src/gromacs/mdlib/force.cpp

   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) 2013,2014,2015, 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 #include "gmxpre.h"
  38
  39 #include "gromacs/legacyheaders/force.h"
  40
  41 #include "config.h"
  42
  43 #include <assert.h>
  44 #include <math.h>
  45 #include <string.h>
  46
  47 #include "gromacs/domdec/domdec.h"
  48 #include "gromacs/ewald/ewald.h"
  49 #include "gromacs/ewald/long-range-correction.h"
  50 #include "gromacs/ewald/pme.h"
  51 #include "gromacs/legacyheaders/gmx_omp_nthreads.h"
  52 #include "gromacs/legacyheaders/macros.h"
  53 #include "gromacs/legacyheaders/mdrun.h"
  54 #include "gromacs/legacyheaders/names.h"
  55 #include "gromacs/legacyheaders/network.h"
  56 #include "gromacs/legacyheaders/nonbonded.h"
  57 #include "gromacs/legacyheaders/nrnb.h"
  58 #include "gromacs/legacyheaders/ns.h"
  59 #include "gromacs/legacyheaders/qmmm.h"
  60 #include "gromacs/legacyheaders/txtdump.h"
  61 #include "gromacs/legacyheaders/typedefs.h"
  62 #include "gromacs/legacyheaders/types/commrec.h"
  63 #include "gromacs/listed-forces/listed-forces.h"
  64 #include "gromacs/math/vec.h"
  65 #include "gromacs/mdlib/forcerec-threading.h"
  66 #include "gromacs/pbcutil/ishift.h"
  67 #include "gromacs/pbcutil/mshift.h"
  68 #include "gromacs/pbcutil/pbc.h"
  69 #include "gromacs/timing/wallcycle.h"
  70 #include "gromacs/utility/cstringutil.h"
  71 #include "gromacs/utility/fatalerror.h"
  72 #include "gromacs/utility/smalloc.h"
  73
  74 void ns(FILE              *fp,
  75         t_forcerec        *fr,
  76         matrix             box,
  77         gmx_groups_t      *groups,
  78         gmx_localtop_t    *top,
  79         t_mdatoms         *md,
  80         t_commrec         *cr,
  81         t_nrnb            *nrnb,
  82         gmx_bool           bFillGrid,
  83         gmx_bool           bDoLongRangeNS)
  84 {
  85     int     nsearch;
  86
  87
  88     if (!fr->ns.nblist_initialized)
  89     {
  90         init_neighbor_list(fp, fr, md->homenr);
  91     }
  92
  93     if (fr->bTwinRange)
  94     {
  95         fr->nlr = 0;
  96     }
  97
  98     nsearch = search_neighbours(fp, fr, box, top, groups, cr, nrnb, md,
  99                                 bFillGrid, bDoLongRangeNS);
 100     if (debug)
 101     {
 102         fprintf(debug, "nsearch = %d\n", nsearch);
 103     }
 104
 105     /* Check whether we have to do dynamic load balancing */
 106     /*if ((nsb->nstDlb > 0) && (mod(step,nsb->nstDlb) == 0))
 107        count_nb(cr,nsb,&(top->blocks[ebCGS]),nns,fr->nlr,
 108        &(top->idef),opts->ngener);
 109      */
 110     if (fr->ns.dump_nl > 0)
 111     {
 112         dump_nblist(fp, cr, fr, fr->ns.dump_nl);
 113     }
 114 }
 115
 116 static void reduce_thread_forces(int n, rvec *f,
 117                                  tensor vir_q, tensor vir_lj,
 118                                  real *Vcorr_q, real *Vcorr_lj,
 119                                  real *dvdl_q, real *dvdl_lj,
 120                                  int nthreads, f_thread_t *f_t)
 121 {
 122     int t, i;
 123     int nthreads_loop gmx_unused;
 124
 125     // cppcheck-suppress unreadVariable
 126     nthreads_loop = gmx_omp_nthreads_get(emntBonded);
 127     /* This reduction can run over any number of threads */
 128 #pragma omp parallel for num_threads(nthreads_loop) private(t) schedule(static)
 129     for (i = 0; i < n; i++)
 130     {
 131         for (t = 1; t < nthreads; t++)
 132         {
 133             rvec_inc(f[i], f_t[t].f[i]);
 134         }
 135     }
 136     for (t = 1; t < nthreads; t++)
 137     {
 138         *Vcorr_q  += f_t[t].Vcorr_q;
 139         *Vcorr_lj += f_t[t].Vcorr_lj;
 140         *dvdl_q   += f_t[t].dvdl[efptCOUL];
 141         *dvdl_lj  += f_t[t].dvdl[efptVDW];
 142         m_add(vir_q, f_t[t].vir_q, vir_q);
 143         m_add(vir_lj, f_t[t].vir_lj, vir_lj);
 144     }
 145 }
 146
 147 void do_force_lowlevel(t_forcerec *fr,      t_inputrec *ir,
 148                        t_idef     *idef,    t_commrec  *cr,
 149                        t_nrnb     *nrnb,    gmx_wallcycle_t wcycle,
 150                        t_mdatoms  *md,
