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 <assert.h>
  42 #include <math.h>
  43 #include <stdio.h>
  44 #include <string.h>
  45 #ifdef HAVE_SYS_TIME_H
  46 #include <sys/time.h>
  47 #endif
  48
  49 #include "typedefs.h"
  50 #include "gromacs/utility/cstringutil.h"
  51 #include "gromacs/utility/smalloc.h"
  52 #include "names.h"
  53 #include "txtdump.h"
  54 #include "pbc.h"
  55 #include "chargegroup.h"
  56 #include "vec.h"
  57 #include "nrnb.h"
  58 #include "mshift.h"
  59 #include "mdrun.h"
  60 #include "sim_util.h"
  61 #include "update.h"
  62 #include "physics.h"
  63 #include "main.h"
  64 #include "mdatoms.h"
  65 #include "force.h"
  66 #include "bondf.h"
  67 #include "pme.h"
  68 #include "disre.h"
  69 #include "orires.h"
  70 #include "network.h"
  71 #include "calcmu.h"
  72 #include "constr.h"
  73 #include "xvgr.h"
  74 #include "copyrite.h"
  75 #include "domdec.h"
  76 #include "genborn.h"
  77 #include "nbnxn_atomdata.h"
  78 #include "nbnxn_search.h"
  79 #include "nbnxn_kernels/nbnxn_kernel_ref.h"
  80 #include "nbnxn_kernels/simd_4xn/nbnxn_kernel_simd_4xn.h"
  81 #include "nbnxn_kernels/simd_2xnn/nbnxn_kernel_simd_2xnn.h"
  82 #include "nbnxn_kernels/nbnxn_kernel_gpu_ref.h"
  83 #include "nonbonded.h"
  84 #include "../gmxlib/nonbonded/nb_kernel.h"
  85 #include "../gmxlib/nonbonded/nb_free_energy.h"
  86
  87 #include "gromacs/timing/wallcycle.h"
  88 #include "gromacs/timing/walltime_accounting.h"
  89 #include "gromacs/utility/gmxmpi.h"
  90 #include "gromacs/essentialdynamics/edsam.h"
  91 #include "gromacs/pulling/pull.h"
  92 #include "gromacs/pulling/pull_rotation.h"
  93 #include "gromacs/imd/imd.h"
  94 #include "adress.h"
  95 #include "qmmm.h"
  96
  97 #include "gmx_omp_nthreads.h"
  98
  99 #include "nbnxn_cuda_data_mgmt.h"
 100 #include "nbnxn_cuda/nbnxn_cuda.h"
 101
 102 void print_time(FILE                     *out,
 103                 gmx_walltime_accounting_t walltime_accounting,
 104                 gmx_int64_t               step,
 105                 t_inputrec               *ir,
 106                 t_commrec gmx_unused     *cr)
 107 {
 108     time_t finish;
 109     char   timebuf[STRLEN];
 110     double dt, elapsed_seconds, time_per_step;
 111     char   buf[48];
 112
 113 #ifndef GMX_THREAD_MPI
 114     if (!PAR(cr))
 115 #endif
 116     {
 117         fprintf(out, "\r");
 118     }
 119     fprintf(out, "step %s", gmx_step_str(step, buf));
 120     if ((step >= ir->nstlist))
 121     {
 122         double seconds_since_epoch = gmx_gettime();
 123         elapsed_seconds = seconds_since_epoch - walltime_accounting_get_start_time_stamp(walltime_accounting);
 124         time_per_step   = elapsed_seconds/(step - ir->init_step + 1);
 125         dt              = (ir->nsteps + ir->init_step - step) * time_per_step;
 126
 127         if (ir->nsteps >= 0)
 128         {
 129             if (dt >= 300)
 130             {
 131                 finish = (time_t) (seconds_since_epoch + dt);
 132                 gmx_ctime_r(&finish, timebuf, STRLEN);
 133                 sprintf(buf, "%s", timebuf);
 134                 buf[strlen(buf)-1] = '\0';
 135                 fprintf(out, ", will finish %s", buf);
 136             }
 137             else
 138             {
 139                 fprintf(out, ", remaining wall clock time: %5d s          ", (int)dt);
 140             }
 141         }
 142         else
 143         {
 144             fprintf(out, " performance: %.1f ns/day    ",
 145                     ir->delta_t/1000*24*60*60/time_per_step);
 146         }
 147     }
 148 #ifndef GMX_THREAD_MPI
 149     if (PAR(cr))
 150     {
 151         fprintf(out, "\n");
 152     }
 153 #endif
 154
 155     fflush(out);
 156 }
 157
 158 void print_date_and_time(FILE *fplog, int nodeid, const char *title,
 159                          double the_time)
 160 {
 161     char   time_string[STRLEN];
 162
 163     if (!fplog)
 164     {
 165         return;
 166     }
 167
 168     {
 169         int    i;
 170         char   timebuf[STRLEN];
 171         time_t temp_time = (time_t) the_time;
 172
 173         gmx_ctime_r(&temp_time, timebuf, STRLEN);
 174         for (i = 0; timebuf[i] >= ' '; i++)
 175         {
 176             time_string[i] = timebuf[i];
 177         }
 178         time_string[i] = '\0';
 179     }
 180
 181     fprintf(fplog, "%s on node %d %s\n", title, nodeid, time_string);
 182 }
 183
 184 void print_start(FILE *fplog, t_commrec *cr,
 185                  gmx_walltime_accounting_t walltime_accounting,
 186                  const char *name)
 187 {
 188     char buf[STRLEN];
 189
 190     sprintf(buf, "Started %s", name);
 191     print_date_and_time(fplog, cr->nodeid, buf,
 192                         walltime_accounting_get_start_time_stamp(walltime_accounting));
 193 }
 194
 195 static void sum_forces(int start, int end, rvec f[], rvec flr[])
 196 {
 197     int i;
 198
 199     if (gmx_debug_at)
 200     {
 201         pr_rvecs(debug, 0, "fsr", f+start, end-start);
 202         pr_rvecs(debug, 0, "flr", flr+start, end-start);
 203     }
 204     for (i = start; (i < end); i++)
 205     {
 206         rvec_inc(f[i], flr[i]);
 207     }
 208 }
 209
 210 /*
 211  * calc_f_el calculates forces due to an electric field.
 212  *
 213  * force is kJ mol^-1 nm^-1 = e * kJ mol^-1 nm^-1 / e
 214  *
 215  * Et[] contains the parameters for the time dependent
 216  * part of the field (not yet used).
 217  * Ex[] contains the parameters for
 218  * the spatial dependent part of the field. You can have cool periodic
 219  * fields in principle, but only a constant field is supported
 220  * now.
 221  * The function should return the energy due to the electric field
 222  * (if any) but for now returns 0.
 223  *
 224  * WARNING:
 225  * There can be problems with the virial.
 226  * Since the field is not self-consistent this is unavoidable.
 227  * For neutral molecules the virial is correct within this approximation.
 228  * For neutral systems with many charged molecules the error is small.
 229  * But for systems with a net charge or a few charged molecules
 230  * the error can be significant when the field is high.
 231  * Solution: implement a self-consitent electric field into PME.
 232  */
 233 static void calc_f_el(FILE *fp, int  start, int homenr,
 234                       real charge[], rvec f[],
 235                       t_cosines Ex[], t_cosines Et[], double t)
 236 {
 237     rvec Ext;
 238     real t0;
 239     int  i, m;
 240
 241     for (m = 0; (m < DIM); m++)
 242     {
 243         if (Et[m].n > 0)
 244         {
 245             if (Et[m].n == 3)
 246             {
 247                 t0     = Et[m].a[1];
 248                 Ext[m] = cos(Et[m].a[0]*(t-t0))*exp(-sqr(t-t0)/(2.0*sqr(Et[m].a[2])));
 249             }
 250             else
 251             {
 252                 Ext[m] = cos(Et[m].a[0]*t);
 253             }
 254         }
 255         else
 256         {
 257             Ext[m] = 1.0;
 258         }
 259         if (Ex[m].n > 0)
 260         {
 261             /* Convert the field strength from V/nm to MD-units */
 262             Ext[m] *= Ex[m].a[0]*FIELDFAC;
 263             for (i = start; (i < start+homenr); i++)
 264             {
 265                 f[i][m] += charge[i]*Ext[m];
 266             }
 267         }
 268         else
 269         {
 270             Ext[m] = 0;
 271         }
 272     }
 273     if (fp != NULL)
 274     {
 275         fprintf(fp, "%10g  %10g  %10g  %10g #FIELD\n", t,
 276                 Ext[XX]/FIELDFAC, Ext[YY]/FIELDFAC, Ext[ZZ]/FIELDFAC);
 277     }
 278 }
 279
 280 static void calc_virial(int start, int homenr, rvec x[], rvec f[],
 281                         tensor vir_part, t_graph *graph, matrix box,
 282                         t_nrnb *nrnb, const t_forcerec *fr, int ePBC)
 283 {
 284     int    i, j;
 285     tensor virtest;
 286
 287     /* The short-range virial from surrounding boxes */
 288     clear_mat(vir_part);
 289     calc_vir(SHIFTS, fr->shift_vec, fr->fshift, vir_part, ePBC == epbcSCREW, box);
 290     inc_nrnb(nrnb, eNR_VIRIAL, SHIFTS);
 291
 292     /* Calculate partial virial, for local atoms only, based on short range.
 293      * Total virial is computed in global_stat, called from do_md
 294      */
 295     f_calc_vir(start, start+homenr, x, f, vir_part, graph, box);
 296     inc_nrnb(nrnb, eNR_VIRIAL, homenr);
 297
 298     /* Add position restraint contribution */
 299     for (i = 0; i < DIM; i++)
 300     {
 301         vir_part[i][i] += fr->vir_diag_posres[i];
 302     }
 303
 304     /* Add wall contribution */
 305     for (i = 0; i < DIM; i++)
 306     {
 307         vir_part[i][ZZ] += fr->vir_wall_z[i];
 308     }
 309
 310     if (debug)
 311     {
 312         pr_rvecs(debug, 0, "vir_part", vir_part, DIM);
 313     }
 314 }
 315
 316 static void posres_wrapper(FILE *fplog,
 317                            int flags,
 318                            gmx_bool bSepDVDL,
 319                            t_inputrec *ir,
 320                            t_nrnb *nrnb,
 321                            gmx_localtop_t *top,
 322                            matrix box, rvec x[],
 323                            gmx_enerdata_t *enerd,
 324                            real *lambda,
 325                            t_forcerec *fr)
 326 {
 327     t_pbc pbc;
 328     real  v, dvdl;
 329     int   i;
 330
 331     /* Position restraints always require full pbc */
 332     set_pbc(&pbc, ir->ePBC, box);
 333     dvdl = 0;
 334     v    = posres(top->idef.il[F_POSRES].nr, top->idef.il[F_POSRES].iatoms,
 335                   top->idef.iparams_posres,
 336                   (const rvec*)x, fr->f_novirsum, fr->vir_diag_posres,
 337                   ir->ePBC == epbcNONE ? NULL : &pbc,
 338                   lambda[efptRESTRAINT], &dvdl,
 339                   fr->rc_scaling, fr->ePBC, fr->posres_com, fr->posres_comB);
 340     if (bSepDVDL)
 341     {
 342         gmx_print_sepdvdl(fplog, interaction_function[F_POSRES].longname, v, dvdl);
 343     }
 344     enerd->term[F_POSRES] += v;
 345     /* If just the force constant changes, the FEP term is linear,
 346      * but if k changes, it is not.
 347      */
 348     enerd->dvdl_nonlin[efptRESTRAINT] += dvdl;
 349     inc_nrnb(nrnb, eNR_POSRES, top->idef.il[F_POSRES].nr/2);
 350
 351     if ((ir->fepvals->n_lambda > 0) && (flags & GMX_FORCE_DHDL))
 352     {
 353         for (i = 0; i < enerd->n_lambda; i++)
 354         {
 355             real dvdl_dum, lambda_dum;
 356
 357             lambda_dum = (i == 0 ? lambda[efptRESTRAINT] : ir->fepvals->all_lambda[efptRESTRAINT][i-1]);
 358             v          = posres(top->idef.il[F_POSRES].nr, top->idef.il[F_POSRES].iatoms,
 359                                 top->idef.iparams_posres,
 360                                 (const rvec*)x, NULL, NULL,
 361                                 ir->ePBC == epbcNONE ? NULL : &pbc, lambda_dum, &dvdl,
 362                                 fr->rc_scaling, fr->ePBC, fr->posres_com, fr->posres_comB);
 363             enerd->enerpart_lambda[i] += v;
 364         }
 365     }
 366 }
 367
 368 static void fbposres_wrapper(t_inputrec *ir,
 369                              t_nrnb *nrnb,
 370                              gmx_localtop_t *top,
 371                              matrix box, rvec x[],
 372                              gmx_enerdata_t *enerd,
 373                              t_forcerec *fr)
 374 {
 375     t_pbc pbc;
 376     real  v;
 377
 378     /* Flat-bottomed position restraints always require full pbc */
 379     set_pbc(&pbc, ir->ePBC, box);
 380     v = fbposres(top->idef.il[F_FBPOSRES].nr, top->idef.il[F_FBPOSRES].iatoms,
 381                  top->idef.iparams_fbposres,
 382                  (const rvec*)x, fr->f_novirsum, fr->vir_diag_posres,
 383                  ir->ePBC == epbcNONE ? NULL : &pbc,
 384                  fr->rc_scaling, fr->ePBC, fr->posres_com);
 385     enerd->term[F_FBPOSRES] += v;
 386     inc_nrnb(nrnb, eNR_FBPOSRES, top->idef.il[F_FBPOSRES].nr/2);
 387 }
 388
 389 static void pull_potential_wrapper(FILE *fplog,
 390                                    gmx_bool bSepDVDL,
 391                                    t_commrec *cr,
 392                                    t_inputrec *ir,
 393                                    matrix box, rvec x[],
 394                                    rvec f[],
 395                                    tensor vir_force,
 396                                    t_mdatoms *mdatoms,
 397                                    gmx_enerdata_t *enerd,
 398                                    real *lambda,
 399                                    double t)
 400 {
 401     t_pbc  pbc;
 402     real   dvdl;
 403
 404     /* Calculate the center of mass forces, this requires communication,
 405      * which is why pull_potential is called close to other communication.
 406      * The virial contribution is calculated directly,
 407      * which is why we call pull_potential after calc_virial.
