src/gromacs/mdlib/force.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 <math.h>
  42 #include <string.h>
  43 #include <assert.h>
  44 #include "sysstuff.h"
  45 #include "typedefs.h"
  46 #include "macros.h"
  47 #include "smalloc.h"
  48 #include "macros.h"
  49 #include "physics.h"
  50 #include "force.h"
  51 #include "nonbonded.h"
  52 #include "names.h"
  53 #include "network.h"
  54 #include "pbc.h"
  55 #include "ns.h"
  56 #include "nrnb.h"
  57 #include "bondf.h"
  58 #include "mshift.h"
  59 #include "txtdump.h"
  60 #include "coulomb.h"
  61 #include "pme.h"
  62 #include "mdrun.h"
  63 #include "domdec.h"
  64 #include "qmmm.h"
  65 #include "gmx_omp_nthreads.h"
  66
  67 #include "gromacs/timing/wallcycle.h"
  68
  69 void ns(FILE              *fp,
  70         t_forcerec        *fr,
  71         matrix             box,
  72         gmx_groups_t      *groups,
  73         gmx_localtop_t    *top,
  74         t_mdatoms         *md,
  75         t_commrec         *cr,
  76         t_nrnb            *nrnb,
  77         gmx_bool           bFillGrid,
  78         gmx_bool           bDoLongRangeNS)
  79 {
  80     char   *ptr;
  81     int     nsearch;
  82
  83
  84     if (!fr->ns.nblist_initialized)
  85     {
  86         init_neighbor_list(fp, fr, md->homenr);
  87     }
  88
  89     if (fr->bTwinRange)
  90     {
  91         fr->nlr = 0;
  92     }
  93
  94     nsearch = search_neighbours(fp, fr, box, top, groups, cr, nrnb, md,
  95                                 bFillGrid, bDoLongRangeNS);
  96     if (debug)
  97     {
  98         fprintf(debug, "nsearch = %d\n", nsearch);
  99     }
 100
 101     /* Check whether we have to do dynamic load balancing */
 102     /*if ((nsb->nstDlb > 0) && (mod(step,nsb->nstDlb) == 0))
 103        count_nb(cr,nsb,&(top->blocks[ebCGS]),nns,fr->nlr,
 104        &(top->idef),opts->ngener);
 105      */
 106     if (fr->ns.dump_nl > 0)
 107     {
 108         dump_nblist(fp, cr, fr, fr->ns.dump_nl);
 109     }
 110 }
 111
 112 static void reduce_thread_forces(int n, rvec *f,
 113                                  tensor vir_q, tensor vir_lj,
 114                                  real *Vcorr_q, real *Vcorr_lj,
 115                                  real *dvdl_q, real *dvdl_lj,
 116                                  int nthreads, f_thread_t *f_t)
 117 {
 118     int t, i;
 119
 120     /* This reduction can run over any number of threads */
 121 #pragma omp parallel for num_threads(gmx_omp_nthreads_get(emntBonded)) private(t) schedule(static)
 122     for (i = 0; i < n; i++)
 123     {
 124         for (t = 1; t < nthreads; t++)
 125         {
 126             rvec_inc(f[i], f_t[t].f[i]);
 127         }
 128     }
 129     for (t = 1; t < nthreads; t++)
 130     {
 131         *Vcorr_q  += f_t[t].Vcorr_q;
 132         *Vcorr_lj += f_t[t].Vcorr_lj;
 133         *dvdl_q   += f_t[t].dvdl[efptCOUL];
 134         *dvdl_lj  += f_t[t].dvdl[efptVDW];
 135         m_add(vir_q, f_t[t].vir_q, vir_q);
 136         m_add(vir_lj, f_t[t].vir_lj, vir_lj);
 137     }
 138 }
 139
 140 void gmx_print_sepdvdl(FILE *fplog, const char *s, real v, real dvdlambda)
 141 {
 142     fprintf(fplog, "  %-30s V %12.5e  dVdl %12.5e\n", s, v, dvdlambda);
 143 }
 144
 145 void do_force_lowlevel(FILE       *fplog,   gmx_int64_t step,
 146                        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                        t_atomtypes *atype,
 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    bDoEpot, bSepDVDL, bSB;
 171     int         pme_flags;
 172     matrix      boxs;