 151                        rvec       x[],      history_t  *hist,
 152                        rvec       f[],
 153                        rvec       f_longrange[],
 154                        gmx_enerdata_t *enerd,
 155                        t_fcdata   *fcd,
 156                        gmx_localtop_t *top,
 157                        gmx_genborn_t *born,
 158                        gmx_bool       bBornRadii,
 159                        matrix     box,
 160                        t_lambda   *fepvals,
 161                        real       *lambda,
 162                        t_graph    *graph,
 163                        t_blocka   *excl,
 164                        rvec       mu_tot[],
 165                        int        flags,
 166                        float      *cycles_pme)
 167 {
 168     int         i, j;
 169     int         donb_flags;
 170     gmx_bool    bSB;
 171     int         pme_flags;
 172     matrix      boxs;
 173     rvec        box_size;
 174     t_pbc       pbc;
 175     real        dvdl_dum[efptNR], dvdl_nb[efptNR];
 176
 177 #ifdef GMX_MPI
 178     double  t0 = 0.0, t1, t2, t3; /* time measurement for coarse load balancing */
 179 #endif
 180
 181     set_pbc(&pbc, fr->ePBC, box);
 182
 183     /* reset free energy components */
 184     for (i = 0; i < efptNR; i++)
 185     {
 186         dvdl_nb[i]  = 0;
 187         dvdl_dum[i] = 0;
 188     }
 189
 190     /* Reset box */
 191     for (i = 0; (i < DIM); i++)
 192     {
 193         box_size[i] = box[i][i];
 194     }
 195
 196     debug_gmx();
 197
 198     /* do QMMM first if requested */
 199     if (fr->bQMMM)
 200     {
 201         enerd->term[F_EQM] = calculate_QMMM(cr, x, f, fr);
 202     }
 203
 204     /* Call the short range functions all in one go. */
 205
 206 #ifdef GMX_MPI
 207     /*#define TAKETIME ((cr->npmenodes) && (fr->timesteps < 12))*/
 208 #define TAKETIME FALSE
 209     if (TAKETIME)
 210     {
 211         MPI_Barrier(cr->mpi_comm_mygroup);
 212         t0 = MPI_Wtime();
 213     }
 214 #endif
 215
 216     if (ir->nwall)
 217     {
 218         /* foreign lambda component for walls */
 219         real dvdl_walls = do_walls(ir, fr, box, md, x, f, lambda[efptVDW],
 220                                    enerd->grpp.ener[egLJSR], nrnb);
 221         enerd->dvdl_lin[efptVDW] += dvdl_walls;
 222     }
 223
 224     /* If doing GB, reset dvda and calculate the Born radii */
 225     if (ir->implicit_solvent)
 226     {
 227         wallcycle_sub_start(wcycle, ewcsNONBONDED);
 228
 229         for (i = 0; i < born->nr; i++)
 230         {
 231             fr->dvda[i] = 0;
 232         }
 233
 234         if (bBornRadii)
 235         {
 236             calc_gb_rad(cr, fr, ir, top, x, &(fr->gblist), born, md, nrnb);
 237         }
 238
 239         wallcycle_sub_stop(wcycle, ewcsNONBONDED);
 240     }
 241
 242     where();
 243     /* We only do non-bonded calculation with group scheme here, the verlet
 244      * calls are done from do_force_cutsVERLET(). */
 245     if (fr->cutoff_scheme == ecutsGROUP && (flags & GMX_FORCE_NONBONDED))
 246     {
 247         donb_flags = 0;
 248         /* Add short-range interactions */
 249         donb_flags |= GMX_NONBONDED_DO_SR;
 250
 251         /* Currently all group scheme kernels always calculate (shift-)forces */
 252         if (flags & GMX_FORCE_FORCES)
 253         {
 254             donb_flags |= GMX_NONBONDED_DO_FORCE;
 255         }
 256         if (flags & GMX_FORCE_VIRIAL)
 257         {
 258             donb_flags |= GMX_NONBONDED_DO_SHIFTFORCE;
 259         }
 260         if (flags & GMX_FORCE_ENERGY)
 261         {
 262             donb_flags |= GMX_NONBONDED_DO_POTENTIAL;
 263         }
 264         if (flags & GMX_FORCE_DO_LR)
 265         {
 266             donb_flags |= GMX_NONBONDED_DO_LR;
 267         }
 268
 269         wallcycle_sub_start(wcycle, ewcsNONBONDED);
 270         do_nonbonded(fr, x, f, f_longrange, md, excl,
 271                      &enerd->grpp, nrnb,
 272                      lambda, dvdl_nb, -1, -1, donb_flags);
 273
 274         /* If we do foreign lambda and we have soft-core interactions
 275          * we have to recalculate the (non-linear) energies contributions.