 408      */
 409     set_pbc(&pbc, ir->ePBC, box);
 410     dvdl                     = 0;
 411     enerd->term[F_COM_PULL] +=
 412         pull_potential(ir->ePull, ir->pull, mdatoms, &pbc,
 413                        cr, t, lambda[efptRESTRAINT], x, f, vir_force, &dvdl);
 414     if (bSepDVDL)
 415     {
 416         gmx_print_sepdvdl(fplog, "Com pull", enerd->term[F_COM_PULL], dvdl);
 417     }
 418     enerd->dvdl_lin[efptRESTRAINT] += dvdl;
 419 }
 420
 421 static void pme_receive_force_ener(FILE           *fplog,
 422                                    gmx_bool        bSepDVDL,
 423                                    t_commrec      *cr,
 424                                    gmx_wallcycle_t wcycle,
 425                                    gmx_enerdata_t *enerd,
 426                                    t_forcerec     *fr)
 427 {
 428     real   e_q, e_lj, v, dvdl_q, dvdl_lj;
 429     float  cycles_ppdpme, cycles_seppme;
 430
 431     cycles_ppdpme = wallcycle_stop(wcycle, ewcPPDURINGPME);
 432     dd_cycles_add(cr->dd, cycles_ppdpme, ddCyclPPduringPME);
 433
 434     /* In case of node-splitting, the PP nodes receive the long-range
 435      * forces, virial and energy from the PME nodes here.
 436      */
 437     wallcycle_start(wcycle, ewcPP_PMEWAITRECVF);
 438     dvdl_q  = 0;
 439     dvdl_lj = 0;
 440     gmx_pme_receive_f(cr, fr->f_novirsum, fr->vir_el_recip, &e_q,
 441                       fr->vir_lj_recip, &e_lj, &dvdl_q, &dvdl_lj,
 442                       &cycles_seppme);
 443     if (bSepDVDL)
 444     {
 445         gmx_print_sepdvdl(fplog, "Electrostatic PME mesh", e_q, dvdl_q);
 446         gmx_print_sepdvdl(fplog, "Lennard-Jones PME mesh", e_lj, dvdl_lj);
 447     }
 448     enerd->term[F_COUL_RECIP] += e_q;
 449     enerd->term[F_LJ_RECIP]   += e_lj;
 450     enerd->dvdl_lin[efptCOUL] += dvdl_q;
 451     enerd->dvdl_lin[efptVDW]  += dvdl_lj;
 452
 453     if (wcycle)
 454     {
 455         dd_cycles_add(cr->dd, cycles_seppme, ddCyclPME);
 456     }
 457     wallcycle_stop(wcycle, ewcPP_PMEWAITRECVF);
 458 }
 459
 460 static void print_large_forces(FILE *fp, t_mdatoms *md, t_commrec *cr,
 461                                gmx_int64_t step, real pforce, rvec *x, rvec *f)
 462 {
 463     int  i;
 464     real pf2, fn2;
 465     char buf[STEPSTRSIZE];
 466
 467     pf2 = sqr(pforce);
 468     for (i = 0; i < md->homenr; i++)
 469     {
 470         fn2 = norm2(f[i]);
 471         /* We also catch NAN, if the compiler does not optimize this away. */
 472         if (fn2 >= pf2 || fn2 != fn2)
 473         {
 474             fprintf(fp, "step %s  atom %6d  x %8.3f %8.3f %8.3f  force %12.5e\n",
 475                     gmx_step_str(step, buf),
 476                     ddglatnr(cr->dd, i), x[i][XX], x[i][YY], x[i][ZZ], sqrt(fn2));
 477         }
 478     }
 479 }
 480
 481 static void post_process_forces(t_commrec *cr,
 482                                 gmx_int64_t step,
 483                                 t_nrnb *nrnb, gmx_wallcycle_t wcycle,
 484                                 gmx_localtop_t *top,
 485                                 matrix box, rvec x[],
 486                                 rvec f[],
 487                                 tensor vir_force,
 488                                 t_mdatoms *mdatoms,
 489                                 t_graph *graph,
 490                                 t_forcerec *fr, gmx_vsite_t *vsite,
 491                                 int flags)
 492 {
 493     if (fr->bF_NoVirSum)
 494     {
 495         if (vsite)
 496         {
 497             /* Spread the mesh force on virtual sites to the other particles...
 498              * This is parallellized. MPI communication is performed
 499              * if the constructing atoms aren't local.
 500              */
 501             wallcycle_start(wcycle, ewcVSITESPREAD);
 502             spread_vsite_f(vsite, x, fr->f_novirsum, NULL,
 503                            (flags & GMX_FORCE_VIRIAL), fr->vir_el_recip,
 504                            nrnb,
 505                            &top->idef, fr->ePBC, fr->bMolPBC, graph, box, cr);
 506             wallcycle_stop(wcycle, ewcVSITESPREAD);
 507         }
 508         if (flags & GMX_FORCE_VIRIAL)
 509         {
 510             /* Now add the forces, this is local */
 511             if (fr->bDomDec)
 512             {
 513                 sum_forces(0, fr->f_novirsum_n, f, fr->f_novirsum);
 514             }
 515             else
 516             {
 517                 sum_forces(0, mdatoms->homenr,
 518                            f, fr->f_novirsum);
 519             }
 520             if (EEL_FULL(fr->eeltype))
 521             {
 522                 /* Add the mesh contribution to the virial */
 523                 m_add(vir_force, fr->vir_el_recip, vir_force);
 524             }
 525             if (EVDW_PME(fr->vdwtype))
 526             {
 527                 /* Add the mesh contribution to the virial */
 528                 m_add(vir_force, fr->vir_lj_recip, vir_force);
 529             }
 530             if (debug)
 531             {
 532                 pr_rvecs(debug, 0, "vir_force", vir_force, DIM);
 533             }
 534         }
 535     }
 536
 537     if (fr->print_force >= 0)
 538     {
 539         print_large_forces(stderr, mdatoms, cr, step, fr->print_force, x, f);
 540     }
 541 }
 542
 543 static void do_nb_verlet(t_forcerec *fr,
 544                          interaction_const_t *ic,
 545                          gmx_enerdata_t *enerd,
 546                          int flags, int ilocality,
 547                          int clearF,
 548                          t_nrnb *nrnb,
 549                          gmx_wallcycle_t wcycle)
 550 {
 551     int                        nnbl, kernel_type, enr_nbnxn_kernel_ljc, enr_nbnxn_kernel_lj;
 552     char                      *env;
 553     nonbonded_verlet_group_t  *nbvg;
 554     gmx_bool                   bCUDA;
 555
 556     if (!(flags & GMX_FORCE_NONBONDED))
 557     {
 558         /* skip non-bonded calculation */
 559         return;
 560     }
 561
 562     nbvg = &fr->nbv->grp[ilocality];
 563
 564     /* CUDA kernel launch overhead is already timed separately */
 565     if (fr->cutoff_scheme != ecutsVERLET)
 566     {
 567         gmx_incons("Invalid cut-off scheme passed!");
 568     }
 569
 570     bCUDA = (nbvg->kernel_type == nbnxnk8x8x8_CUDA);
 571
 572     if (!bCUDA)
 573     {
 574         wallcycle_sub_start(wcycle, ewcsNONBONDED);
 575     }
 576     switch (nbvg->kernel_type)
 577     {
 578         case nbnxnk4x4_PlainC:
 579             nbnxn_kernel_ref(&nbvg->nbl_lists,
 580                              nbvg->nbat, ic,
 581                              fr->shift_vec,
 582                              flags,
 583                              clearF,
 584                              fr->fshift[0],
 585                              enerd->grpp.ener[egCOULSR],
 586                              fr->bBHAM ?
 587                              enerd->grpp.ener[egBHAMSR] :
 588                              enerd->grpp.ener[egLJSR]);
 589             break;
 590
 591         case nbnxnk4xN_SIMD_4xN:
 592             nbnxn_kernel_simd_4xn(&nbvg->nbl_lists,
 593                                   nbvg->nbat, ic,
 594                                   nbvg->ewald_excl,
 595                                   fr->shift_vec,
 596                                   flags,
 597                                   clearF,
 598                                   fr->fshift[0],
 599                                   enerd->grpp.ener[egCOULSR],
 600                                   fr->bBHAM ?
 601                                   enerd->grpp.ener[egBHAMSR] :
 602                                   enerd->grpp.ener[egLJSR]);
 603             break;
 604         case nbnxnk4xN_SIMD_2xNN:
 605             nbnxn_kernel_simd_2xnn(&nbvg->nbl_lists,
 606                                    nbvg->nbat, ic,
 607                                    nbvg->ewald_excl,
 608                                    fr->shift_vec,
 609                                    flags,
 610                                    clearF,
 611                                    fr->fshift[0],
 612                                    enerd->grpp.ener[egCOULSR],
 613                                    fr->bBHAM ?
 614                                    enerd->grpp.ener[egBHAMSR] :
 615                                    enerd->grpp.ener[egLJSR]);
 616             break;
 617
 618         case nbnxnk8x8x8_CUDA:
 619             nbnxn_cuda_launch_kernel(fr->nbv->cu_nbv, nbvg->nbat, flags, ilocality);
 620             break;
 621
 622         case nbnxnk8x8x8_PlainC:
 623             nbnxn_kernel_gpu_ref(nbvg->nbl_lists.nbl[0],
 624                                  nbvg->nbat, ic,
 625                                  fr->shift_vec,
 626                                  flags,
 627                                  clearF,
 628                                  nbvg->nbat->out[0].f,
 629                                  fr->fshift[0],
 630                                  enerd->grpp.ener[egCOULSR],
 631                                  fr->bBHAM ?
 632                                  enerd->grpp.ener[egBHAMSR] :
 633                                  enerd->grpp.ener[egLJSR]);
 634             break;
 635
 636         default:
 637             gmx_incons("Invalid nonbonded kernel type passed!");
 638
 639     }
 640     if (!bCUDA)
 641     {
 642         wallcycle_sub_stop(wcycle, ewcsNONBONDED);
 643     }
 644
 645     if (EEL_RF(ic->eeltype) || ic->eeltype == eelCUT)
 646     {
 647         enr_nbnxn_kernel_ljc = eNR_NBNXN_LJ_RF;
 648     }
 649     else if ((!bCUDA && nbvg->ewald_excl == ewaldexclAnalytical) ||
 650              (bCUDA && nbnxn_cuda_is_kernel_ewald_analytical(fr->nbv->cu_nbv)))
 651     {
 652         enr_nbnxn_kernel_ljc = eNR_NBNXN_LJ_EWALD;
 653     }
 654     else
 655     {
 656         enr_nbnxn_kernel_ljc = eNR_NBNXN_LJ_TAB;
 657     }
 658     enr_nbnxn_kernel_lj = eNR_NBNXN_LJ;
 659     if (flags & GMX_FORCE_ENERGY)
 660     {
 661         /* In eNR_??? the nbnxn F+E kernels are always the F kernel + 1 */
 662         enr_nbnxn_kernel_ljc += 1;
 663         enr_nbnxn_kernel_lj  += 1;
 664     }
 665
 666     inc_nrnb(nrnb, enr_nbnxn_kernel_ljc,
 667              nbvg->nbl_lists.natpair_ljq);
 668     inc_nrnb(nrnb, enr_nbnxn_kernel_lj,
 669              nbvg->nbl_lists.natpair_lj);
 670     /* The Coulomb-only kernels are offset -eNR_NBNXN_LJ_RF+eNR_NBNXN_RF */
 671     inc_nrnb(nrnb, enr_nbnxn_kernel_ljc-eNR_NBNXN_LJ_RF+eNR_NBNXN_RF,
 672              nbvg->nbl_lists.natpair_q);
 673
 674     if (ic->vdw_modifier == eintmodFORCESWITCH)
 675     {
 676         /* We add up the switch cost separately */
 677         inc_nrnb(nrnb, eNR_NBNXN_ADD_LJ_FSW+((flags & GMX_FORCE_ENERGY) ? 1 : 0),
 678                  nbvg->nbl_lists.natpair_ljq + nbvg->nbl_lists.natpair_lj);
 679     }
 680     if (ic->vdw_modifier == eintmodPOTSWITCH)
 681     {
 682         /* We add up the switch cost separately */
 683         inc_nrnb(nrnb, eNR_NBNXN_ADD_LJ_PSW+((flags & GMX_FORCE_ENERGY) ? 1 : 0),
 684                  nbvg->nbl_lists.natpair_ljq + nbvg->nbl_lists.natpair_lj);
 685     }
 686     if (ic->vdwtype == evdwPME)
 687     {
 688         /* We add up the LJ Ewald cost separately */
 689         inc_nrnb(nrnb, eNR_NBNXN_ADD_LJ_EWALD+((flags & GMX_FORCE_ENERGY) ? 1 : 0),
 690                  nbvg->nbl_lists.natpair_ljq + nbvg->nbl_lists.natpair_lj);
 691     }
 692 }
 693
 694 static void do_nb_verlet_fep(nbnxn_pairlist_set_t *nbl_lists,
 695                              t_forcerec           *fr,
 696                              rvec                  x[],
 697                              rvec                  f[],
 698                              t_mdatoms            *mdatoms,
 699                              t_lambda             *fepvals,
 700                              real                 *lambda,
 701                              gmx_enerdata_t       *enerd,
 702                              int                   flags,
 703                              t_nrnb               *nrnb,
 704                              gmx_wallcycle_t       wcycle)
 705 {
 706     int              donb_flags;
 707     nb_kernel_data_t kernel_data;
 708     real             lam_i[efptNR];
 709     real             dvdl_nb[efptNR];
 710     int              th;
 711     int              i, j;
 712
 713     donb_flags = 0;
 714     /* Add short-range interactions */
 715     donb_flags |= GMX_NONBONDED_DO_SR;
 716
 717     /* Currently all group scheme kernels always calculate (shift-)forces */
 718     if (flags & GMX_FORCE_FORCES)
 719     {
 720         donb_flags |= GMX_NONBONDED_DO_FORCE;
 721     }
 722     if (flags & GMX_FORCE_VIRIAL)
 723     {
 724         donb_flags |= GMX_NONBONDED_DO_SHIFTFORCE;
 725     }
 726     if (flags & GMX_FORCE_ENERGY)
 727     {
 728         donb_flags |= GMX_NONBONDED_DO_POTENTIAL;
 729     }
 730     if (flags & GMX_FORCE_DO_LR)
 731     {
 732         donb_flags |= GMX_NONBONDED_DO_LR;
 733     }
 734
 735     kernel_data.flags  = donb_flags;
 736     kernel_data.lambda = lambda;
 737     kernel_data.dvdl   = dvdl_nb;
 738
 739     kernel_data.energygrp_elec = enerd->grpp.ener[egCOULSR];
 740     kernel_data.energygrp_vdw  = enerd->grpp.ener[egLJSR];
 741
 742     /* reset free energy components */
 743     for (i = 0; i < efptNR; i++)
 744     {
 745         dvdl_nb[i]  = 0;
 746     }
 747
 748     assert(gmx_omp_nthreads_get(emntNonbonded) == nbl_lists->nnbl);
 749
 750     wallcycle_sub_start(wcycle, ewcsNONBONDED);
 751 #pragma omp parallel for schedule(static) num_threads(nbl_lists->nnbl)
 752     for (th = 0; th < nbl_lists->nnbl; th++)
 753     {
 754         gmx_nb_free_energy_kernel(nbl_lists->nbl_fep[th],
 755                                   x, f, fr, mdatoms, &kernel_data, nrnb);
 756     }
 757
 758     if (fepvals->sc_alpha != 0)
 759     {
 760         enerd->dvdl_nonlin[efptVDW]  += dvdl_nb[efptVDW];
 761         enerd->dvdl_nonlin[efptCOUL] += dvdl_nb[efptCOUL];
 762     }
 763     else
 764     {
 765         enerd->dvdl_lin[efptVDW]  += dvdl_nb[efptVDW];
 766         enerd->dvdl_lin[efptCOUL] += dvdl_nb[efptCOUL];
 767     }
 768
 769     /* If we do foreign lambda and we have soft-core interactions
 770      * we have to recalculate the (non-linear) energies contributions.