 173     rvec        box_size;
 174     t_pbc       pbc;
 175     char        buf[22];
 176     double      clam_i, vlam_i;
 177     real        dvdl_dum[efptNR], dvdl_nb[efptNR], lam_i[efptNR];
 178     real        dvdl_q, dvdl_lj;
 179
 180 #ifdef GMX_MPI
 181     double  t0 = 0.0, t1, t2, t3; /* time measurement for coarse load balancing */
 182 #endif
 183
 184 #define PRINT_SEPDVDL(s, v, dvdlambda) if (bSepDVDL) { gmx_print_sepdvdl(fplog, s, v, dvdlambda); }
 185
 186     set_pbc(&pbc, fr->ePBC, box);
 187
 188     /* reset free energy components */
 189     for (i = 0; i < efptNR; i++)
 190     {
 191         dvdl_nb[i]  = 0;
 192         dvdl_dum[i] = 0;
 193     }
 194
 195     /* Reset box */
 196     for (i = 0; (i < DIM); i++)
 197     {
 198         box_size[i] = box[i][i];
 199     }
 200
 201     bSepDVDL = (fr->bSepDVDL && do_per_step(step, ir->nstlog));
 202     debug_gmx();
 203
 204     /* do QMMM first if requested */
 205     if (fr->bQMMM)
 206     {
 207         enerd->term[F_EQM] = calculate_QMMM(cr, x, f, fr);
 208     }
 209
 210     if (bSepDVDL)
 211     {
 212         fprintf(fplog, "Step %s: non-bonded V and dVdl for node %d:\n",
 213                 gmx_step_str(step, buf), cr->nodeid);
 214     }
 215
 216     /* Call the short range functions all in one go. */
 217
 218 #ifdef GMX_MPI
 219     /*#define TAKETIME ((cr->npmenodes) && (fr->timesteps < 12))*/
 220 #define TAKETIME FALSE
 221     if (TAKETIME)
 222     {
 223         MPI_Barrier(cr->mpi_comm_mygroup);
 224         t0 = MPI_Wtime();
 225     }
 226 #endif
 227
 228     if (ir->nwall)
 229     {
 230         /* foreign lambda component for walls */
 231         real dvdl_walls = do_walls(ir, fr, box, md, x, f, lambda[efptVDW],
 232                                    enerd->grpp.ener[egLJSR], nrnb);
 233         PRINT_SEPDVDL("Walls", 0.0, dvdl_walls);
 234         enerd->dvdl_lin[efptVDW] += dvdl_walls;
 235     }
 236
 237     /* If doing GB, reset dvda and calculate the Born radii */
 238     if (ir->implicit_solvent)
 239     {
 240         wallcycle_sub_start(wcycle, ewcsNONBONDED);
 241
 242         for (i = 0; i < born->nr; i++)
 243         {
 244             fr->dvda[i] = 0;
 245         }
 246
 247         if (bBornRadii)
 248         {
 249             calc_gb_rad(cr, fr, ir, top, x, &(fr->gblist), born, md, nrnb);
 250         }
 251
 252         wallcycle_sub_stop(wcycle, ewcsNONBONDED);
 253     }
 254
 255     where();
 256     /* We only do non-bonded calculation with group scheme here, the verlet
 257      * calls are done from do_force_cutsVERLET(). */
 258     if (fr->cutoff_scheme == ecutsGROUP && (flags & GMX_FORCE_NONBONDED))
 259     {
 260         donb_flags = 0;
 261         /* Add short-range interactions */
 262         donb_flags |= GMX_NONBONDED_DO_SR;
 263
 264         if (flags & GMX_FORCE_FORCES)
 265         {
 266             donb_flags |= GMX_NONBONDED_DO_FORCE;
 267         }
 268         if (flags & GMX_FORCE_ENERGY)
 269         {
 270             donb_flags |= GMX_NONBONDED_DO_POTENTIAL;
 271         }
 272         if (flags & GMX_FORCE_DO_LR)
 273         {
 274             donb_flags |= GMX_NONBONDED_DO_LR;
 275         }
 276
 277         wallcycle_sub_start(wcycle, ewcsNONBONDED);
 278         do_nonbonded(fr, x, f, f_longrange, md, excl,
 279                      &enerd->grpp, nrnb,
 280                      lambda, dvdl_nb, -1, -1, donb_flags);
 281
 282         /* If we do foreign lambda and we have soft-core interactions
 283          * we have to recalculate the (non-linear) energies contributions.
