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