 771      */
 772     if (fepvals->n_lambda > 0 && (flags & GMX_FORCE_DHDL) && fepvals->sc_alpha != 0)
 773     {
 774         kernel_data.flags          = (donb_flags & ~(GMX_NONBONDED_DO_FORCE | GMX_NONBONDED_DO_SHIFTFORCE)) | GMX_NONBONDED_DO_FOREIGNLAMBDA;
 775         kernel_data.lambda         = lam_i;
 776         kernel_data.energygrp_elec = enerd->foreign_grpp.ener[egCOULSR];
 777         kernel_data.energygrp_vdw  = enerd->foreign_grpp.ener[egLJSR];
 778         /* Note that we add to kernel_data.dvdl, but ignore the result */
 779
 780         for (i = 0; i < enerd->n_lambda; i++)
 781         {
 782             for (j = 0; j < efptNR; j++)
 783             {
 784                 lam_i[j] = (i == 0 ? lambda[j] : fepvals->all_lambda[j][i-1]);
 785             }
 786             reset_foreign_enerdata(enerd);
 787 #pragma omp parallel for schedule(static) num_threads(nbl_lists->nnbl)
 788             for (th = 0; th < nbl_lists->nnbl; th++)
 789             {
 790                 gmx_nb_free_energy_kernel(nbl_lists->nbl_fep[th],
 791                                           x, f, fr, mdatoms, &kernel_data, nrnb);
 792             }
 793
 794             sum_epot(&(enerd->foreign_grpp), enerd->foreign_term);
 795             enerd->enerpart_lambda[i] += enerd->foreign_term[F_EPOT];
 796         }
 797     }
 798
 799     wallcycle_sub_stop(wcycle, ewcsNONBONDED);
 800 }
 801
 802 void do_force_cutsVERLET(FILE *fplog, t_commrec *cr,
 803                          t_inputrec *inputrec,
 804                          gmx_int64_t step, t_nrnb *nrnb, gmx_wallcycle_t wcycle,
 805                          gmx_localtop_t *top,
 806                          gmx_groups_t gmx_unused *groups,
 807                          matrix box, rvec x[], history_t *hist,
 808                          rvec f[],
 809                          tensor vir_force,
 810                          t_mdatoms *mdatoms,
 811                          gmx_enerdata_t *enerd, t_fcdata *fcd,
 812                          real *lambda, t_graph *graph,
 813                          t_forcerec *fr, interaction_const_t *ic,
 814                          gmx_vsite_t *vsite, rvec mu_tot,
 815                          double t, FILE *field, gmx_edsam_t ed,
 816                          gmx_bool bBornRadii,
 817                          int flags)
 818 {
 819     int                 cg0, cg1, i, j;
 820     int                 start, homenr;
 821     int                 nb_kernel_type;
 822     double              mu[2*DIM];
 823     gmx_bool            bSepDVDL, bStateChanged, bNS, bFillGrid, bCalcCGCM, bBS;
 824     gmx_bool            bDoLongRange, bDoForces, bSepLRF, bUseGPU, bUseOrEmulGPU;
 825     gmx_bool            bDiffKernels = FALSE;
 826     matrix              boxs;
 827     rvec                vzero, box_diag;
 828     real                e, v, dvdl;
 829     float               cycles_pme, cycles_force, cycles_wait_gpu;
 830     nonbonded_verlet_t *nbv;
 831
 832     cycles_force    = 0;
 833     cycles_wait_gpu = 0;
 834     nbv             = fr->nbv;
 835     nb_kernel_type  = fr->nbv->grp[0].kernel_type;
 836
 837     start  = 0;
 838     homenr = mdatoms->homenr;
 839
 840     bSepDVDL = (fr->bSepDVDL && do_per_step(step, inputrec->nstlog));
 841
 842     clear_mat(vir_force);
 843
 844     cg0 = 0;
 845     if (DOMAINDECOMP(cr))
 846     {
 847         cg1 = cr->dd->ncg_tot;
 848     }
 849     else
 850     {
 851         cg1 = top->cgs.nr;
 852     }
 853     if (fr->n_tpi > 0)
 854     {
 855         cg1--;
 856     }
 857
 858     bStateChanged = (flags & GMX_FORCE_STATECHANGED);
 859     bNS           = (flags & GMX_FORCE_NS) && (fr->bAllvsAll == FALSE);
 860     bFillGrid     = (bNS && bStateChanged);
 861     bCalcCGCM     = (bFillGrid && !DOMAINDECOMP(cr));
 862     bDoLongRange  = (fr->bTwinRange && bNS && (flags & GMX_FORCE_DO_LR));
 863     bDoForces     = (flags & GMX_FORCE_FORCES);
 864     bSepLRF       = (bDoLongRange && bDoForces && (flags & GMX_FORCE_SEPLRF));
 865     bUseGPU       = fr->nbv->bUseGPU;
 866     bUseOrEmulGPU = bUseGPU || (nbv->grp[0].kernel_type == nbnxnk8x8x8_PlainC);
 867
 868     if (bStateChanged)
 869     {
 870         update_forcerec(fr, box);
 871
 872         if (NEED_MUTOT(*inputrec))
 873         {
 874             /* Calculate total (local) dipole moment in a temporary common array.
 875              * This makes it possible to sum them over nodes faster.
 876              */
 877             calc_mu(start, homenr,
 878                     x, mdatoms->chargeA, mdatoms->chargeB, mdatoms->nChargePerturbed,
 879                     mu, mu+DIM);
 880         }
 881     }
 882
 883     if (fr->ePBC != epbcNONE)
 884     {
 885         /* Compute shift vectors every step,
 886          * because of pressure coupling or box deformation!
 887          */
 888         if ((flags & GMX_FORCE_DYNAMICBOX) && bStateChanged)
 889         {
 890             calc_shifts(box, fr->shift_vec);
 891         }
 892
 893         if (bCalcCGCM)
 894         {
 895             put_atoms_in_box_omp(fr->ePBC, box, homenr, x);
 896             inc_nrnb(nrnb, eNR_SHIFTX, homenr);
 897         }
 898         else if (EI_ENERGY_MINIMIZATION(inputrec->eI) && graph)
 899         {
 900             unshift_self(graph, box, x);
 901         }
 902     }
 903
 904     nbnxn_atomdata_copy_shiftvec(flags & GMX_FORCE_DYNAMICBOX,
 905                                  fr->shift_vec, nbv->grp[0].nbat);
 906
 907 #ifdef GMX_MPI
 908     if (!(cr->duty & DUTY_PME))
 909     {
 910         /* Send particle coordinates to the pme nodes.
 911          * Since this is only implemented for domain decomposition
 912          * and domain decomposition does not use the graph,
 913          * we do not need to worry about shifting.
 914          */
 915
 916         int pme_flags = 0;
 917
 918         wallcycle_start(wcycle, ewcPP_PMESENDX);
 919
 920         bBS = (inputrec->nwall == 2);
 921         if (bBS)
 922         {
 923             copy_mat(box, boxs);
 924             svmul(inputrec->wall_ewald_zfac, boxs[ZZ], boxs[ZZ]);
 925         }
 926
 927         if (EEL_PME(fr->eeltype))
 928         {
 929             pme_flags |= GMX_PME_DO_COULOMB;
 930         }
 931
 932         if (EVDW_PME(fr->vdwtype))
 933         {
 934             pme_flags |= GMX_PME_DO_LJ;
 935         }
 936
 937         gmx_pme_send_coordinates(cr, bBS ? boxs : box, x,
 938                                  mdatoms->nChargePerturbed, mdatoms->nTypePerturbed, lambda[efptCOUL], lambda[efptVDW],
 939                                  (flags & (GMX_FORCE_VIRIAL | GMX_FORCE_ENERGY)),
 940                                  pme_flags, step);
 941
 942         wallcycle_stop(wcycle, ewcPP_PMESENDX);
 943     }
 944 #endif /* GMX_MPI */
 945
 946     /* do gridding for pair search */
 947     if (bNS)
 948     {
 949         if (graph && bStateChanged)
 950         {
 951             /* Calculate intramolecular shift vectors to make molecules whole */
 952             mk_mshift(fplog, graph, fr->ePBC, box, x);
 953         }
 954
 955         clear_rvec(vzero);
 956         box_diag[XX] = box[XX][XX];
 957         box_diag[YY] = box[YY][YY];
 958         box_diag[ZZ] = box[ZZ][ZZ];
 959
 960         wallcycle_start(wcycle, ewcNS);
 961         if (!fr->bDomDec)
 962         {
 963             wallcycle_sub_start(wcycle, ewcsNBS_GRID_LOCAL);
 964             nbnxn_put_on_grid(nbv->nbs, fr->ePBC, box,
 965                               0, vzero, box_diag,
 966                               0, mdatoms->homenr, -1, fr->cginfo, x,
 967                               0, NULL,
 968                               nbv->grp[eintLocal].kernel_type,
 969                               nbv->grp[eintLocal].nbat);
 970             wallcycle_sub_stop(wcycle, ewcsNBS_GRID_LOCAL);
 971         }
 972         else
 973         {
 974             wallcycle_sub_start(wcycle, ewcsNBS_GRID_NONLOCAL);
 975             nbnxn_put_on_grid_nonlocal(nbv->nbs, domdec_zones(cr->dd),
 976                                        fr->cginfo, x,
 977                                        nbv->grp[eintNonlocal].kernel_type,
 978                                        nbv->grp[eintNonlocal].nbat);
 979             wallcycle_sub_stop(wcycle, ewcsNBS_GRID_NONLOCAL);
 980         }
 981
 982         if (nbv->ngrp == 1 ||
 983             nbv->grp[eintNonlocal].nbat == nbv->grp[eintLocal].nbat)
 984         {
 985             nbnxn_atomdata_set(nbv->grp[eintLocal].nbat, eatAll,
 986                                nbv->nbs, mdatoms, fr->cginfo);
 987         }
 988         else
 989         {
 990             nbnxn_atomdata_set(nbv->grp[eintLocal].nbat, eatLocal,
 991                                nbv->nbs, mdatoms, fr->cginfo);
 992             nbnxn_atomdata_set(nbv->grp[eintNonlocal].nbat, eatAll,
 993                                nbv->nbs, mdatoms, fr->cginfo);
 994         }
 995         wallcycle_stop(wcycle, ewcNS);
 996     }
 997
 998     /* initialize the GPU atom data and copy shift vector */
 999     if (bUseGPU)
1000     {
1001         if (bNS)
1002         {
1003             wallcycle_start_nocount(wcycle, ewcLAUNCH_GPU_NB);
1004             nbnxn_cuda_init_atomdata(nbv->cu_nbv, nbv->grp[eintLocal].nbat);
1005             wallcycle_stop(wcycle, ewcLAUNCH_GPU_NB);
1006         }
1007
1008         wallcycle_start_nocount(wcycle, ewcLAUNCH_GPU_NB);
1009         nbnxn_cuda_upload_shiftvec(nbv->cu_nbv, nbv->grp[eintLocal].nbat);
1010         wallcycle_stop(wcycle, ewcLAUNCH_GPU_NB);
1011     }
1012
1013     /* do local pair search */
1014     if (bNS)
1015     {
1016         wallcycle_start_nocount(wcycle, ewcNS);
1017         wallcycle_sub_start(wcycle, ewcsNBS_SEARCH_LOCAL);
1018         nbnxn_make_pairlist(nbv->nbs, nbv->grp[eintLocal].nbat,
1019                             &top->excls,
1020                             ic->rlist,
1021                             nbv->min_ci_balanced,
1022                             &nbv->grp[eintLocal].nbl_lists,
1023                             eintLocal,
1024                             nbv->grp[eintLocal].kernel_type,
1025                             nrnb);
1026         wallcycle_sub_stop(wcycle, ewcsNBS_SEARCH_LOCAL);
1027
1028         if (bUseGPU)
1029         {
1030             /* initialize local pair-list on the GPU */
1031             nbnxn_cuda_init_pairlist(nbv->cu_nbv,
1032                                      nbv->grp[eintLocal].nbl_lists.nbl[0],
1033                                      eintLocal);
1034         }
1035         wallcycle_stop(wcycle, ewcNS);
1036     }
1037     else
1038     {
1039         wallcycle_start(wcycle, ewcNB_XF_BUF_OPS);
1040         wallcycle_sub_start(wcycle, ewcsNB_X_BUF_OPS);
1041         nbnxn_atomdata_copy_x_to_nbat_x(nbv->nbs, eatLocal, FALSE, x,
1042                                         nbv->grp[eintLocal].nbat);
1043         wallcycle_sub_stop(wcycle, ewcsNB_X_BUF_OPS);
1044         wallcycle_stop(wcycle, ewcNB_XF_BUF_OPS);
1045     }
1046
1047     if (bUseGPU)
1048     {
1049         wallcycle_start(wcycle, ewcLAUNCH_GPU_NB);
1050         /* launch local nonbonded F on GPU */
1051         do_nb_verlet(fr, ic, enerd, flags, eintLocal, enbvClearFNo,
1052                      nrnb, wcycle);
1053         wallcycle_stop(wcycle, ewcLAUNCH_GPU_NB);
1054     }
1055
1056     /* Communicate coordinates and sum dipole if necessary +
1057        do non-local pair search */
1058     if (DOMAINDECOMP(cr))
1059     {
1060         bDiffKernels = (nbv->grp[eintNonlocal].kernel_type !=
1061                         nbv->grp[eintLocal].kernel_type);
1062
1063         if (bDiffKernels)
1064         {
1065             /* With GPU+CPU non-bonded calculations we need to copy
1066              * the local coordinates to the non-local nbat struct
1067              * (in CPU format) as the non-local kernel call also
1068              * calculates the local - non-local interactions.