 284          */
 285         if (fepvals->n_lambda > 0 && (flags & GMX_FORCE_DHDL) && fepvals->sc_alpha != 0)
 286         {
 287             for (i = 0; i < enerd->n_lambda; i++)
 288             {
 289                 for (j = 0; j < efptNR; j++)
 290                 {
 291                     lam_i[j] = (i == 0 ? lambda[j] : fepvals->all_lambda[j][i-1]);
 292                 }
 293                 reset_foreign_enerdata(enerd);
 294                 do_nonbonded(fr, x, f, f_longrange, md, excl,
 295                              &(enerd->foreign_grpp), nrnb,
 296                              lam_i, dvdl_dum, -1, -1,
 297                              (donb_flags & ~GMX_NONBONDED_DO_FORCE) | GMX_NONBONDED_DO_FOREIGNLAMBDA);
 298                 sum_epot(&(enerd->foreign_grpp), enerd->foreign_term);
 299                 enerd->enerpart_lambda[i] += enerd->foreign_term[F_EPOT];
 300             }
 301         }
 302         wallcycle_sub_stop(wcycle, ewcsNONBONDED);
 303         where();
 304     }
 305
 306     /* If we are doing GB, calculate bonded forces and apply corrections
 307      * to the solvation forces */
 308     /* MRS: Eventually, many need to include free energy contribution here! */
 309     if (ir->implicit_solvent)
 310     {
 311         wallcycle_sub_start(wcycle, ewcsBONDED);
 312         calc_gb_forces(cr, md, born, top, x, f, fr, idef,
 313                        ir->gb_algorithm, ir->sa_algorithm, nrnb, &pbc, graph, enerd);
 314         wallcycle_sub_stop(wcycle, ewcsBONDED);
 315     }
 316
 317 #ifdef GMX_MPI
 318     if (TAKETIME)
 319     {
 320         t1          = MPI_Wtime();
 321         fr->t_fnbf += t1-t0;
 322     }
 323 #endif
 324
 325     if (fepvals->sc_alpha != 0)
 326     {
 327         enerd->dvdl_nonlin[efptVDW] += dvdl_nb[efptVDW];
 328     }
 329     else
 330     {
 331         enerd->dvdl_lin[efptVDW] += dvdl_nb[efptVDW];
 332     }
 333
 334     if (fepvals->sc_alpha != 0)
 335
 336     /* even though coulomb part is linear, we already added it, beacuse we
 337        need to go through the vdw calculation anyway */
 338     {
 339         enerd->dvdl_nonlin[efptCOUL] += dvdl_nb[efptCOUL];
 340     }
 341     else
 342     {
 343         enerd->dvdl_lin[efptCOUL] += dvdl_nb[efptCOUL];
 344     }
 345
 346     if (bSepDVDL)
 347     {
 348         real V_short_range    = 0;
 349         real dvdl_short_range = 0;
 350
 351         for (i = 0; i < enerd->grpp.nener; i++)
 352         {
 353             V_short_range +=
 354                 (fr->bBHAM ?
 355                  enerd->grpp.ener[egBHAMSR][i] :
 356                  enerd->grpp.ener[egLJSR][i])
 357                 + enerd->grpp.ener[egCOULSR][i] + enerd->grpp.ener[egGB][i];
 358         }
 359         dvdl_short_range = dvdl_nb[efptVDW] + dvdl_nb[efptCOUL];
 360         PRINT_SEPDVDL("VdW and Coulomb SR particle-p.",
 361                       V_short_range,
 362                       dvdl_short_range);
 363     }
 364     debug_gmx();
 365
 366
 367     if (debug)
 368     {
 369         pr_rvecs(debug, 0, "fshift after SR", fr->fshift, SHIFTS);
 370     }
 371
 372     /* Shift the coordinates. Must be done before bonded forces and PPPM,
 373      * but is also necessary for SHAKE and update, therefore it can NOT
 374      * go when no bonded forces have to be evaluated.
 375      */
 376
 377     /* Here sometimes we would not need to shift with NBFonly,
 378      * but we do so anyhow for consistency of the returned coordinates.
 379      */
 380     if (graph)
 381     {
 382         shift_self(graph, box, x);
 383         if (TRICLINIC(box))
 384         {
 385             inc_nrnb(nrnb, eNR_SHIFTX, 2*graph->nnodes);
 386         }
 387         else
 388         {
 389             inc_nrnb(nrnb, eNR_SHIFTX, graph->nnodes);
 390         }
 391     }
 392     /* Check whether we need to do bondeds or correct for exclusions */
 393     if (fr->bMolPBC &&
 394         ((flags & GMX_FORCE_BONDED)
 395          || EEL_RF(fr->eeltype) || EEL_FULL(fr->eeltype) || EVDW_PME(fr->vdwtype)))
 396     {
 397         /* Since all atoms are in the rectangular or triclinic unit-cell,
 398          * only single box vector shifts (2 in x) are required.