 276          */
 277         if (fepvals->n_lambda > 0 && (flags & GMX_FORCE_DHDL) && fepvals->sc_alpha != 0)
 278         {
 279             for (i = 0; i < enerd->n_lambda; i++)
 280             {
 281                 real lam_i[efptNR];
 282
 283                 for (j = 0; j < efptNR; j++)
 284                 {
 285                     lam_i[j] = (i == 0 ? lambda[j] : fepvals->all_lambda[j][i-1]);
 286                 }
 287                 reset_foreign_enerdata(enerd);
 288                 do_nonbonded(fr, x, f, f_longrange, md, excl,
 289                              &(enerd->foreign_grpp), nrnb,
 290                              lam_i, dvdl_dum, -1, -1,
 291                              (donb_flags & ~GMX_NONBONDED_DO_FORCE) | GMX_NONBONDED_DO_FOREIGNLAMBDA);
 292                 sum_epot(&(enerd->foreign_grpp), enerd->foreign_term);
 293                 enerd->enerpart_lambda[i] += enerd->foreign_term[F_EPOT];
 294             }
 295         }
 296         wallcycle_sub_stop(wcycle, ewcsNONBONDED);
 297         where();
 298     }
 299
 300     /* If we are doing GB, calculate bonded forces and apply corrections
 301      * to the solvation forces */
 302     /* MRS: Eventually, many need to include free energy contribution here! */
 303     if (ir->implicit_solvent)
 304     {
 305         wallcycle_sub_start(wcycle, ewcsLISTED);
 306         calc_gb_forces(cr, md, born, top, x, f, fr, idef,
 307                        ir->gb_algorithm, ir->sa_algorithm, nrnb, &pbc, graph, enerd);
 308         wallcycle_sub_stop(wcycle, ewcsLISTED);
 309     }
 310
 311 #ifdef GMX_MPI
 312     if (TAKETIME)
 313     {
 314         t1          = MPI_Wtime();
 315         fr->t_fnbf += t1-t0;
 316     }
 317 #endif
 318
 319     if (fepvals->sc_alpha != 0)
 320     {
 321         enerd->dvdl_nonlin[efptVDW] += dvdl_nb[efptVDW];
 322     }
 323     else
 324     {
 325         enerd->dvdl_lin[efptVDW] += dvdl_nb[efptVDW];
 326     }
 327
 328     if (fepvals->sc_alpha != 0)
 329
 330     /* even though coulomb part is linear, we already added it, beacuse we
 331        need to go through the vdw calculation anyway */
 332     {
 333         enerd->dvdl_nonlin[efptCOUL] += dvdl_nb[efptCOUL];
 334     }
 335     else
 336     {
 337         enerd->dvdl_lin[efptCOUL] += dvdl_nb[efptCOUL];
 338     }
 339
 340     debug_gmx();
 341
 342
 343     if (debug)
 344     {
 345         pr_rvecs(debug, 0, "fshift after SR", fr->fshift, SHIFTS);
 346     }
 347
 348     /* Shift the coordinates. Must be done before listed forces and PPPM,
 349      * but is also necessary for SHAKE and update, therefore it can NOT
 350      * go when no listed forces have to be evaluated.
 351      *
 352      * The shifting and PBC code is deliberately not timed, since with
 353      * the Verlet scheme it only takes non-zero time with triclinic
 354      * boxes, and even then the time is around a factor of 100 less
 355      * than the next smallest counter.
 356      */
 357
 358
 359     /* Here sometimes we would not need to shift with NBFonly,
 360      * but we do so anyhow for consistency of the returned coordinates.
 361      */
 362     if (graph)
 363     {
 364         shift_self(graph, box, x);
 365         if (TRICLINIC(box))
 366         {
 367             inc_nrnb(nrnb, eNR_SHIFTX, 2*graph->nnodes);
 368         }
 369         else
 370         {
 371             inc_nrnb(nrnb, eNR_SHIFTX, graph->nnodes);
 372         }
 373     }
 374     /* Check whether we need to do listed interactions or correct for exclusions */
 375     if (fr->bMolPBC &&
 376         ((flags & GMX_FORCE_LISTED)
 377          || EEL_RF(fr->eeltype) || EEL_FULL(fr->eeltype) || EVDW_PME(fr->vdwtype)))
 378     {
 379         /* TODO There are no electrostatics methods that require this
 380            transformation, when using the Verlet scheme, so update the
 381            above conditional. */
 382         /* Since all atoms are in the rectangular or triclinic unit-cell,
 383          * only single box vector shifts (2 in x) are required.