1069              */
1070             wallcycle_start(wcycle, ewcNB_XF_BUF_OPS);
1071             wallcycle_sub_start(wcycle, ewcsNB_X_BUF_OPS);
1072             nbnxn_atomdata_copy_x_to_nbat_x(nbv->nbs, eatLocal, TRUE, x,
1073                                             nbv->grp[eintNonlocal].nbat);
1074             wallcycle_sub_stop(wcycle, ewcsNB_X_BUF_OPS);
1075             wallcycle_stop(wcycle, ewcNB_XF_BUF_OPS);
1076         }
1077
1078         if (bNS)
1079         {
1080             wallcycle_start_nocount(wcycle, ewcNS);
1081             wallcycle_sub_start(wcycle, ewcsNBS_SEARCH_NONLOCAL);
1082
1083             if (bDiffKernels)
1084             {
1085                 nbnxn_grid_add_simple(nbv->nbs, nbv->grp[eintNonlocal].nbat);
1086             }
1087
1088             nbnxn_make_pairlist(nbv->nbs, nbv->grp[eintNonlocal].nbat,
1089                                 &top->excls,
1090                                 ic->rlist,
1091                                 nbv->min_ci_balanced,
1092                                 &nbv->grp[eintNonlocal].nbl_lists,
1093                                 eintNonlocal,
1094                                 nbv->grp[eintNonlocal].kernel_type,
1095                                 nrnb);
1096
1097             wallcycle_sub_stop(wcycle, ewcsNBS_SEARCH_NONLOCAL);
1098
1099             if (nbv->grp[eintNonlocal].kernel_type == nbnxnk8x8x8_CUDA)
1100             {
1101                 /* initialize non-local pair-list on the GPU */
1102                 nbnxn_cuda_init_pairlist(nbv->cu_nbv,
1103                                          nbv->grp[eintNonlocal].nbl_lists.nbl[0],
1104                                          eintNonlocal);
1105             }
1106             wallcycle_stop(wcycle, ewcNS);
1107         }
1108         else
1109         {
1110             wallcycle_start(wcycle, ewcMOVEX);
1111             dd_move_x(cr->dd, box, x);
1112
1113             /* When we don't need the total dipole we sum it in global_stat */
1114             if (bStateChanged && NEED_MUTOT(*inputrec))
1115             {
1116                 gmx_sumd(2*DIM, mu, cr);
1117             }
1118             wallcycle_stop(wcycle, ewcMOVEX);
1119
1120             wallcycle_start(wcycle, ewcNB_XF_BUF_OPS);
1121             wallcycle_sub_start(wcycle, ewcsNB_X_BUF_OPS);
1122             nbnxn_atomdata_copy_x_to_nbat_x(nbv->nbs, eatNonlocal, FALSE, x,
1123                                             nbv->grp[eintNonlocal].nbat);
1124             wallcycle_sub_stop(wcycle, ewcsNB_X_BUF_OPS);
1125             cycles_force += wallcycle_stop(wcycle, ewcNB_XF_BUF_OPS);
1126         }
1127
1128         if (bUseGPU && !bDiffKernels)
1129         {
1130             wallcycle_start(wcycle, ewcLAUNCH_GPU_NB);
1131             /* launch non-local nonbonded F on GPU */
1132             do_nb_verlet(fr, ic, enerd, flags, eintNonlocal, enbvClearFNo,
1133                          nrnb, wcycle);
1134             cycles_force += wallcycle_stop(wcycle, ewcLAUNCH_GPU_NB);
1135         }
1136     }
1137
1138     if (bUseGPU)
1139     {
1140         /* launch D2H copy-back F */
1141         wallcycle_start_nocount(wcycle, ewcLAUNCH_GPU_NB);
1142         if (DOMAINDECOMP(cr) && !bDiffKernels)
1143         {
1144             nbnxn_cuda_launch_cpyback(nbv->cu_nbv, nbv->grp[eintNonlocal].nbat,
1145                                       flags, eatNonlocal);
1146         }
1147         nbnxn_cuda_launch_cpyback(nbv->cu_nbv, nbv->grp[eintLocal].nbat,
1148                                   flags, eatLocal);
1149         cycles_force += wallcycle_stop(wcycle, ewcLAUNCH_GPU_NB);
1150     }
1151
1152     if (bStateChanged && NEED_MUTOT(*inputrec))
1153     {
1154         if (PAR(cr))
1155         {
1156             gmx_sumd(2*DIM, mu, cr);
1157         }
1158
1159         for (i = 0; i < 2; i++)
1160         {
1161             for (j = 0; j < DIM; j++)
1162             {
1163                 fr->mu_tot[i][j] = mu[i*DIM + j];
1164             }
1165         }
1166     }
1167     if (fr->efep == efepNO)
1168     {
1169         copy_rvec(fr->mu_tot[0], mu_tot);
1170     }
1171     else
1172     {
1173         for (j = 0; j < DIM; j++)
1174         {
1175             mu_tot[j] =
1176                 (1.0 - lambda[efptCOUL])*fr->mu_tot[0][j] +
1177                 lambda[efptCOUL]*fr->mu_tot[1][j];
1178         }
1179     }
1180
1181     /* Reset energies */
1182     reset_enerdata(fr, bNS, enerd, MASTER(cr));
1183     clear_rvecs(SHIFTS, fr->fshift);
1184
1185     if (DOMAINDECOMP(cr) && !(cr->duty & DUTY_PME))
1186     {
1187         wallcycle_start(wcycle, ewcPPDURINGPME);
1188         dd_force_flop_start(cr->dd, nrnb);
1189     }
1190
1191     if (inputrec->bRot)
1192     {
1193         /* Enforced rotation has its own cycle counter that starts after the collective
1194          * coordinates have been communicated. It is added to ddCyclF to allow
1195          * for proper load-balancing */
1196         wallcycle_start(wcycle, ewcROT);
1197         do_rotation(cr, inputrec, box, x, t, step, wcycle, bNS);
1198         wallcycle_stop(wcycle, ewcROT);
1199     }
1200
1201     /* Start the force cycle counter.
1202      * This counter is stopped in do_forcelow_level.
1203      * No parallel communication should occur while this counter is running,
1204      * since that will interfere with the dynamic load balancing.
1205      */
1206     wallcycle_start(wcycle, ewcFORCE);
1207     if (bDoForces)
1208     {
1209         /* Reset forces for which the virial is calculated separately:
1210          * PME/Ewald forces if necessary */
1211         if (fr->bF_NoVirSum)
1212         {
1213             if (flags & GMX_FORCE_VIRIAL)
1214             {
1215                 fr->f_novirsum = fr->f_novirsum_alloc;
1216                 if (fr->bDomDec)
1217                 {
1218                     clear_rvecs(fr->f_novirsum_n, fr->f_novirsum);
1219                 }
1220                 else
1221                 {
1222                     clear_rvecs(homenr, fr->f_novirsum+start);
1223                 }
1224             }
1225             else
1226             {
1227                 /* We are not calculating the pressure so we do not need
1228                  * a separate array for forces that do not contribute
1229                  * to the pressure.
1230                  */
1231                 fr->f_novirsum = f;
1232             }
1233         }
1234
1235         /* Clear the short- and long-range forces */
1236         clear_rvecs(fr->natoms_force_constr, f);
1237         if (bSepLRF && do_per_step(step, inputrec->nstcalclr))
1238         {
1239             clear_rvecs(fr->natoms_force_constr, fr->f_twin);
1240         }
1241
1242         clear_rvec(fr->vir_diag_posres);
1243     }
1244
1245     if (inputrec->ePull == epullCONSTRAINT)
1246     {
1247         clear_pull_forces(inputrec->pull);
1248     }
1249
1250     /* We calculate the non-bonded forces, when done on the CPU, here.
1251      * We do this before calling do_force_lowlevel, as in there bondeds
1252      * forces are calculated before PME, which does communication.
1253      * With this order, non-bonded and bonded force calculation imbalance
1254      * can be balanced out by the domain decomposition load balancing.
1255      */
1256
1257     if (!bUseOrEmulGPU)
1258     {
1259         /* Maybe we should move this into do_force_lowlevel */
1260         do_nb_verlet(fr, ic, enerd, flags, eintLocal, enbvClearFYes,
1261                      nrnb, wcycle);
1262     }
1263
1264     if (fr->efep != efepNO)
1265     {
1266         /* Calculate the local and non-local free energy interactions here.
1267          * Happens here on the CPU both with and without GPU.
1268          */
1269         if (fr->nbv->grp[eintLocal].nbl_lists.nbl_fep[0]->nrj > 0)
1270         {
1271             do_nb_verlet_fep(&fr->nbv->grp[eintLocal].nbl_lists,
1272                              fr, x, f, mdatoms,
1273                              inputrec->fepvals, lambda,
1274                              enerd, flags, nrnb, wcycle);
1275         }
1276
1277         if (DOMAINDECOMP(cr) &&
1278             fr->nbv->grp[eintNonlocal].nbl_lists.nbl_fep[0]->nrj > 0)
1279         {
1280             do_nb_verlet_fep(&fr->nbv->grp[eintNonlocal].nbl_lists,
1281                              fr, x, f, mdatoms,
1282                              inputrec->fepvals, lambda,
1283                              enerd, flags, nrnb, wcycle);
1284         }
1285     }
1286
1287     if (!bUseOrEmulGPU || bDiffKernels)
1288     {
1289         int aloc;
1290
1291         if (DOMAINDECOMP(cr))
1292         {
1293             do_nb_verlet(fr, ic, enerd, flags, eintNonlocal,
1294                          bDiffKernels ? enbvClearFYes : enbvClearFNo,
1295                          nrnb, wcycle);
1296         }
1297
1298         if (!bUseOrEmulGPU)
1299         {
1300             aloc = eintLocal;
1301         }
1302         else
1303         {
1304             aloc = eintNonlocal;
1305         }
1306
1307         /* Add all the non-bonded force to the normal force array.
1308          * This can be split into a local a non-local part when overlapping
1309          * communication with calculation with domain decomposition.
1310          */
1311         cycles_force += wallcycle_stop(wcycle, ewcFORCE);
1312         wallcycle_start(wcycle, ewcNB_XF_BUF_OPS);
1313         wallcycle_sub_start(wcycle, ewcsNB_F_BUF_OPS);
1314         nbnxn_atomdata_add_nbat_f_to_f(nbv->nbs, eatAll, nbv->grp[aloc].nbat, f);
1315         wallcycle_sub_stop(wcycle, ewcsNB_F_BUF_OPS);
1316         cycles_force += wallcycle_stop(wcycle, ewcNB_XF_BUF_OPS);
1317         wallcycle_start_nocount(wcycle, ewcFORCE);
1318
1319         /* if there are multiple fshift output buffers reduce them */
1320         if ((flags & GMX_FORCE_VIRIAL) &&
1321             nbv->grp[aloc].nbl_lists.nnbl > 1)
1322         {
1323             nbnxn_atomdata_add_nbat_fshift_to_fshift(nbv->grp[aloc].nbat,
1324                                                      fr->fshift);
1325         }
1326     }
1327
1328     /* update QMMMrec, if necessary */
1329     if (fr->bQMMM)
1330     {
1331         update_QMMMrec(cr, fr, x, mdatoms, box, top);
1332     }
1333
1334     if ((flags & GMX_FORCE_BONDED) && top->idef.il[F_POSRES].nr > 0)
1335     {
1336         posres_wrapper(fplog, flags, bSepDVDL, inputrec, nrnb, top, box, x,
1337                        enerd, lambda, fr);
1338     }
1339
1340     if ((flags & GMX_FORCE_BONDED) && top->idef.il[F_FBPOSRES].nr > 0)
1341     {
1342         fbposres_wrapper(inputrec, nrnb, top, box, x, enerd, fr);
1343     }
1344
1345     /* Compute the bonded and non-bonded energies and optionally forces */
1346     do_force_lowlevel(fplog, step, fr, inputrec, &(top->idef),
1347                       cr, nrnb, wcycle, mdatoms,
1348                       x, hist, f, bSepLRF ? fr->f_twin : f, enerd, fcd, top, fr->born,
1349                       &(top->atomtypes), bBornRadii, box,
1350                       inputrec->fepvals, lambda, graph, &(top->excls), fr->mu_tot,
1351                       flags, &cycles_pme);
1352
1353     if (bSepLRF)
1354     {
1355         if (do_per_step(step, inputrec->nstcalclr))
1356         {
1357             /* Add the long range forces to the short range forces */
1358             for (i = 0; i < fr->natoms_force_constr; i++)
1359             {
1360                 rvec_add(fr->f_twin[i], f[i], f[i]);
1361             }
1362         }
1363     }
1364
1365     cycles_force += wallcycle_stop(wcycle, ewcFORCE);
1366
1367     if (ed)
1368     {
1369         do_flood(cr, inputrec, x, f, ed, box, step, bNS);
1370     }
1371
1372     if (bUseOrEmulGPU && !bDiffKernels)
1373     {
1374         /* wait for non-local forces (or calculate in emulation mode) */
1375         if (DOMAINDECOMP(cr))
1376         {
1377             if (bUseGPU)
1378             {
1379                 float cycles_tmp;
1380
1381                 wallcycle_start(wcycle, ewcWAIT_GPU_NB_NL);
1382                 nbnxn_cuda_wait_gpu(nbv->cu_nbv,
1383                                     nbv->grp[eintNonlocal].nbat,
1384                                     flags, eatNonlocal,
1385                                     enerd->grpp.ener[egLJSR], enerd->grpp.ener[egCOULSR],
1386                                     fr->fshift);
1387                 cycles_tmp       = wallcycle_stop(wcycle, ewcWAIT_GPU_NB_NL);
1388                 cycles_wait_gpu += cycles_tmp;
1389                 cycles_force    += cycles_tmp;
1390             }
1391             else
1392             {
1393                 wallcycle_start_nocount(wcycle, ewcFORCE);
1394                 do_nb_verlet(fr, ic, enerd, flags, eintNonlocal, enbvClearFYes,
1395                              nrnb, wcycle);
1396                 cycles_force += wallcycle_stop(wcycle, ewcFORCE);
1397             }
1398             wallcycle_start(wcycle, ewcNB_XF_BUF_OPS);
1399             wallcycle_sub_start(wcycle, ewcsNB_F_BUF_OPS);
1400             /* skip the reduction if there was no non-local work to do */
1401             if (nbv->grp[eintLocal].nbl_lists.nbl[0]->nsci > 0)
1402             {
1403                 nbnxn_atomdata_add_nbat_f_to_f(nbv->nbs, eatNonlocal,
1404                                                nbv->grp[eintNonlocal].nbat, f);
1405             }
1406             wallcycle_sub_stop(wcycle, ewcsNB_F_BUF_OPS);
1407             cycles_force += wallcycle_stop(wcycle, ewcNB_XF_BUF_OPS);
1408         }
1409     }
1410
1411     if (bDoForces && DOMAINDECOMP(cr))
1412     {
1413         /* Communicate the forces */
1414         wallcycle_start(wcycle, ewcMOVEF);
1415         dd_move_f(cr->dd, f, fr->fshift);
1416         /* Do we need to communicate the separate force array
1417          * for terms that do not contribute to the single sum virial?