 399          */
 400         set_pbc_dd(&pbc, fr->ePBC, cr->dd, TRUE, box);
 401     }
 402     debug_gmx();
 403
 404     if (flags & GMX_FORCE_BONDED)
 405     {
 406         wallcycle_sub_start(wcycle, ewcsBONDED);
 407         calc_bonds(fplog, cr->ms,
 408                    idef, x, hist, f, fr, &pbc, graph, enerd, nrnb, lambda, md, fcd,
 409                    DOMAINDECOMP(cr) ? cr->dd->gatindex : NULL, atype, born,
 410                    flags,
 411                    fr->bSepDVDL && do_per_step(step, ir->nstlog), step);
 412
 413         /* Check if we have to determine energy differences
 414          * at foreign lambda's.
 415          */
 416         if (fepvals->n_lambda > 0 && (flags & GMX_FORCE_DHDL) &&
 417             idef->ilsort != ilsortNO_FE)
 418         {
 419             if (idef->ilsort != ilsortFE_SORTED)
 420             {
 421                 gmx_incons("The bonded interactions are not sorted for free energy");
 422             }
 423             for (i = 0; i < enerd->n_lambda; i++)
 424             {
 425                 reset_foreign_enerdata(enerd);
 426                 for (j = 0; j < efptNR; j++)
 427                 {
 428                     lam_i[j] = (i == 0 ? lambda[j] : fepvals->all_lambda[j][i-1]);
 429                 }
 430                 calc_bonds_lambda(fplog, idef, x, fr, &pbc, graph, &(enerd->foreign_grpp), enerd->foreign_term, nrnb, lam_i, md,
 431                                   fcd, DOMAINDECOMP(cr) ? cr->dd->gatindex : NULL);
 432                 sum_epot(&(enerd->foreign_grpp), enerd->foreign_term);
 433                 enerd->enerpart_lambda[i] += enerd->foreign_term[F_EPOT];
 434             }
 435         }
 436         debug_gmx();
 437
 438         wallcycle_sub_stop(wcycle, ewcsBONDED);
 439     }
 440
 441     where();
 442
 443     *cycles_pme = 0;
 444     if (EEL_FULL(fr->eeltype) || EVDW_PME(fr->vdwtype))
 445     {
 446         real Vlr             = 0, Vcorr = 0;
 447         real dvdl_long_range = 0;
 448         int  status          = 0;
 449
 450         bSB = (ir->nwall == 2);
 451         if (bSB)
 452         {
 453             copy_mat(box, boxs);
 454             svmul(ir->wall_ewald_zfac, boxs[ZZ], boxs[ZZ]);
 455             box_size[ZZ] *= ir->wall_ewald_zfac;
 456         }
 457     }
 458
 459     /* Do long-range electrostatics and/or LJ-PME, including related short-range
 460      * corrections.
 461      */
 462
 463     clear_mat(fr->vir_el_recip);
 464     clear_mat(fr->vir_lj_recip);
 465
 466     if (EEL_FULL(fr->eeltype) || EVDW_PME(fr->vdwtype))
 467     {
 468         real Vlr_q             = 0, Vlr_lj = 0, Vcorr_q = 0, Vcorr_lj = 0;
 469         real dvdl_long_range_q = 0, dvdl_long_range_lj = 0;
 470         int  status            = 0;
 471
 472         if (EEL_EWALD(fr->eeltype) || EVDW_PME(fr->vdwtype))
 473         {
 474             real dvdl_long_range_correction_q   = 0;
 475             real dvdl_long_range_correction_lj  = 0;
 476             /* With the Verlet scheme exclusion forces are calculated
 477              * in the non-bonded kernel.
 478              */
 479             /* The TPI molecule does not have exclusions with the rest
 480              * of the system and no intra-molecular PME grid
 481              * contributions will be calculated in
 482              * gmx_pme_calc_energy.