 384          */
 385         set_pbc_dd(&pbc, fr->ePBC, cr->dd, TRUE, box);
 386     }
 387     debug_gmx();
 388
 389     do_force_listed(wcycle, box, ir->fepvals, cr->ms,
 390                     idef, (const rvec *) x, hist, f, fr,
 391                     &pbc, graph, enerd, nrnb, lambda, md, fcd,
 392                     DOMAINDECOMP(cr) ? cr->dd->gatindex : NULL,
 393                     flags);
 394
 395     where();
 396
 397     *cycles_pme = 0;
 398     clear_mat(fr->vir_el_recip);
 399     clear_mat(fr->vir_lj_recip);
 400
 401     /* Do long-range electrostatics and/or LJ-PME, including related short-range
 402      * corrections.
 403      */
 404     if (EEL_FULL(fr->eeltype) || EVDW_PME(fr->vdwtype))
 405     {
 406         int  status            = 0;
 407         real Vlr_q             = 0, Vlr_lj = 0, Vcorr_q = 0, Vcorr_lj = 0;
 408         real dvdl_long_range_q = 0, dvdl_long_range_lj = 0;
 409
 410         bSB = (ir->nwall == 2);
 411         if (bSB)
 412         {
 413             copy_mat(box, boxs);
 414             svmul(ir->wall_ewald_zfac, boxs[ZZ], boxs[ZZ]);
 415             box_size[ZZ] *= ir->wall_ewald_zfac;
 416         }
 417
 418         if (EEL_PME_EWALD(fr->eeltype) || EVDW_PME(fr->vdwtype))
 419         {
 420             real dvdl_long_range_correction_q   = 0;
 421             real dvdl_long_range_correction_lj  = 0;
 422             /* With the Verlet scheme exclusion forces are calculated
 423              * in the non-bonded kernel.
 424              */
 425             /* The TPI molecule does not have exclusions with the rest
 426              * of the system and no intra-molecular PME grid
 427              * contributions will be calculated in
 428              * gmx_pme_calc_energy.
 429              */
 430             if ((ir->cutoff_scheme == ecutsGROUP && fr->n_tpi == 0) ||
 431                 ir->ewald_geometry != eewg3D ||
 432                 ir->epsilon_surface != 0)
 433             {
 434                 int nthreads, t;
 435
 436                 wallcycle_sub_start(wcycle, ewcsEWALD_CORRECTION);
 437
 438                 if (fr->n_tpi > 0)
 439                 {
 440                     gmx_fatal(FARGS, "TPI with PME currently only works in a 3D geometry with tin-foil boundary conditions");
 441                 }
 442
 443                 nthreads = gmx_omp_nthreads_get(emntBonded);
 444 #pragma omp parallel for num_threads(nthreads) schedule(static)
 445                 for (t = 0; t < nthreads; t++)
 446                 {
 447                     int     i;
 448                     rvec   *fnv;
 449                     tensor *vir_q, *vir_lj;
 450                     real   *Vcorrt_q, *Vcorrt_lj, *dvdlt_q, *dvdlt_lj;
 451                     if (t == 0)
 452                     {
 453                         fnv       = fr->f_novirsum;
 454                         vir_q     = &fr->vir_el_recip;
 455                         vir_lj    = &fr->vir_lj_recip;
 456                         Vcorrt_q  = &Vcorr_q;
 457                         Vcorrt_lj = &Vcorr_lj;
 458                         dvdlt_q   = &dvdl_long_range_correction_q;
 459                         dvdlt_lj  = &dvdl_long_range_correction_lj;
 460                     }
 461                     else
 462                     {
 463                         fnv       = fr->f_t[t].f;
 464                         vir_q     = &fr->f_t[t].vir_q;
 465                         vir_lj    = &fr->f_t[t].vir_lj;
 466                         Vcorrt_q  = &fr->f_t[t].Vcorr_q;
 467                         Vcorrt_lj = &fr->f_t[t].Vcorr_lj;
 468                         dvdlt_q   = &fr->f_t[t].dvdl[efptCOUL];
 469                         dvdlt_lj  = &fr->f_t[t].dvdl[efptVDW];
 470                         for (i = 0; i < fr->natoms_force; i++)
 471                         {
 472                             clear_rvec(fnv[i]);
 473                         }
 474                         clear_mat(*vir_q);
 475                         clear_mat(*vir_lj);
 476                     }
 477                     *dvdlt_q  = 0;
 478                     *dvdlt_lj = 0;
 479
 480                     ewald_LRcorrection(fr->excl_load[t], fr->excl_load[t+1],
 481                                        cr, t, fr,
 482                                        md->chargeA, md->chargeB,
 483                                        md->sqrt_c6A, md->sqrt_c6B,
 484                                        md->sigmaA, md->sigmaB,
 485                                        md->sigma3A, md->sigma3B,
 486                                        md->nChargePerturbed || md->nTypePerturbed,
 487                                        ir->cutoff_scheme != ecutsVERLET,
 488                                        excl, x, bSB ? boxs : box, mu_tot,
 489                                        ir->ewald_geometry,
 490                                        ir->epsilon_surface,
 491                                        fnv, *vir_q, *vir_lj,
 492                                        Vcorrt_q, Vcorrt_lj,
 493                                        lambda[efptCOUL], lambda[efptVDW],