1418          * Position restraints and electric fields do not introduce
1419          * inter-cg forces, only full electrostatics methods do.
1420          * When we do not calculate the virial, fr->f_novirsum = f,
1421          * so we have already communicated these forces.
1422          */
1423         if (EEL_FULL(fr->eeltype) && cr->dd->n_intercg_excl &&
1424             (flags & GMX_FORCE_VIRIAL))
1425         {
1426             dd_move_f(cr->dd, fr->f_novirsum, NULL);
1427         }
1428         if (bSepLRF)
1429         {
1430             /* We should not update the shift forces here,
1431              * since f_twin is already included in f.
1432              */
1433             dd_move_f(cr->dd, fr->f_twin, NULL);
1434         }
1435         wallcycle_stop(wcycle, ewcMOVEF);
1436     }
1437
1438     if (bUseOrEmulGPU)
1439     {
1440         /* wait for local forces (or calculate in emulation mode) */
1441         if (bUseGPU)
1442         {
1443             wallcycle_start(wcycle, ewcWAIT_GPU_NB_L);
1444             nbnxn_cuda_wait_gpu(nbv->cu_nbv,
1445                                 nbv->grp[eintLocal].nbat,
1446                                 flags, eatLocal,
1447                                 enerd->grpp.ener[egLJSR], enerd->grpp.ener[egCOULSR],
1448                                 fr->fshift);
1449             cycles_wait_gpu += wallcycle_stop(wcycle, ewcWAIT_GPU_NB_L);
1450
1451             /* now clear the GPU outputs while we finish the step on the CPU */
1452
1453             wallcycle_start_nocount(wcycle, ewcLAUNCH_GPU_NB);
1454             nbnxn_cuda_clear_outputs(nbv->cu_nbv, flags);
1455             wallcycle_stop(wcycle, ewcLAUNCH_GPU_NB);
1456         }
1457         else
1458         {
1459             wallcycle_start_nocount(wcycle, ewcFORCE);
1460             do_nb_verlet(fr, ic, enerd, flags, eintLocal,
1461                          DOMAINDECOMP(cr) ? enbvClearFNo : enbvClearFYes,
1462                          nrnb, wcycle);
1463             wallcycle_stop(wcycle, ewcFORCE);
1464         }
1465         wallcycle_start(wcycle, ewcNB_XF_BUF_OPS);
1466         wallcycle_sub_start(wcycle, ewcsNB_F_BUF_OPS);
1467         if (nbv->grp[eintLocal].nbl_lists.nbl[0]->nsci > 0)
1468         {
1469             /* skip the reduction if there was no non-local work to do */
1470             nbnxn_atomdata_add_nbat_f_to_f(nbv->nbs, eatLocal,
1471                                            nbv->grp[eintLocal].nbat, f);
1472         }
1473         wallcycle_sub_stop(wcycle, ewcsNB_F_BUF_OPS);
1474         wallcycle_stop(wcycle, ewcNB_XF_BUF_OPS);
1475     }
1476
1477     if (DOMAINDECOMP(cr))
1478     {
1479         dd_force_flop_stop(cr->dd, nrnb);
1480         if (wcycle)
1481         {
1482             dd_cycles_add(cr->dd, cycles_force-cycles_pme, ddCyclF);
1483             if (bUseGPU)
1484             {
1485                 dd_cycles_add(cr->dd, cycles_wait_gpu, ddCyclWaitGPU);
1486             }
1487         }
1488     }
1489
1490     if (bDoForces)
1491     {
1492         if (IR_ELEC_FIELD(*inputrec))
1493         {
1494             /* Compute forces due to electric field */
1495             calc_f_el(MASTER(cr) ? field : NULL,
1496                       start, homenr, mdatoms->chargeA, fr->f_novirsum,
1497                       inputrec->ex, inputrec->et, t);
1498         }
1499
1500         /* If we have NoVirSum forces, but we do not calculate the virial,
1501          * we sum fr->f_novirum=f later.
1502          */
1503         if (vsite && !(fr->bF_NoVirSum && !(flags & GMX_FORCE_VIRIAL)))
1504         {
1505             wallcycle_start(wcycle, ewcVSITESPREAD);
1506             spread_vsite_f(vsite, x, f, fr->fshift, FALSE, NULL, nrnb,
1507                            &top->idef, fr->ePBC, fr->bMolPBC, graph, box, cr);
1508             wallcycle_stop(wcycle, ewcVSITESPREAD);
1509
1510             if (bSepLRF)
1511             {
1512                 wallcycle_start(wcycle, ewcVSITESPREAD);
1513                 spread_vsite_f(vsite, x, fr->f_twin, NULL, FALSE, NULL,
1514                                nrnb,
1515                                &top->idef, fr->ePBC, fr->bMolPBC, graph, box, cr);
1516                 wallcycle_stop(wcycle, ewcVSITESPREAD);
1517             }
1518         }
1519
1520         if (flags & GMX_FORCE_VIRIAL)
1521         {
1522             /* Calculation of the virial must be done after vsites! */
1523             calc_virial(0, mdatoms->homenr, x, f,
1524                         vir_force, graph, box, nrnb, fr, inputrec->ePBC);
1525         }
1526     }
1527
1528     if (inputrec->ePull == epullUMBRELLA || inputrec->ePull == epullCONST_F)
1529     {
1530         pull_potential_wrapper(fplog, bSepDVDL, cr, inputrec, box, x,
1531                                f, vir_force, mdatoms, enerd, lambda, t);
1532     }
1533
1534     /* Add the forces from enforced rotation potentials (if any) */
1535     if (inputrec->bRot)
1536     {
1537         wallcycle_start(wcycle, ewcROTadd);
1538         enerd->term[F_COM_PULL] += add_rot_forces(inputrec->rot, f, cr, step, t);
1539         wallcycle_stop(wcycle, ewcROTadd);
1540     }
1541
1542     /* Add forces from interactive molecular dynamics (IMD), if bIMD == TRUE. */
1543     IMD_apply_forces(inputrec->bIMD, inputrec->imd, cr, f, wcycle);
1544
1545     if (PAR(cr) && !(cr->duty & DUTY_PME))
1546     {
1547         /* In case of node-splitting, the PP nodes receive the long-range
1548          * forces, virial and energy from the PME nodes here.
1549          */
1550         pme_receive_force_ener(fplog, bSepDVDL, cr, wcycle, enerd, fr);
1551     }
1552
1553     if (bDoForces)
1554     {
1555         post_process_forces(cr, step, nrnb, wcycle,
1556                             top, box, x, f, vir_force, mdatoms, graph, fr, vsite,
1557                             flags);
1558     }
1559
1560     /* Sum the potential energy terms from group contributions */
1561     sum_epot(&(enerd->grpp), enerd->term);
1562 }
1563
1564 void do_force_cutsGROUP(FILE *fplog, t_commrec *cr,
1565                         t_inputrec *inputrec,
1566                         gmx_int64_t step, t_nrnb *nrnb, gmx_wallcycle_t wcycle,
1567                         gmx_localtop_t *top,
1568                         gmx_groups_t *groups,
1569                         matrix box, rvec x[], history_t *hist,
1570                         rvec f[],
1571                         tensor vir_force,
1572                         t_mdatoms *mdatoms,
1573                         gmx_enerdata_t *enerd, t_fcdata *fcd,
1574                         real *lambda, t_graph *graph,
1575                         t_forcerec *fr, gmx_vsite_t *vsite, rvec mu_tot,
1576                         double t, FILE *field, gmx_edsam_t ed,
1577                         gmx_bool bBornRadii,
1578                         int flags)
1579 {
1580     int        cg0, cg1, i, j;
1581     int        start, homenr;
1582     double     mu[2*DIM];
1583     gmx_bool   bSepDVDL, bStateChanged, bNS, bFillGrid, bCalcCGCM, bBS;
1584     gmx_bool   bDoLongRangeNS, bDoForces, bDoPotential, bSepLRF;
1585     gmx_bool   bDoAdressWF;
1586     matrix     boxs;
1587     rvec       vzero, box_diag;
1588     real       e, v, dvdlambda[efptNR];
1589     t_pbc      pbc;
1590     float      cycles_pme, cycles_force;
1591
1592     start  = 0;
1593     homenr = mdatoms->homenr;
1594
1595     bSepDVDL = (fr->bSepDVDL && do_per_step(step, inputrec->nstlog));
1596
1597     clear_mat(vir_force);
1598
1599     cg0 = 0;
1600     if (DOMAINDECOMP(cr))
1601     {
1602         cg1 = cr->dd->ncg_tot;
1603     }
1604     else
1605     {
1606         cg1 = top->cgs.nr;
1607     }
1608     if (fr->n_tpi > 0)
1609     {
1610         cg1--;
1611     }
1612
1613     bStateChanged  = (flags & GMX_FORCE_STATECHANGED);
1614     bNS            = (flags & GMX_FORCE_NS) && (fr->bAllvsAll == FALSE);
1615     /* Should we update the long-range neighborlists at this step? */
1616     bDoLongRangeNS = fr->bTwinRange && bNS;
1617     /* Should we perform the long-range nonbonded evaluation inside the neighborsearching? */
1618     bFillGrid      = (bNS && bStateChanged);
1619     bCalcCGCM      = (bFillGrid && !DOMAINDECOMP(cr));
1620     bDoForces      = (flags & GMX_FORCE_FORCES);
1621     bDoPotential   = (flags & GMX_FORCE_ENERGY);
1622     bSepLRF        = ((inputrec->nstcalclr > 1) && bDoForces &&
1623                       (flags & GMX_FORCE_SEPLRF) && (flags & GMX_FORCE_DO_LR));
1624
1625     /* should probably move this to the forcerec since it doesn't change */
1626     bDoAdressWF   = ((fr->adress_type != eAdressOff));
1627
1628     if (bStateChanged)
1629     {
1630         update_forcerec(fr, box);
1631
1632         if (NEED_MUTOT(*inputrec))
1633         {
1634             /* Calculate total (local) dipole moment in a temporary common array.
1635              * This makes it possible to sum them over nodes faster.
1636              */
1637             calc_mu(start, homenr,
1638                     x, mdatoms->chargeA, mdatoms->chargeB, mdatoms->nChargePerturbed,
1639                     mu, mu+DIM);
1640         }
1641     }
1642
1643     if (fr->ePBC != epbcNONE)
1644     {
1645         /* Compute shift vectors every step,
1646          * because of pressure coupling or box deformation!
1647          */
1648         if ((flags & GMX_FORCE_DYNAMICBOX) && bStateChanged)
1649         {
1650             calc_shifts(box, fr->shift_vec);
1651         }
1652
1653         if (bCalcCGCM)
1654         {
1655             put_charge_groups_in_box(fplog, cg0, cg1, fr->ePBC, box,
1656                                      &(top->cgs), x, fr->cg_cm);
1657             inc_nrnb(nrnb, eNR_CGCM, homenr);
1658             inc_nrnb(nrnb, eNR_RESETX, cg1-cg0);
1659         }
1660         else if (EI_ENERGY_MINIMIZATION(inputrec->eI) && graph)
1661         {
1662             unshift_self(graph, box, x);
1663         }
1664     }
1665     else if (bCalcCGCM)
1666     {
1667         calc_cgcm(fplog, cg0, cg1, &(top->cgs), x, fr->cg_cm);
1668         inc_nrnb(nrnb, eNR_CGCM, homenr);
1669     }
1670
1671     if (bCalcCGCM && gmx_debug_at)
1672     {
1673         pr_rvecs(debug, 0, "cgcm", fr->cg_cm, top->cgs.nr);
1674     }
1675
1676 #ifdef GMX_MPI
1677     if (!(cr->duty & DUTY_PME))
1678     {
1679         /* Send particle coordinates to the pme nodes.
1680          * Since this is only implemented for domain decomposition
1681          * and domain decomposition does not use the graph,
1682          * we do not need to worry about shifting.
1683          */
1684
1685         int pme_flags = 0;
1686
1687         wallcycle_start(wcycle, ewcPP_PMESENDX);
1688
1689         bBS = (inputrec->nwall == 2);
1690         if (bBS)
1691         {
1692             copy_mat(box, boxs);
1693             svmul(inputrec->wall_ewald_zfac, boxs[ZZ], boxs[ZZ]);
1694         }
1695
1696         if (EEL_PME(fr->eeltype))
1697         {
1698             pme_flags |= GMX_PME_DO_COULOMB;
1699         }
1700
1701         if (EVDW_PME(fr->vdwtype))
1702         {
1703             pme_flags |= GMX_PME_DO_LJ;
1704         }
1705
1706         gmx_pme_send_coordinates(cr, bBS ? boxs : box, x,
1707                                  mdatoms->nChargePerturbed, mdatoms->nTypePerturbed, lambda[efptCOUL], lambda[efptVDW],
1708                                  (flags & (GMX_FORCE_VIRIAL | GMX_FORCE_ENERGY)),
1709                                  pme_flags, step);
1710
1711         wallcycle_stop(wcycle, ewcPP_PMESENDX);
1712     }
1713 #endif /* GMX_MPI */
1714
1715     /* Communicate coordinates and sum dipole if necessary */
1716     if (DOMAINDECOMP(cr))
1717     {
1718         wallcycle_start(wcycle, ewcMOVEX);
1719         dd_move_x(cr->dd, box, x);
1720         wallcycle_stop(wcycle, ewcMOVEX);
1721     }
1722
1723     /* update adress weight beforehand */
1724     if (bStateChanged && bDoAdressWF)
1725     {
1726         /* need pbc for adress weight calculation with pbc_dx */
1727         set_pbc(&pbc, inputrec->ePBC, box);
1728         if (fr->adress_site == eAdressSITEcog)
1729         {
1730             update_adress_weights_cog(top->idef.iparams, top->idef.il, x, fr, mdatoms,
1731                                       inputrec->ePBC == epbcNONE ? NULL : &pbc);
1732         }
1733         else if (fr->adress_site == eAdressSITEcom)
1734         {
1735             update_adress_weights_com(fplog, cg0, cg1, &(top->cgs), x, fr, mdatoms,
1736                                       inputrec->ePBC == epbcNONE ? NULL : &pbc);
1737         }
1738         else if (fr->adress_site == eAdressSITEatomatom)
1739         {
1740             update_adress_weights_atom_per_atom(cg0, cg1, &(top->cgs), x, fr, mdatoms,
1741                                                 inputrec->ePBC == epbcNONE ? NULL : &pbc);
1742         }
1743         else
1744         {
1745             update_adress_weights_atom(cg0, cg1, &(top->cgs), x, fr, mdatoms,
1746                                        inputrec->ePBC == epbcNONE ? NULL : &pbc);
1747         }
1748     }
1749
1750     if (NEED_MUTOT(*inputrec))
1751     {
1752
1753         if (bStateChanged)
1754         {
1755             if (PAR(cr))
1756             {
1757                 gmx_sumd(2*DIM, mu, cr);
1758             }
1759             for (i = 0; i < 2; i++)
1760             {
1761                 for (j = 0; j < DIM; j++)
1762                 {
1763                     fr->mu_tot[i][j] = mu[i*DIM + j];
1764                 }
1765             }
1766         }
1767         if (fr->efep == efepNO)
1768         {
1769             copy_rvec(fr->mu_tot[0], mu_tot);
1770         }
1771         else
1772         {
1773             for (j = 0; j < DIM; j++)
1774             {
1775                 mu_tot[j] =
1776                     (1.0 - lambda[efptCOUL])*fr->mu_tot[0][j] + lambda[efptCOUL]*fr->mu_tot[1][j];
1777             }
1778         }
1779     }
1780
1781     /* Reset energies */
1782     reset_enerdata(fr, bNS, enerd, MASTER(cr));
1783     clear_rvecs(SHIFTS, fr->fshift);
1784
1785     if (bNS)
1786     {
1787         wallcycle_start(wcycle, ewcNS);
1788
1789         if (graph && bStateChanged)
1790         {
1791             /* Calculate intramolecular shift vectors to make molecules whole */
1792             mk_mshift(fplog, graph, fr->ePBC, box, x);
1793         }
1794
1795         /* Do the actual neighbour searching */
1796         ns(fplog, fr, box,
1797            groups, top, mdatoms,
1798            cr, nrnb, bFillGrid,
1799            bDoLongRangeNS);
1800
1801         wallcycle_stop(wcycle, ewcNS);
1802     }
1803
1804     if (inputrec->implicit_solvent && bNS)
1805     {
1806         make_gb_nblist(cr, inputrec->gb_algorithm,
1807                        x, box, fr, &top->idef, graph, fr->born);
1808     }
1809
1810     if (DOMAINDECOMP(cr) && !(cr->duty & DUTY_PME))
1811     {
1812         wallcycle_start(wcycle, ewcPPDURINGPME);
1813         dd_force_flop_start(cr->dd, nrnb);
1814     }
1815
1816     if (inputrec->bRot)
1817     {
1818         /* Enforced rotation has its own cycle counter that starts after the collective
1819          * coordinates have been communicated. It is added to ddCyclF to allow
1820          * for proper load-balancing */
1821         wallcycle_start(wcycle, ewcROT);
1822         do_rotation(cr, inputrec, box, x, t, step, wcycle, bNS);
1823         wallcycle_stop(wcycle, ewcROT);
1824     }
1825
1826     /* Start the force cycle counter.
