lambdas(FILE *fplog, t_inputrec *ir, int *fep_state, real *lambda, double *lam0)
2574 {
2575     /* this function works, but could probably use a logic rewrite to keep all the different
2576        types of efep straight. */
2577
2578     int       i;
2579     t_lambda *fep = ir->fepvals;
2580
2581     if ((ir->efep == efepNO) && (ir->bSimTemp == FALSE))
2582     {
2583         for (i = 0; i < efptNR; i++)
2584         {
2585             lambda[i] = 0.0;
2586             if (lam0)
2587             {
2588                 lam0[i] = 0.0;
2589             }
2590         }
2591         return;
2592     }
2593     else
2594     {
2595         *fep_state = fep->init_fep_state; /* this might overwrite the checkpoint
2596                                              if checkpoint is set -- a kludge is in for now
2597                                              to prevent this.*/
2598         for (i = 0; i < efptNR; i++)
2599         {
2600             /* overwrite lambda state with init_lambda for now for backwards compatibility */
2601             if (fep->init_lambda >= 0) /* if it's -1, it was never initializd */
2602             {
2603                 lambda[i] = fep->init_lambda;
2604                 if (lam0)
2605                 {
2606                     lam0[i] = lambda[i];
2607                 }
2608             }
2609             else
2610             {
2611                 lambda[i] = fep->all_lambda[i][*fep_state];
2612                 if (lam0)
2613                 {
2614                     lam0[i] = lambda[i];
2615                 }
2616             }
2617         }
2618         if (ir->bSimTemp)
2619         {
2620             /* need to rescale control temperatures to match current state */
2621             for (i = 0; i < ir->opts.ngtc; i++)
2622             {
2623                 if (ir->opts.ref_t[i] > 0)
2624                 {
2625                     ir->opts.ref_t[i] = ir->simtempvals->temperatures[*fep_state];
2626                 }
2627             }
2628         }
2629     }
2630
2631     /* Send to the log the information on the current lambdas */
2632     if (fplog != NULL)
2633     {
2634         fprintf(fplog, "Initial vector of lambda components:[ ");
2635         for (i = 0; i < efptNR; i++)
2636         {
2637             fprintf(fplog, "%10.4f ", lambda[i]);
2638         }
2639         fprintf(fplog, "]\n");
2640     }
2641     return;
2642 }
2643
2644
2645 void init_md(FILE *fplog,
2646              t_commrec *cr, t_inputrec *ir, const output_env_t oenv,
2647              double *t, double *t0,
2648              real *lambda, int *fep_state, double *lam0,
2649              t_nrnb *nrnb, gmx_mtop_t *mtop,
2650              gmx_update_t *upd,
2651              int nfile, const t_filenm fnm[],
2652              gmx_mdoutf_t *outf, t_mdebin **mdebin,
2653              tensor force_vir, tensor shake_vir, rvec mu_tot,
2654              gmx_bool *bSimAnn, t_vcm **vcm, unsigned long Flags)
2655 {
2656     int  i, j, n;
2657     real tmpt, mod;
2658
2659     /* Initial values */
2660     *t = *t0       = ir->init_t;
2661
2662     *bSimAnn = FALSE;
2663     for (i = 0; i < ir->opts.ngtc; i++)
2664     {
2665         /* set bSimAnn if any group is being annealed */
2666         if (ir->opts.annealing[i] != eannNO)
2667         {
2668             *bSimAnn = TRUE;
2669         }
2670     }
2671     if (*bSimAnn)
2672     {
2673         update_annealing_target_temp(&(ir->opts), ir->init_t);
2674     }
2675
2676     /* Initialize lambda variables */
2677     initialize_lambdas(fplog, ir, fep_state, lambda, lam0);
2678
2679     if (upd)
2680     {
2681         *upd = init_update(ir);
2682     }
2683
2684
2685     if (vcm != NULL)
2686     {
2687         *vcm = init_vcm(fplog, &mtop->groups, ir);
2688     }
2689
2690     if (EI_DYNAMICS(ir->eI) && !(Flags & MD_APPENDFILES))
2691     {
2692         if (ir->etc == etcBERENDSEN)
2693         {
2694             please_cite(fplog, "Berendsen84a");
2695         }
2696         if (ir->etc == etcVRESCALE)
2697         {
2698             please_cite(fplog, "Bussi2007a");
2699         }
2700     }
2701
2702     init_nrnb(nrnb);
2703
2704     if (nfile != -1)
2705     {
2706         *outf = init_mdoutf(fplog, nfile, fnm, Flags, cr, ir, mtop, oenv);
2707
2708         *mdebin = init_mdebin((Flags & MD_APPENDFILES) ? NULL : mdoutf_get_fp_ene(*outf),
2709                               mtop, ir, mdoutf_get_fp_dhdl(*outf));
2710     }
2711
2712     if (ir->bAdress)
2713     {
2714         please_cite(fplog, "Fritsch12");
2715         please_cite(fplog, "Junghans10");
2716     }
2717     /* Initiate variables */
2718     clear_mat(force_vir);
2719     clear_mat(shake_vir);
2720     clear_rvec(mu_tot);
2721
2722     debug_gmx();
2723 }