 494                                        dvdlt_q, dvdlt_lj);
 495                 }
 496                 if (nthreads > 1)
 497                 {
 498                     reduce_thread_forces(fr->natoms_force, fr->f_novirsum,
 499                                          fr->vir_el_recip, fr->vir_lj_recip,
 500                                          &Vcorr_q, &Vcorr_lj,
 501                                          &dvdl_long_range_correction_q,
 502                                          &dvdl_long_range_correction_lj,
 503                                          nthreads, fr->f_t);
 504                 }
 505                 wallcycle_sub_stop(wcycle, ewcsEWALD_CORRECTION);
 506             }
 507
 508             if (EEL_PME_EWALD(fr->eeltype) && fr->n_tpi == 0)
 509             {
 510                 /* This is not in a subcounter because it takes a
 511                    negligible and constant-sized amount of time */
 512                 Vcorr_q += ewald_charge_correction(cr, fr, lambda[efptCOUL], box,
 513                                                    &dvdl_long_range_correction_q,
 514                                                    fr->vir_el_recip);
 515             }
 516
 517             enerd->dvdl_lin[efptCOUL] += dvdl_long_range_correction_q;
 518             enerd->dvdl_lin[efptVDW]  += dvdl_long_range_correction_lj;
 519
 520             if ((EEL_PME(fr->eeltype) || EVDW_PME(fr->vdwtype)) && (cr->duty & DUTY_PME))
 521             {
 522                 /* Do reciprocal PME for Coulomb and/or LJ. */
 523                 assert(fr->n_tpi >= 0);
 524                 if (fr->n_tpi == 0 || (flags & GMX_FORCE_STATECHANGED))
 525                 {
 526                     pme_flags = GMX_PME_SPREAD | GMX_PME_SOLVE;
 527                     if (EEL_PME(fr->eeltype))
 528                     {
 529                         pme_flags     |= GMX_PME_DO_COULOMB;
 530                     }
 531                     if (EVDW_PME(fr->vdwtype))
 532                     {
 533                         pme_flags |= GMX_PME_DO_LJ;
 534                     }
 535                     if (flags & GMX_FORCE_FORCES)
 536                     {
 537                         pme_flags |= GMX_PME_CALC_F;
 538                     }
 539                     if (flags & GMX_FORCE_VIRIAL)
 540                     {
 541                         pme_flags |= GMX_PME_CALC_ENER_VIR;
 542                     }
 543                     if (fr->n_tpi > 0)
 544                     {
 545                         /* We don't calculate f, but we do want the potential */
 546                         pme_flags |= GMX_PME_CALC_POT;
 547                     }
 548                     wallcycle_start(wcycle, ewcPMEMESH);
 549                     status = gmx_pme_do(fr->pmedata,
 550                                         0, md->homenr - fr->n_tpi,
 551                                         x, fr->f_novirsum,
 552                                         md->chargeA, md->chargeB,
 553                                         md->sqrt_c6A, md->sqrt_c6B,
 554                                         md->sigmaA, md->sigmaB,
 555                                         bSB ? boxs : box, cr,
 556                                         DOMAINDECOMP(cr) ? dd_pme_maxshift_x(cr->dd) : 0,
 557                                         DOMAINDECOMP(cr) ? dd_pme_maxshift_y(cr->dd) : 0,
 558                                         nrnb, wcycle,
 559                                         fr->vir_el_recip, fr->ewaldcoeff_q,
 560                                         fr->vir_lj_recip, fr->ewaldcoeff_lj,
 561                                         &Vlr_q, &Vlr_lj,
 562                                         lambda[efptCOUL], lambda[efptVDW],
 563                                         &dvdl_long_range_q, &dvdl_long_range_lj, pme_flags);
 564                     *cycles_pme = wallcycle_stop(wcycle, ewcPMEMESH);
 565                     if (status != 0)
 566                     {
 567                         gmx_fatal(FARGS, "Error %d in reciprocal PME routine", status);
 568                     }
 569                     /* We should try to do as little computation after
 570                      * this as possible, because parallel PME synchronizes
 571                      * the nodes, so we want all load imbalance of the
 572                      * rest of the force calculation to be before the PME
 573                      * call.  DD load balancing is done on the whole time
 574                      * of the force call (without PME).
 575                      */
 576                 }
 577                 if (fr->n_tpi > 0)
 578                 {
 579                     if (EVDW_PME(ir->vdwtype))
 580                     {
 581
 582                         gmx_fatal(FARGS, "Test particle insertion not implemented with LJ-PME");
 583                     }
 584                     /* Determine the PME grid energy of the test molecule
 585                      * with the PME grid potential of the other charges.