1827      * This counter is stopped in do_forcelow_level.
1828      * No parallel communication should occur while this counter is running,
1829      * since that will interfere with the dynamic load balancing.
1830      */
1831     wallcycle_start(wcycle, ewcFORCE);
1832
1833     if (bDoForces)
1834     {
1835         /* Reset forces for which the virial is calculated separately:
1836          * PME/Ewald forces if necessary */
1837         if (fr->bF_NoVirSum)
1838         {
1839             if (flags & GMX_FORCE_VIRIAL)
1840             {
1841                 fr->f_novirsum = fr->f_novirsum_alloc;
1842                 if (fr->bDomDec)
1843                 {
1844                     clear_rvecs(fr->f_novirsum_n, fr->f_novirsum);
1845                 }
1846                 else
1847                 {
1848                     clear_rvecs(homenr, fr->f_novirsum+start);
1849                 }
1850             }
1851             else
1852             {
1853                 /* We are not calculating the pressure so we do not need
1854                  * a separate array for forces that do not contribute
1855                  * to the pressure.
1856                  */
1857                 fr->f_novirsum = f;
1858             }
1859         }
1860
1861         /* Clear the short- and long-range forces */
1862         clear_rvecs(fr->natoms_force_constr, f);
1863         if (bSepLRF && do_per_step(step, inputrec->nstcalclr))
1864         {
1865             clear_rvecs(fr->natoms_force_constr, fr->f_twin);
1866         }
1867
1868         clear_rvec(fr->vir_diag_posres);
1869     }
1870     if (inputrec->ePull == epullCONSTRAINT)
1871     {
1872         clear_pull_forces(inputrec->pull);
1873     }
1874
1875     /* update QMMMrec, if necessary */
1876     if (fr->bQMMM)
1877     {
1878         update_QMMMrec(cr, fr, x, mdatoms, box, top);
1879     }
1880
1881     if ((flags & GMX_FORCE_BONDED) && top->idef.il[F_POSRES].nr > 0)
1882     {
1883         posres_wrapper(fplog, flags, bSepDVDL, inputrec, nrnb, top, box, x,
1884                        enerd, lambda, fr);
1885     }
1886
1887     if ((flags & GMX_FORCE_BONDED) && top->idef.il[F_FBPOSRES].nr > 0)
1888     {
1889         fbposres_wrapper(inputrec, nrnb, top, box, x, enerd, fr);
1890     }
1891
1892     /* Compute the bonded and non-bonded energies and optionally forces */
1893     do_force_lowlevel(fplog, step, fr, inputrec, &(top->idef),
1894                       cr, nrnb, wcycle, mdatoms,
1895                       x, hist, f, bSepLRF ? fr->f_twin : f, enerd, fcd, top, fr->born,
1896                       &(top->atomtypes), bBornRadii, box,
1897                       inputrec->fepvals, lambda,
1898                       graph, &(top->excls), fr->mu_tot,
1899                       flags,
1900                       &cycles_pme);
1901
1902     if (bSepLRF)
1903     {
1904         if (do_per_step(step, inputrec->nstcalclr))
1905         {
1906             /* Add the long range forces to the short range forces */
1907             for (i = 0; i < fr->natoms_force_constr; i++)
1908             {
1909                 rvec_add(fr->f_twin[i], f[i], f[i]);
1910             }
1911         }
1912     }
1913
1914     cycles_force = wallcycle_stop(wcycle, ewcFORCE);
1915
1916     if (ed)
1917     {
1918         do_flood(cr, inputrec, x, f, ed, box, step, bNS);
1919     }
1920
1921     if (DOMAINDECOMP(cr))
1922     {
1923         dd_force_flop_stop(cr->dd, nrnb);
1924         if (wcycle)
1925         {
1926             dd_cycles_add(cr->dd, cycles_force-cycles_pme, ddCyclF);
1927         }
1928     }
1929
1930     if (bDoForces)
1931     {
1932         if (IR_ELEC_FIELD(*inputrec))
1933         {
1934             /* Compute forces due to electric field */
1935             calc_f_el(MASTER(cr) ? field : NULL,
1936                       start, homenr, mdatoms->chargeA, fr->f_novirsum,
1937                       inputrec->ex, inputrec->et, t);
1938         }
1939
1940         if (bDoAdressWF && fr->adress_icor == eAdressICThermoForce)
1941         {
1942             /* Compute thermodynamic force in hybrid AdResS region */
1943             adress_thermo_force(start, homenr, &(top->cgs), x, fr->f_novirsum, fr, mdatoms,
1944                                 inputrec->ePBC == epbcNONE ? NULL : &pbc);
1945         }
1946
1947         /* Communicate the forces */
1948         if (DOMAINDECOMP(cr))
1949         {
1950             wallcycle_start(wcycle, ewcMOVEF);
1951             dd_move_f(cr->dd, f, fr->fshift);
1952             /* Do we need to communicate the separate force array
1953              * for terms that do not contribute to the single sum virial?
1954              * Position restraints and electric fields do not introduce
1955              * inter-cg forces, only full electrostatics methods do.
1956              * When we do not calculate the virial, fr->f_novirsum = f,
1957              * so we have already communicated these forces.
1958              */
1959             if (EEL_FULL(fr->eeltype) && cr->dd->n_intercg_excl &&
1960                 (flags & GMX_FORCE_VIRIAL))
1961             {
1962                 dd_move_f(cr->dd, fr->f_novirsum, NULL);
1963             }
1964             if (bSepLRF)
1965             {
1966                 /* We should not update the shift forces here,
1967                  * since f_twin is already included in f.
1968                  */
1969                 dd_move_f(cr->dd, fr->f_twin, NULL);
1970             }
1971             wallcycle_stop(wcycle, ewcMOVEF);
1972         }
1973
1974         /* If we have NoVirSum forces, but we do not calculate the virial,
1975          * we sum fr->f_novirum=f later.
1976          */
1977         if (vsite && !(fr->bF_NoVirSum && !(flags & GMX_FORCE_VIRIAL)))
1978         {
1979             wallcycle_start(wcycle, ewcVSITESPREAD);
1980             spread_vsite_f(vsite, x, f, fr->fshift, FALSE, NULL, nrnb,
1981                            &top->idef, fr->ePBC, fr->bMolPBC, graph, box, cr);
1982             wallcycle_stop(wcycle, ewcVSITESPREAD);
1983
1984             if (bSepLRF)
1985             {
1986                 wallcycle_start(wcycle, ewcVSITESPREAD);
1987                 spread_vsite_f(vsite, x, fr->f_twin, NULL, FALSE, NULL,
1988                                nrnb,
1989                                &top->idef, fr->ePBC, fr->bMolPBC, graph, box, cr);
1990                 wallcycle_stop(wcycle, ewcVSITESPREAD);
1991             }
1992         }
1993
1994         if (flags & GMX_FORCE_VIRIAL)
1995         {
1996             /* Calculation of the virial must be done after vsites! */
1997             calc_virial(0, mdatoms->homenr, x, f,
1998                         vir_force, graph, box, nrnb, fr, inputrec->ePBC);
1999         }
2000     }
2001
2002     if (inputrec->ePull == epullUMBRELLA || inputrec->ePull == epullCONST_F)
2003     {
2004         pull_potential_wrapper(fplog, bSepDVDL, cr, inputrec, box, x,
2005                                f, vir_force, mdatoms, enerd, lambda, t);
2006     }
2007
2008     /* Add the forces from enforced rotation potentials (if any) */
2009     if (inputrec->bRot)
2010     {
2011         wallcycle_start(wcycle, ewcROTadd);
2012         enerd->term[F_COM_PULL] += add_rot_forces(inputrec->rot, f, cr, step, t);
2013         wallcycle_stop(wcycle, ewcROTadd);
2014     }
2015
2016     /* Add forces from interactive molecular dynamics (IMD), if bIMD == TRUE. */
2017     IMD_apply_forces(inputrec->bIMD, inputrec->imd, cr, f, wcycle);
2018
2019     if (PAR(cr) && !(cr->duty & DUTY_PME))
2020     {
2021         /* In case of node-splitting, the PP nodes receive the long-range
2022          * forces, virial and energy from the PME nodes here.
2023          */
2024         pme_receive_force_ener(fplog, bSepDVDL, cr, wcycle, enerd, fr);
2025     }
2026
2027     if (bDoForces)
2028     {
2029         post_process_forces(cr, step, nrnb, wcycle,
2030                             top, box, x, f, vir_force, mdatoms, graph, fr, vsite,
2031                             flags);
2032     }
2033
2034     /* Sum the potential energy terms from group contributions */
2035     sum_epot(&(enerd->grpp), enerd->term);
2036 }
2037
2038 void do_force(FILE *fplog, t_commrec *cr,
2039               t_inputrec *inputrec,
2040               gmx_int64_t step, t_nrnb *nrnb, gmx_wallcycle_t wcycle,
2041               gmx_localtop_t *top,
2042               gmx_groups_t *groups,
2043               matrix box, rvec x[], history_t *hist,
2044               rvec f[],
2045               tensor vir_force,
2046               t_mdatoms *mdatoms,
2047               gmx_enerdata_t *enerd, t_fcdata *fcd,
2048               real *lambda, t_graph *graph,
2049               t_forcerec *fr,
2050               gmx_vsite_t *vsite, rvec mu_tot,
2051               double t, FILE *field, gmx_edsam_t ed,
2052               gmx_bool bBornRadii,
2053               int flags)
2054 {
2055     /* modify force flag if not doing nonbonded */
2056     if (!fr->bNonbonded)
2057     {
2058         flags &= ~GMX_FORCE_NONBONDED;
2059     }
2060
2061     switch (inputrec->cutoff_scheme)
2062     {
2063         case ecutsVERLET:
2064             do_force_cutsVERLET(fplog, cr, inputrec,
2065                                 step, nrnb, wcycle,
2066                                 top,
2067                                 groups,
2068                                 box, x, hist,
2069                                 f, vir_force,
2070                                 mdatoms,
2071                                 enerd, fcd,
2072                                 lambda, graph,
2073                                 fr, fr->ic,
2074                                 vsite, mu_tot,
2075                                 t, field, ed,
2076                                 bBornRadii,
2077                                 flags);
2078             break;
2079         case ecutsGROUP:
2080             do_force_cutsGROUP(fplog, cr, inputrec,
2081                                step, nrnb, wcycle,
2082                                top,
2083                                groups,
2084                                box, x, hist,
2085                                f, vir_force,
2086                                mdatoms,
2087                                enerd, fcd,
2088                                lambda, graph,
2089                                fr, vsite, mu_tot,
2090                                t, field, ed,
2091                                bBornRadii,
2092                                flags);
2093             break;
2094         default:
2095             gmx_incons("Invalid cut-off scheme passed!");
2096     }
2097 }
2098
2099
2100 void do_constrain_first(FILE *fplog, gmx_constr_t constr,
2101                         t_inputrec *ir, t_mdatoms *md,
2102                         t_state *state, t_commrec *cr, t_nrnb *nrnb,
2103                         t_forcerec *fr, gmx_localtop_t *top)
2104 {
2105     int             i, m, start, end;
2106     gmx_int64_t     step;
2107     real            dt = ir->delta_t;
2108     real            dvdl_dum;
2109     rvec           *savex;
2110
2111     snew(savex, state->natoms);
2112
2113     start = 0;
2114     end   = md->homenr;
2115
2116     if (debug)
2117     {
2118         fprintf(debug, "vcm: start=%d, homenr=%d, end=%d\n",
2119                 start, md->homenr, end);
2120     }
2121     /* Do a first constrain to reset particles... */
2122     step = ir->init_step;
2123     if (fplog)
2124     {
2125         char buf[STEPSTRSIZE];
2126         fprintf(fplog, "\nConstraining the starting coordinates (step %s)\n",
2127                 gmx_step_str(step, buf));
2128     }
2129     dvdl_dum = 0;
2130
2131     /* constrain the current position */
2132     constrain(NULL, TRUE, FALSE, constr, &(top->idef),
2133               ir, NULL, cr, step, 0, md,
2134               state->x, state->x, NULL,
2135               fr->bMolPBC, state->box,
2136               state->lambda[efptBONDED], &dvdl_dum,
2137               NULL, NULL, nrnb, econqCoord,
2138               ir->epc == epcMTTK, state->veta, state->veta);
2139     if (EI_VV(ir->eI))
2140     {
2141         /* constrain the inital velocity, and save it */
2142         /* also may be useful if we need the ekin from the halfstep for velocity verlet */
2143         /* might not yet treat veta correctly */
2144         constrain(NULL, TRUE, FALSE, constr, &(top->idef),
2145                   ir, NULL, cr, step, 0, md,
2146                   state->x, state->v, state->v,
2147                   fr->bMolPBC, state->box,
2148                   state->lambda[efptBONDED], &dvdl_dum,
2149                   NULL, NULL, nrnb, econqVeloc,
2150                   ir->epc == epcMTTK, state->veta, state->veta);
2151     }
2152     /* constrain the inital velocities at t-dt/2 */
2153     if (EI_STATE_VELOCITY(ir->eI) && ir->eI != eiVV)
2154     {
2155         for (i = start; (i < end); i++)
2156         {
2157             for (m = 0; (m < DIM); m++)
2158             {
2159                 /* Reverse the velocity */
2160                 state->v[i][m] = -state->v[i][m];
2161                 /* Store the position at t-dt in buf */
2162                 savex[i][m] = state->x[i][m] + dt*state->v[i][m];
2163             }
2164         }
2165         /* Shake the positions at t=-dt with the positions at t=0
2166          * as reference coordinates.