 586                      */
 587                     gmx_pme_calc_energy(fr->pmedata, fr->n_tpi,
 588                                         x + md->homenr - fr->n_tpi,
 589                                         md->chargeA + md->homenr - fr->n_tpi,
 590                                         &Vlr_q);
 591                 }
 592             }
 593         }
 594
 595         if (!EEL_PME(fr->eeltype) && EEL_PME_EWALD(fr->eeltype))
 596         {
 597             Vlr_q = do_ewald(ir, x, fr->f_novirsum,
 598                              md->chargeA, md->chargeB,
 599                              box_size, cr, md->homenr,
 600                              fr->vir_el_recip, fr->ewaldcoeff_q,
 601                              lambda[efptCOUL], &dvdl_long_range_q, fr->ewald_table);
 602         }
 603
 604         /* Note that with separate PME nodes we get the real energies later */
 605         enerd->dvdl_lin[efptCOUL] += dvdl_long_range_q;
 606         enerd->dvdl_lin[efptVDW]  += dvdl_long_range_lj;
 607         enerd->term[F_COUL_RECIP]  = Vlr_q + Vcorr_q;
 608         enerd->term[F_LJ_RECIP]    = Vlr_lj + Vcorr_lj;
 609         if (debug)
 610         {
 611             fprintf(debug, "Vlr_q = %g, Vcorr_q = %g, Vlr_corr_q = %g\n",
 612                     Vlr_q, Vcorr_q, enerd->term[F_COUL_RECIP]);
 613             pr_rvecs(debug, 0, "vir_el_recip after corr", fr->vir_el_recip, DIM);
 614             pr_rvecs(debug, 0, "fshift after LR Corrections", fr->fshift, SHIFTS);
 615             fprintf(debug, "Vlr_lj: %g, Vcorr_lj = %g, Vlr_corr_lj = %g\n",
 616                     Vlr_lj, Vcorr_lj, enerd->term[F_LJ_RECIP]);
 617             pr_rvecs(debug, 0, "vir_lj_recip after corr", fr->vir_lj_recip, DIM);
 618         }
 619     }
 620     else
 621     {
 622         /* Is there a reaction-field exclusion correction needed? */
 623         if (EEL_RF(fr->eeltype) && eelRF_NEC != fr->eeltype)
 624         {
 625             /* With the Verlet scheme, exclusion forces are calculated
 626              * in the non-bonded kernel.
 627              */
 628             if (ir->cutoff_scheme != ecutsVERLET)
 629             {
 630                 real dvdl_rf_excl      = 0;
 631                 enerd->term[F_RF_EXCL] =
 632                     RF_excl_correction(fr, graph, md, excl, x, f,
 633                                        fr->fshift, &pbc, lambda[efptCOUL], &dvdl_rf_excl);
 634
 635                 enerd->dvdl_lin[efptCOUL] += dvdl_rf_excl;
 636             }
 637         }
 638     }
 639     where();
 640     debug_gmx();
 641
 642     if (debug)
 643     {
 644         print_nrnb(debug, nrnb);
 645     }
 646     debug_gmx();
 647
 648 #ifdef GMX_MPI
 649     if (TAKETIME)
 650     {
 651         t2 = MPI_Wtime();
 652         MPI_Barrier(cr->mpi_comm_mygroup);
 653         t3          = MPI_Wtime();
 654         fr->t_wait += t3-t2;
 655         if (fr->timesteps == 11)
 656         {
 657             char buf[22];
 658             fprintf(stderr, "* PP load balancing info: rank %d, step %s, rel wait time=%3.0f%% , load string value: %7.2f\n",
 659                     cr->nodeid, gmx_step_str(fr->timesteps, buf),
 660                     100*fr->t_wait/(fr->t_wait+fr->t_fnbf),
 661                     (fr->t_fnbf+fr->t_wait)/fr->t_fnbf);
 662         }
 663         fr->timesteps++;
 664     }
 665 #endif
 666
 667     if (debug)
 668     {
 669         pr_rvecs(debug, 0, "fshift after bondeds", fr->fshift, SHIFTS);
 670     }
 671
 672 }
 673
 674 void init_enerdata(int ngener, int n_lambda, gmx_enerdata_t *enerd)
 675 {
 676     int i, n2;
 677
 678     for (i = 0; i < F_NRE; i++)
 679     {
 680         enerd->term[i]         = 0;
 681         enerd->foreign_term[i] = 0;
 682     }
 683
 684
 685     for (i = 0; i < efptNR; i++)
 686     {
 687         enerd->dvdl_lin[i]     = 0;
 688         enerd->dvdl_nonlin[i]  = 0;
 689     }
 690
 691     n2 = ngener*ngener;
 692     if (debug)
 693     {
 694         fprintf(debug, "Creating %d sized group matrix for energies\n", n2);
 695     }
 696     enerd->grpp.nener         = n2;
 697     enerd->foreign_grpp.nener = n2;
 698     for (i = 0; (i < egNR); i++)
 699     {
 700         snew(enerd->grpp.ener[i], n2);