2167          */
2168         if (fplog)
2169         {
2170             char buf[STEPSTRSIZE];
2171             fprintf(fplog, "\nConstraining the coordinates at t0-dt (step %s)\n",
2172                     gmx_step_str(step, buf));
2173         }
2174         dvdl_dum = 0;
2175         constrain(NULL, TRUE, FALSE, constr, &(top->idef),
2176                   ir, NULL, cr, step, -1, md,
2177                   state->x, savex, NULL,
2178                   fr->bMolPBC, state->box,
2179                   state->lambda[efptBONDED], &dvdl_dum,
2180                   state->v, NULL, nrnb, econqCoord,
2181                   ir->epc == epcMTTK, state->veta, state->veta);
2182
2183         for (i = start; i < end; i++)
2184         {
2185             for (m = 0; m < DIM; m++)
2186             {
2187                 /* Re-reverse the velocities */
2188                 state->v[i][m] = -state->v[i][m];
2189             }
2190         }
2191     }
2192     sfree(savex);
2193 }
2194
2195
2196 static void
2197 integrate_table(real vdwtab[], real scale, int offstart, int rstart, int rend,
2198                 double *enerout, double *virout)
2199 {
2200     double enersum, virsum;
2201     double invscale, invscale2, invscale3;
2202     double r, ea, eb, ec, pa, pb, pc, pd;
2203     double y0, f, g, h;
2204     int    ri, offset, tabfactor;
2205
2206     invscale  = 1.0/scale;
2207     invscale2 = invscale*invscale;
2208     invscale3 = invscale*invscale2;
2209
2210     /* Following summation derived from cubic spline definition,
2211      * Numerical Recipies in C, second edition, p. 113-116.  Exact for
2212      * the cubic spline.  We first calculate the negative of the
2213      * energy from rvdw to rvdw_switch, assuming that g(r)=1, and then
2214      * add the more standard, abrupt cutoff correction to that result,
2215      * yielding the long-range correction for a switched function.  We
2216      * perform both the pressure and energy loops at the same time for
2217      * simplicity, as the computational cost is low. */
2218
2219     if (offstart == 0)
2220     {
2221         /* Since the dispersion table has been scaled down a factor
2222          * 6.0 and the repulsion a factor 12.0 to compensate for the
2223          * c6/c12 parameters inside nbfp[] being scaled up (to save
2224          * flops in kernels), we need to correct for this.
2225          */
2226         tabfactor = 6.0;
2227     }
2228     else
2229     {
2230         tabfactor = 12.0;
2231     }
2232
2233     enersum = 0.0;
2234     virsum  = 0.0;
2235     for (ri = rstart; ri < rend; ++ri)
2236     {
2237         r  = ri*invscale;
2238         ea = invscale3;
2239         eb = 2.0*invscale2*r;
2240         ec = invscale*r*r;
2241
2242         pa = invscale3;
2243         pb = 3.0*invscale2*r;
2244         pc = 3.0*invscale*r*r;
2245         pd = r*r*r;
2246
2247         /* this "8" is from the packing in the vdwtab array - perhaps
2248            should be defined? */
2249
2250         offset = 8*ri + offstart;
2251         y0     = vdwtab[offset];
2252         f      = vdwtab[offset+1];
2253         g      = vdwtab[offset+2];
2254         h      = vdwtab[offset+3];
2255
2256         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);
2257         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);
2258     }
2259     *enerout = 4.0*M_PI*enersum*tabfactor;
2260     *virout  = 4.0*M_PI*virsum*tabfactor;
2261 }
2262
2263 void calc_enervirdiff(FILE *fplog, int eDispCorr, t_forcerec *fr)
2264 {
2265     double   eners[2], virs[2], enersum, virsum, y0, f, g, h;
2266     double   r0, r1, r, rc3, rc9, ea, eb, ec, pa, pb, pc, pd;
2267     double   invscale, invscale2, invscale3;
2268     int      ri0, ri1, ri, i, offstart, offset;
2269     real     scale, *vdwtab, tabfactor, tmp;
2270
2271     fr->enershiftsix    = 0;
2272     fr->enershifttwelve = 0;
2273     fr->enerdiffsix     = 0;
2274     fr->enerdifftwelve  = 0;
2275     fr->virdiffsix      = 0;
2276     fr->virdifftwelve   = 0;
2277
2278     if (eDispCorr != edispcNO)
2279     {
2280         for (i = 0; i < 2; i++)
2281         {
2282             eners[i] = 0;
2283             virs[i]  = 0;
2284         }
2285         if ((fr->vdw_modifier == eintmodPOTSHIFT) ||
2286             (fr->vdw_modifier == eintmodPOTSWITCH) ||
2287             (fr->vdw_modifier == eintmodFORCESWITCH) ||
2288             (fr->vdwtype == evdwSHIFT) ||
2289             (fr->vdwtype == evdwSWITCH))
2290         {
2291             if (((fr->vdw_modifier == eintmodPOTSWITCH) ||
2292                  (fr->vdw_modifier == eintmodFORCESWITCH) ||
2293                  (fr->vdwtype == evdwSWITCH)) && fr->rvdw_switch == 0)
2294             {
2295                 gmx_fatal(FARGS,
2296                           "With dispersion correction rvdw-switch can not be zero "
2297                           "for vdw-type = %s", evdw_names[fr->vdwtype]);
2298             }
2299
2300             scale  = fr->nblists[0].table_vdw.scale;
2301             vdwtab = fr->nblists[0].table_vdw.data;
2302
2303             /* Round the cut-offs to exact table values for precision */
2304             ri0  = floor(fr->rvdw_switch*scale);
2305             ri1  = ceil(fr->rvdw*scale);
2306
2307             /* The code below has some support for handling force-switching, i.e.
2308              * when the force (instead of potential) is switched over a limited
2309              * region. This leads to a constant shift in the potential inside the
2310              * switching region, which we can handle by adding a constant energy
2311              * term in the force-switch case just like when we do potential-shift.
2312              *
2313              * For now this is not enabled, but to keep the functionality in the
2314              * code we check separately for switch and shift. When we do force-switch
2315              * the shifting point is rvdw_switch, while it is the cutoff when we
2316              * have a classical potential-shift.
2317              *
2318              * For a pure potential-shift the potential has a constant shift
2319              * all the way out to the cutoff, and that is it. For other forms
2320              * we need to calculate the constant shift up to the point where we
2321              * start modifying the potential.
2322              */
2323             ri0  = (fr->vdw_modifier == eintmodPOTSHIFT) ? ri1 : ri0;
2324
2325             r0   = ri0/scale;
2326             r1   = ri1/scale;
2327             rc3  = r0*r0*r0;
2328             rc9  = rc3*rc3*rc3;
2329
2330             if ((fr->vdw_modifier == eintmodFORCESWITCH) ||
2331                 (fr->vdwtype == evdwSHIFT))
2332             {
2333                 /* Determine the constant energy shift below rvdw_switch.
2334                  * Table has a scale factor since we have scaled it down to compensate
2335                  * for scaling-up c6/c12 with the derivative factors to save flops in analytical kernels.
2336                  */
2337                 fr->enershiftsix    = (real)(-1.0/(rc3*rc3)) - 6.0*vdwtab[8*ri0];
2338                 fr->enershifttwelve = (real)( 1.0/(rc9*rc3)) - 12.0*vdwtab[8*ri0 + 4];
2339             }
2340             else if (fr->vdw_modifier == eintmodPOTSHIFT)
2341             {
2342                 fr->enershiftsix    = (real)(-1.0/(rc3*rc3));
2343                 fr->enershifttwelve = (real)( 1.0/(rc9*rc3));
2344             }
2345
2346             /* Add the constant part from 0 to rvdw_switch.
2347              * This integration from 0 to rvdw_switch overcounts the number
2348              * of interactions by 1, as it also counts the self interaction.
2349              * We will correct for this later.
2350              */
2351             eners[0] += 4.0*M_PI*fr->enershiftsix*rc3/3.0;
2352             eners[1] += 4.0*M_PI*fr->enershifttwelve*rc3/3.0;
2353
2354             /* Calculate the contribution in the range [r0,r1] where we
2355              * modify the potential. For a pure potential-shift modifier we will
2356              * have ri0==ri1, and there will not be any contribution here.
2357              */
2358             for (i = 0; i < 2; i++)
2359             {
2360                 enersum = 0;
2361                 virsum  = 0;
2362                 integrate_table(vdwtab, scale, (i == 0 ? 0 : 4), ri0, ri1, &enersum, &virsum);
2363                 eners[i] -= enersum;
2364                 virs[i]  -= virsum;
2365             }
2366
2367             /* Alright: Above we compensated by REMOVING the parts outside r0
2368              * corresponding to the ideal VdW 1/r6 and /r12 potentials.
2369              *
2370              * Regardless of whether r0 is the point where we start switching,
2371              * or the cutoff where we calculated the constant shift, we include
2372              * all the parts we are missing out to infinity from r0 by
2373              * calculating the analytical dispersion correction.
2374              */
2375             eners[0] += -4.0*M_PI/(3.0*rc3);
2376             eners[1] +=  4.0*M_PI/(9.0*rc9);
2377             virs[0]  +=  8.0*M_PI/rc3;
2378             virs[1]  += -16.0*M_PI/(3.0*rc9);
2379         }
2380         else if (fr->vdwtype == evdwCUT ||
2381                  EVDW_PME(fr->vdwtype) ||
2382                  fr->vdwtype == evdwUSER)
2383         {
2384             if (fr->vdwtype == evdwUSER && fplog)
2385             {
2386                 fprintf(fplog,
2387                         "WARNING: using dispersion correction with user tables\n");
2388             }
2389
2390             /* Note that with LJ-PME, the dispersion correction is multiplied
2391              * by the difference between the actual C6 and the value of C6
2392              * that would produce the combination rule.
2393              * This means the normal energy and virial difference formulas
2394              * can be used here.
2395              */
2396
2397             rc3  = fr->rvdw*fr->rvdw*fr->rvdw;
2398             rc9  = rc3*rc3*rc3;
2399             /* Contribution beyond the cut-off */
2400             eners[0] += -4.0*M_PI/(3.0*rc3);
2401             eners[1] +=  4.0*M_PI/(9.0*rc9);
2402             if (fr->vdw_modifier == eintmodPOTSHIFT)
2403             {
2404                 /* Contribution within the cut-off */
2405                 eners[0] += -4.0*M_PI/(3.0*rc3);
2406                 eners[1] +=  4.0*M_PI/(3.0*rc9);
2407             }
2408             /* Contribution beyond the cut-off */
2409             virs[0]  +=  8.0*M_PI/rc3;
2410             virs[1]  += -16.0*M_PI/(3.0*rc9);
2411         }
2412         else
2413         {
2414             gmx_fatal(FARGS,
2415                       "Dispersion correction is not implemented for vdw-type = %s",
2416                       evdw_names[fr->vdwtype]);
2417         }
2418
2419         /* When we deprecate the group kernels the code below can go too */
2420         if (fr->vdwtype == evdwPME && fr->cutoff_scheme == ecutsGROUP)
2421         {
2422             /* Calculate self-interaction coefficient (assuming that
2423              * the reciprocal-space contribution is constant in the
2424              * region that contributes to the self-interaction).