 701         snew(enerd->foreign_grpp.ener[i], n2);
 702     }
 703
 704     if (n_lambda)
 705     {
 706         enerd->n_lambda = 1 + n_lambda;
 707         snew(enerd->enerpart_lambda, enerd->n_lambda);
 708     }
 709     else
 710     {
 711         enerd->n_lambda = 0;
 712     }
 713 }
 714
 715 void destroy_enerdata(gmx_enerdata_t *enerd)
 716 {
 717     int i;
 718
 719     for (i = 0; (i < egNR); i++)
 720     {
 721         sfree(enerd->grpp.ener[i]);
 722     }
 723
 724     for (i = 0; (i < egNR); i++)
 725     {
 726         sfree(enerd->foreign_grpp.ener[i]);
 727     }
 728
 729     if (enerd->n_lambda)
 730     {
 731         sfree(enerd->enerpart_lambda);
 732     }
 733 }
 734
 735 static real sum_v(int n, real v[])
 736 {
 737     real t;
 738     int  i;
 739
 740     t = 0.0;
 741     for (i = 0; (i < n); i++)
 742     {
 743         t = t + v[i];
 744     }
 745
 746     return t;
 747 }
 748
 749 void sum_epot(gmx_grppairener_t *grpp, real *epot)
 750 {
 751     int i;
 752
 753     /* Accumulate energies */
 754     epot[F_COUL_SR]  = sum_v(grpp->nener, grpp->ener[egCOULSR]);
 755     epot[F_LJ]       = sum_v(grpp->nener, grpp->ener[egLJSR]);
 756     epot[F_LJ14]     = sum_v(grpp->nener, grpp->ener[egLJ14]);
 757     epot[F_COUL14]   = sum_v(grpp->nener, grpp->ener[egCOUL14]);
 758     epot[F_COUL_LR]  = sum_v(grpp->nener, grpp->ener[egCOULLR]);
 759     epot[F_LJ_LR]    = sum_v(grpp->nener, grpp->ener[egLJLR]);
 760     /* We have already added 1-2,1-3, and 1-4 terms to F_GBPOL */
 761     epot[F_GBPOL]   += sum_v(grpp->nener, grpp->ener[egGB]);
 762
 763 /* lattice part of LR doesnt belong to any group
 764  * and has been added earlier
 765  */
 766     epot[F_BHAM]     = sum_v(grpp->nener, grpp->ener[egBHAMSR]);
 767     epot[F_BHAM_LR]  = sum_v(grpp->nener, grpp->ener[egBHAMLR]);
 768
 769     epot[F_EPOT] = 0;
 770     for (i = 0; (i < F_EPOT); i++)
 771     {
 772         if (i != F_DISRESVIOL && i != F_ORIRESDEV)
 773         {
 774             epot[F_EPOT] += epot[i];
 775         }
 776     }
 777 }
 778
 779 void sum_dhdl(gmx_enerdata_t *enerd, real *lambda, t_lambda *fepvals)
 780 {
 781     int    i, j, index;
 782     double dlam;
 783
 784     enerd->dvdl_lin[efptVDW] += enerd->term[F_DVDL_VDW];  /* include dispersion correction */
 785     enerd->term[F_DVDL]       = 0.0;
 786     for (i = 0; i < efptNR; i++)
 787     {
 788         if (fepvals->separate_dvdl[i])
 789         {
 790             /* could this be done more readably/compactly? */
 791             switch (i)
 792             {
 793                 case (efptMASS):
 794                     index = F_DKDL;
 795                     break;
 796                 case (efptCOUL):
 797                     index = F_DVDL_COUL;
 798                     break;
 799                 case (efptVDW):
 800                     index = F_DVDL_VDW;
 801                     break;
 802                 case (efptBONDED):
 803                     index = F_DVDL_BONDED;
 804                     break;
 805                 case (efptRESTRAINT):
 806                     index = F_DVDL_RESTRAINT;
 807                     break;
 808                 default:
 809                     index = F_DVDL;
 810                     break;
 811             }
 812             enerd->term[index] = enerd->dvdl_lin[i] + enerd->dvdl_nonlin[i];
 813             if (debug)
 814             {
 815                 fprintf(debug, "dvdl-%s[%2d]: %f: non-linear %f + linear %f\n",
 816                         efpt_names[i], i, enerd->term[index], enerd->dvdl_nonlin[i], enerd->dvdl_lin[i]);
 817             }
 818         }
 819         else
 820         {
 821             enerd->term[F_DVDL] += enerd->dvdl_lin[i] + enerd->dvdl_nonlin[i];
 822             if (debug)
 823             {
 824                 fprintf(debug, "dvd-%sl[%2d]: %f: non-linear %f + linear %f\n",
 825                         efpt_names[0], i, enerd->term[F_DVDL], enerd->dvdl_nonlin[i], enerd->dvdl_lin[i]);
 826             }
 827         }
 828     }
 829
 830     /* Notes on the foreign lambda free energy difference evaluation:
 831      * Adding the potential and ekin terms that depend linearly on lambda
 832      * as delta lam * dvdl to the energy differences is exact.
 833      * For the constraints this is not exact, but we have no other option
 834      * without literally changing the lengths and reevaluating the energies at each step.
 835      * (try to remedy this post 4.6 - MRS)
 836      * For the non-bonded LR term we assume that the soft-core (if present)
 837      * no longer affects the energy beyond the short-range cut-off,
 838      * which is a very good approximation (except for exotic settings).
 839      * (investigate how to overcome this post 4.6 - MRS)
 840      */
 841     if (fepvals->separate_dvdl[efptBONDED])
 842     {
 843         enerd->term[F_DVDL_BONDED] += enerd->term[F_DVDL_CONSTR];
 844     }
 845     else
 846     {
 847         enerd->term[F_DVDL] += enerd->term[F_DVDL_CONSTR];
 848     }
 849     enerd->term[F_DVDL_CONSTR] = 0;
 850
 851     for (i = 0; i < fepvals->n_lambda; i++)
 852     {
 853         /* note we are iterating over fepvals here!
 854            For the current lam, dlam = 0 automatically,
 855            so we don't need to add anything to the
 856            enerd->enerpart_lambda[0] */
 857
 858         /* we don't need to worry about dvdl_lin contributions to dE at
 859            current lambda, because the contributions to the current
 860            lambda are automatically zeroed */
 861
 862         for (j = 0; j < efptNR; j++)
 863         {
 864             /* Note that this loop is over all dhdl components, not just the separated ones */
 865             dlam = (fepvals->all_lambda[j][i]-lambda[j]);
 866             enerd->enerpart_lambda[i+1] += dlam*enerd->dvdl_lin[j];
 867             if (debug)
 868             {
 869                 fprintf(debug, "enerdiff lam %g: (%15s), non-linear %f linear %f*%f\n",
 870                         fepvals->all_lambda[j][i], efpt_names[j],
 871                         (enerd->enerpart_lambda[i+1] - enerd->enerpart_lambda[0]),
 872                         dlam, enerd->dvdl_lin[j]);
 873             }
 874         }
 875     }
 876 }
 877
 878
 879 void reset_foreign_enerdata(gmx_enerdata_t *enerd)
 880 {
 881     int  i, j;
 882
 883     /* First reset all foreign energy components.  Foreign energies always called on
 884        neighbor search steps */
 885     for (i = 0; (i < egNR); i++)
 886     {
 887         for (j = 0; (j < enerd->grpp.nener); j++)
 888         {
 889             enerd->foreign_grpp.ener[i][j] = 0.0;
 890         }
 891     }
 892
 893     /* potential energy components */
 894     for (i = 0; (i <= F_EPOT); i++)
 895     {
 896         enerd->foreign_term[i] = 0.0;
 897     }
 898 }
 899
 900 void reset_enerdata(t_forcerec *fr, gmx_bool bNS,
 901                     gmx_enerdata_t *enerd,
 902                     gmx_bool bMaster)
 903 {
 904     gmx_bool bKeepLR;
 905     int      i, j;
 906
 907     /* First reset all energy components, except for the long range terms
 908      * on the master at non neighbor search steps, since the long range
 909      * terms have already been summed at the last neighbor search step.
 910      */
 911     bKeepLR = (fr->bTwinRange && !bNS);
 912     for (i = 0; (i < egNR); i++)
 913     {
 914         if (!(bKeepLR && bMaster && (i == egCOULLR || i == egLJLR)))
 915         {
 916             for (j = 0; (j < enerd->grpp.nener); j++)
 917             {
 918                 enerd->grpp.ener[i][j] = 0.0;
 919             }
 920         }
 921     }
 922     for (i = 0; i < efptNR; i++)
 923     {
 924         enerd->dvdl_lin[i]    = 0.0;
 925         enerd->dvdl_nonlin[i] = 0.0;
 926     }
 927
 928     /* Normal potential energy components */
 929     for (i = 0; (i <= F_EPOT); i++)
 930     {
 931         enerd->term[i] = 0.0;
 932     }
 933     /* Initialize the dVdlambda term with the long range contribution */
 934     /* Initialize the dvdl term with the long range contribution */
 935     enerd->term[F_DVDL]            = 0.0;
 936     enerd->term[F_DVDL_COUL]       = 0.0;
 937     enerd->term[F_DVDL_VDW]        = 0.0;
 938     enerd->term[F_DVDL_BONDED]     = 0.0;
 939     enerd->term[F_DVDL_RESTRAINT]  = 0.0;
 940     enerd->term[F_DKDL]            = 0.0;
 941     if (enerd->n_lambda > 0)
 942     {
 943         for (i = 0; i < enerd->n_lambda; i++)
 944         {
 945             enerd->enerpart_lambda[i] = 0.0;
 946         }
 947     }
 948     /* reset foreign energy data - separate function since we also call it elsewhere */
 949     reset_foreign_enerdata(enerd);
 950 }