2425              */
2426             fr->enershiftsix = pow(fr->ewaldcoeff_lj, 6) / 6.0;
2427
2428             eners[0] += -pow(sqrt(M_PI)*fr->ewaldcoeff_lj, 3)/3.0;
2429             virs[0]  +=  pow(sqrt(M_PI)*fr->ewaldcoeff_lj, 3);
2430         }
2431
2432         fr->enerdiffsix    = eners[0];
2433         fr->enerdifftwelve = eners[1];
2434         /* The 0.5 is due to the Gromacs definition of the virial */
2435         fr->virdiffsix     = 0.5*virs[0];
2436         fr->virdifftwelve  = 0.5*virs[1];
2437     }
2438 }
2439
2440 void calc_dispcorr(FILE *fplog, t_inputrec *ir, t_forcerec *fr,
2441                    gmx_int64_t step, int natoms,
2442                    matrix box, real lambda, tensor pres, tensor virial,
2443                    real *prescorr, real *enercorr, real *dvdlcorr)
2444 {
2445     gmx_bool bCorrAll, bCorrPres;
2446     real     dvdlambda, invvol, dens, ninter, avcsix, avctwelve, enerdiff, svir = 0, spres = 0;
2447     int      m;
2448
2449     *prescorr = 0;
2450     *enercorr = 0;
2451     *dvdlcorr = 0;
2452
2453     clear_mat(virial);
2454     clear_mat(pres);
2455
2456     if (ir->eDispCorr != edispcNO)
2457     {
2458         bCorrAll  = (ir->eDispCorr == edispcAllEner ||
2459                      ir->eDispCorr == edispcAllEnerPres);
2460         bCorrPres = (ir->eDispCorr == edispcEnerPres ||
2461                      ir->eDispCorr == edispcAllEnerPres);
2462
2463         invvol = 1/det(box);
2464         if (fr->n_tpi)
2465         {
2466             /* Only correct for the interactions with the inserted molecule */
2467             dens   = (natoms - fr->n_tpi)*invvol;
2468             ninter = fr->n_tpi;
2469         }
2470         else
2471         {
2472             dens   = natoms*invvol;
2473             ninter = 0.5*natoms;
2474         }
2475
2476         if (ir->efep == efepNO)
2477         {
2478             avcsix    = fr->avcsix[0];
2479             avctwelve = fr->avctwelve[0];
2480         }
2481         else
2482         {
2483             avcsix    = (1 - lambda)*fr->avcsix[0]    + lambda*fr->avcsix[1];
2484             avctwelve = (1 - lambda)*fr->avctwelve[0] + lambda*fr->avctwelve[1];
2485         }
2486
2487         enerdiff   = ninter*(dens*fr->enerdiffsix - fr->enershiftsix);
2488         *enercorr += avcsix*enerdiff;
2489         dvdlambda  = 0.0;
2490         if (ir->efep != efepNO)
2491         {
2492             dvdlambda += (fr->avcsix[1] - fr->avcsix[0])*enerdiff;
2493         }
2494         if (bCorrAll)
2495         {
2496             enerdiff   = ninter*(dens*fr->enerdifftwelve - fr->enershifttwelve);
2497             *enercorr += avctwelve*enerdiff;
2498             if (fr->efep != efepNO)
2499             {
2500                 dvdlambda += (fr->avctwelve[1] - fr->avctwelve[0])*enerdiff;
2501             }
2502         }
2503
2504         if (bCorrPres)
2505         {
2506             svir = ninter*dens*avcsix*fr->virdiffsix/3.0;
2507             if (ir->eDispCorr == edispcAllEnerPres)
2508             {
2509                 svir += ninter*dens*avctwelve*fr->virdifftwelve/3.0;
2510             }
2511             /* The factor 2 is because of the Gromacs virial definition */
2512             spres = -2.0*invvol*svir*PRESFAC;
2513
2514             for (m = 0; m < DIM; m++)
2515             {
2516                 virial[m][m] += svir;
2517                 pres[m][m]   += spres;
2518             }
2519             *prescorr += spres;
2520         }
2521
2522         /* Can't currently control when it prints, for now, just print when degugging */
2523         if (debug)
2524         {
2525             if (bCorrAll)
2526             {
2527                 fprintf(debug, "Long Range LJ corr.: <C6> %10.4e, <C12> %10.4e\n",
2528                         avcsix, avctwelve);
2529             }
2530             if (bCorrPres)
2531             {
2532                 fprintf(debug,
2533                         "Long Range LJ corr.: Epot %10g, Pres: %10g, Vir: %10g\n",
2534                         *enercorr, spres, svir);
2535             }
2536             else
2537             {
2538                 fprintf(debug, "Long Range LJ corr.: Epot %10g\n", *enercorr);
2539             }
2540         }
2541
2542         if (fr->bSepDVDL && do_per_step(step, ir->nstlog))
2543         {
2544             gmx_print_sepdvdl(fplog, "Dispersion correction", *enercorr, dvdlambda);
2545         }
2546         if (fr->efep != efepNO)
2547         {
2548             *dvdlcorr += dvdlambda;
2549         }
2550     }
2551 }
2552
2553 void do_pbc_first(FILE *fplog, matrix box, t_forcerec *fr,
2554                   t_graph *graph, rvec x[])
2555 {
2556     if (fplog)
2557     {
2558         fprintf(fplog, "Removing pbc first time\n");
2559     }
2560     calc_shifts(box, fr->shift_vec);
2561     if (graph)
2562     {
2563         mk_mshift(fplog, graph, fr->ePBC, box, x);
2564         if (gmx_debug_at)
2565         {
2566             p_graph(debug, "do_pbc_first 1", graph);
2567         }
2568         shift_self(graph, box, x);
2569         /* By doing an extra mk_mshift the molecules that are broken
2570          * because they were e.g. imported from another software
2571          * will be made whole again. Such are the healing powers
2572          * of GROMACS.
2573          */
2574         mk_mshift(fplog, graph, fr->ePBC, box, x);
2575         if (gmx_debug_at)
2576         {
2577             p_graph(debug, "do_pbc_first 2", graph);
2578         }
2579     }
2580     if (fplog)
2581     {
2582         fprintf(fplog, "Done rmpbc\n");
2583     }
2584 }
2585
2586 static void low_do_pbc_mtop(FILE *fplog, int ePBC, matrix box,
2587                             gmx_mtop_t *mtop, rvec x[],
2588                             gmx_bool bFirst)
2589 {
2590     t_graph        *graph;
2591     int             mb, as, mol;
2592     gmx_molblock_t *molb;
2593
2594     if (bFirst && fplog)
2595     {
2596         fprintf(fplog, "Removing pbc first time\n");
2597     }
2598
2599     snew(graph, 1);
2600     as = 0;
2601     for (mb = 0; mb < mtop->nmolblock; mb++)
2602     {
2603         molb = &mtop->molblock[mb];
2604         if (molb->natoms_mol == 1 ||
2605             (!bFirst && mtop->moltype[molb->type].cgs.nr == 1))
2606         {
2607             /* Just one atom or charge group in the molecule, no PBC required */
2608             as += molb->nmol*molb->natoms_mol;
2609         }
2610         else
2611         {
2612             /* Pass NULL iso fplog to avoid graph prints for each molecule type */
2613             mk_graph_ilist(NULL, mtop->moltype[molb->type].ilist,
2614                            0, molb->natoms_mol, FALSE, FALSE, graph);
2615
2616             for (mol = 0; mol < molb->nmol; mol++)
2617             {
2618                 mk_mshift(fplog, graph, ePBC, box, x+as);
2619
2620                 shift_self(graph, box, x+as);
2621                 /* The molecule is whole now.
2622                  * We don't need the second mk_mshift call as in do_pbc_first,
2623                  * since we no longer need this graph.
2624                  */
2625
2626                 as += molb->natoms_mol;
2627             }
2628             done_graph(graph);
2629         }
2630     }
2631     sfree(graph);
2632 }
2633
2634 void do_pbc_first_mtop(FILE *fplog, int ePBC, matrix box,
2635                        gmx_mtop_t *mtop, rvec x[])
2636 {
2637     low_do_pbc_mtop(fplog, ePBC, box, mtop, x, TRUE);
2638 }
2639
2640 void do_pbc_mtop(FILE *fplog, int ePBC, matrix box,
2641                  gmx_mtop_t *mtop, rvec x[])
2642 {
2643     low_do_pbc_mtop(fplog, ePBC, box, mtop, x, FALSE);
2644 }
2645
2646 void finish_run(FILE *fplog, t_commrec *cr,
2647                 t_inputrec *inputrec,
2648                 t_nrnb nrnb[], gmx_wallcycle_t wcycle,
2649                 gmx_walltime_accounting_t walltime_accounting,
2650                 wallclock_gpu_t *gputimes,
2651                 gmx_bool bWriteStat)
2652 {
2653     int     i, j;
2654     t_nrnb *nrnb_tot = NULL;
2655     real    delta_t;
2656     double  nbfs, mflop;
2657     double  elapsed_time,
2658             elapsed_time_over_all_ranks,
2659             elapsed_time_over_all_threads,
2660             elapsed_time_over_all_threads_over_all_ranks;
2661     wallcycle_sum(cr, wcycle);
2662
2663     if (cr->nnodes > 1)
2664     {
2665         snew(nrnb_tot, 1);
2666 #ifdef GMX_MPI
2667         MPI_Allreduce(nrnb->n, nrnb_tot->n, eNRNB, MPI_DOUBLE, MPI_SUM,
2668                       cr->mpi_comm_mysim);
2669 #endif
2670     }
2671     else
2672     {
2673         nrnb_tot = nrnb;
2674     }
2675
2676     elapsed_time                                 = walltime_accounting_get_elapsed_time(walltime_accounting);
2677     elapsed_time_over_all_ranks                  = elapsed_time;
2678     elapsed_time_over_all_threads                = walltime_accounting_get_elapsed_time_over_all_threads(walltime_accounting);
2679     elapsed_time_over_all_threads_over_all_ranks = elapsed_time_over_all_threads;
2680 #ifdef GMX_MPI
2681     if (cr->nnodes > 1)
2682     {
2683         /* reduce elapsed_time over all MPI ranks in the current simulation */
2684         MPI_Allreduce(&elapsed_time,
2685                       &elapsed_time_over_all_ranks,
2686                       1, MPI_DOUBLE, MPI_SUM,
2687                       cr->mpi_comm_mysim);
2688         elapsed_time_over_all_ranks /= cr->nnodes;
2689         /* Reduce elapsed_time_over_all_threads over all MPI ranks in the
2690          * current simulation. */
2691         MPI_Allreduce(&elapsed_time_over_all_threads,
2692                       &elapsed_time_over_all_threads_over_all_ranks,
2693                       1, MPI_DOUBLE, MPI_SUM,
2694                       cr->mpi_comm_mysim);
2695     }
2696 #endif
2697
2698     if (SIMMASTER(cr))
2699     {
2700         print_flop(fplog, nrnb_tot, &nbfs, &mflop);
2701     }
2702     if (cr->nnodes > 1)
2703     {
2704         sfree(nrnb_tot);
2705     }
2706
2707     if ((cr->duty & DUTY_PP) && DOMAINDECOMP(cr))
2708     {
2709         print_dd_statistics(cr, inputrec, fplog);
2710     }
2711
2712     if (SIMMASTER(cr))
2713     {
2714         wallcycle_print(fplog, cr->nnodes, cr->npmenodes,
2715                         elapsed_time_over_all_ranks,
2716                         wcycle, gputimes);
2717
2718         if (EI_DYNAMICS(inputrec->eI))
2719         {
2720             delta_t = inputrec->delta_t;
2721         }
2722         else
2723         {
2724             delta_t = 0;
2725         }
2726
2727         if (fplog)
2728         {
2729             print_perf(fplog, elapsed_time_over_all_threads_over_all_ranks,
2730                        elapsed_time_over_all_ranks,
2731                        walltime_accounting_get_nsteps_done(walltime_accounting),
2732                        delta_t, nbfs, mflop);
2733         }
2734         if (bWriteStat)
2735         {
2736             print_perf(stderr, elapsed_time_over_all_threads_over_all_ranks,
2737                        elapsed_time_over_all_ranks,
2738                        walltime_accounting_get_nsteps_done(walltime_accounting),
2739                        delta_t, nbfs, mflop);
2740         }
2741     }
2742 }
2743
2744 extern void initialize_lambdas(FILE *fplog, t_inputrec *ir, int *fep_state, real *lambda, double *lam0)
2745 {
2746     /* this function works, but could probably use a logic rewrite to keep all the different
2747        types of efep straight. */
2748
2749     int       i;
2750     t_lambda *fep = ir->fepvals;
2751
2752     if ((ir->efep == efepNO) && (ir->bSimTemp == FALSE))
2753     {
2754         for (i = 0; i < efptNR; i++)
2755         {
2756             lambda[i] = 0.0;
2757             if (lam0)
2758             {
2759                 lam0[i] = 0.0;
2760             }
2761         }
2762         return;
2763     }
2764     else
2765     {
2766         *fep_state = fep->init_fep_state; /* this might overwrite the checkpoint
2767                                              if checkpoint is set -- a kludge is in for now
2768                                              to prevent this.*/
2769         for (i = 0; i < efptNR; i++)
2770         {
2771             /* overwrite lambda state with init_lambda for now for backwards compatibility */
2772             if (fep->init_lambda >= 0) /* if it's -1, it was never initializd */
2773             {
2774                 lambda[i] = fep->init_lambda;
2775                 if (lam0)
2776                 {
2777                     lam0[i] = lambda[i];
2778                 }
2779             }
2780             else
2781             {
2782                 lambda[i] = fep->all_lambda[i][*fep_state];
2783                 if (lam0)
2784                 {
2785                     lam0[i] = lambda[i];
2786                 }
2787             }
2788         }
2789         if (ir->bSimTemp)
2790         {
2791             /* need to rescale control temperatures to match current state */
2792             for (i = 0; i < ir->opts.ngtc; i++)
2793             {
2794                 if (ir->opts.ref_t[i] > 0)
2795                 {
2796                     ir->opts.ref_t[i] = ir->simtempvals->temperatures[*fep_state];
2797                 }
2798             }
2799         }
2800     }
2801
2802     /* Send to the log the information on the current lambdas */
2803     if (fplog != NULL)
2804     {
2805         fprintf(fplog, "Initial vector of lambda components:[ ");
2806         for (i = 0; i < efptNR; i++)
2807         {
2808             fprintf(fplog, "%10.4f ", lambda[i]);
2809         }
2810         fprintf(fplog, "]\n");
2811     }
2812     return;
2813 }
2814
2815
2816 void init_md(FILE *fplog,
2817              t_commrec *cr, t_inputrec *ir, const output_env_t oenv,
2818              double *t, double *t0,
2819              real *lambda, int *fep_state, double *lam0,
2820              t_nrnb *nrnb, gmx_mtop_t *mtop,
2821              gmx_update_t *upd,
2822              int nfile, const t_filenm fnm[],
2823              gmx_mdoutf_t *outf, t_mdebin **mdebin,
2824              tensor force_vir, tensor shake_vir, rvec mu_tot,
2825              gmx_bool *bSimAnn, t_vcm **vcm, unsigned long Flags)
2826 {
2827     int  i, j, n;
2828     real tmpt, mod;
2829
2830     /* Initial values */
2831     *t = *t0       = ir->init_t;
2832
2833     *bSimAnn = FALSE;
2834     for (i = 0; i < ir->opts.ngtc; i++)
2835     {
2836         /* set bSimAnn if any group is being annealed */
2837         if (ir->opts.annealing[i] != eannNO)
2838         {
2839             *bSimAnn = TRUE;
2840         }
2841     }
2842     if (*bSimAnn)
2843     {
2844         update_annealing_target_temp(&(ir->opts), ir->init_t);
2845     }
2846
2847     /* Initialize lambda variables */
2848     initialize_lambdas(fplog, ir, fep_state, lambda, lam0);
2849
2850     if (upd)
2851     {
2852         *upd = init_update(ir);
2853     }
2854
2855
2856     if (vcm != NULL)
2857     {
2858         *vcm = init_vcm(fplog, &mtop->groups, ir);
2859     }
2860
2861     if (EI_DYNAMICS(ir->eI) && !(Flags & MD_APPENDFILES))
2862     {
2863         if (ir->etc == etcBERENDSEN)
2864         {
2865             please_cite(fplog, "Berendsen84a");
2866         }
2867         if (ir->etc == etcVRESCALE)
2868         {
2869             please_cite(fplog, "Bussi2007a");
2870         }
2871     }
2872
2873     init_nrnb(nrnb);
2874
2875     if (nfile != -1)
2876     {
2877         *outf = init_mdoutf(fplog, nfile, fnm, Flags, cr, ir, mtop, oenv);
2878
2879         *mdebin = init_mdebin((Flags & MD_APPENDFILES) ? NULL : mdoutf_get_fp_ene(*outf),
2880                               mtop, ir, mdoutf_get_fp_dhdl(*outf));
2881     }
2882
2883     if (ir->bAdress)
2884     {
2885         please_cite(fplog, "Fritsch12");
2886         please_cite(fplog, "Junghans10");
2887     }
2888     /* Initiate variables */
2889     clear_mat(force_vir);
2890     clear_mat(shake_vir);
2891     clear_rvec(mu_tot);
2892
2893     debug_gmx();
2894 }