 483              */
 484             if ((ir->cutoff_scheme == ecutsGROUP && fr->n_tpi == 0) ||
 485                 ir->ewald_geometry != eewg3D ||
 486                 ir->epsilon_surface != 0)
 487             {
 488                 int nthreads, t;
 489
 490                 wallcycle_sub_start(wcycle, ewcsEWALD_CORRECTION);
 491
 492                 if (fr->n_tpi > 0)
 493                 {
 494                     gmx_fatal(FARGS, "TPI with PME currently only works in a 3D geometry with tin-foil boundary conditions");
 495                 }
 496
 497                 nthreads = gmx_omp_nthreads_get(emntBonded);
 498 #pragma omp parallel for num_threads(nthreads) schedule(static)
 499                 for (t = 0; t < nthreads; t++)
 500                 {
 501                     int     s, e, i;
 502                     rvec   *fnv;
 503                     tensor *vir_q, *vir_lj;
 504                     real   *Vcorrt_q, *Vcorrt_lj, *dvdlt_q, *dvdlt_lj;
 505                     if (t == 0)
 506                     {
 507                         fnv       = fr->f_novirsum;
 508                         vir_q     = &fr->vir_el_recip;
 509                         vir_lj    = &fr->vir_lj_recip;
 510                         Vcorrt_q  = &Vcorr_q;
 511                         Vcorrt_lj = &Vcorr_lj;
 512                         dvdlt_q   = &dvdl_long_range_correction_q;
 513                         dvdlt_lj  = &dvdl_long_range_correction_lj;
 514                     }
 515                     else
 516                     {
 517                         fnv       = fr->f_t[t].f;
 518                         vir_q     = &fr->f_t[t].vir_q;
 519                         vir_lj    = &fr->f_t[t].vir_lj;
 520                         Vcorrt_q  = &fr->f_t[t].Vcorr_q;
 521                         Vcorrt_lj = &fr->f_t[t].Vcorr_lj;
 522                         dvdlt_q   = &fr->f_t[t].dvdl[efptCOUL];
 523                         dvdlt_lj  = &fr->f_t[t].dvdl[efptVDW];
 524                         for (i = 0; i < fr->natoms_force; i++)
 525                         {
 526                             clear_rvec(fnv[i]);
 527                         }
 528                         clear_mat(*vir_q);
 529                         clear_mat(*vir_lj);
 530                     }
 531                     *dvdlt_q  = 0;
 532                     *dvdlt_lj = 0;
 533                     ewald_LRcorrection(fr->excl_load[t], fr->excl_load[t+1],
 534                                        cr, t, fr,
 535                                        md->chargeA,
 536                                        md->nChargePerturbed ? md->chargeB : NULL,
 537                                        md->sqrt_c6A,
 538                                        md->nChargePerturbed ? md->sqrt_c6B : NULL,
 539                                        md->sigmaA,
 540                                        md->nChargePerturbed ? md->sigmaB : NULL,
 541                                        md->sigma3A,
 542                                        md->nChargePerturbed ? md->sigma3B : NULL,
 543                                        ir->cutoff_scheme != ecutsVERLET,
 544                                        excl, x, bSB ? boxs : box, mu_tot,
 545                                        ir->ewald_geometry,
 546                                        ir->epsilon_surface,
 547                                        fnv, *vir_q, *vir_lj,
 548                                        Vcorrt_q, Vcorrt_lj,
 549                                        lambda[efptCOUL], lambda[efptVDW],
 550                                        dvdlt_q, dvdlt_lj);
 551                 }
 552                 if (nthreads > 1)
 553                 {
 554                     reduce_thread_forces(fr->natoms_force, fr->f_novirsum,
 555                                          fr->vir_el_recip, fr->vir_lj_recip,
 556                                          &Vcorr_q, &Vcorr_lj,
 557                                          &dvdl_long_range_correction_q,
 558                                          &dvdl_long_range_correction_lj,
 559                                          nthreads, fr->f_t);
 560                 }
 561                 wallcycle_sub_stop(wcycle, ewcsEWALD_CORRECTION);
 562             }
 563
 564             if (EEL_EWALD(fr->eeltype) && fr->n_tpi == 0)
 565             {
 566                 Vcorr_q += ewald_charge_correction(cr, fr, lambda[efptCOUL], box,
 567                                                    &dvdl_long_range_correction_q,
 568                                                    fr->vir_el_recip);
 569             }
 570
 571             PRINT_SEPDVDL("Ewald excl./charge/dip. corr.", Vcorr_q, dvdl_long_range_correction_q);
 572             PRINT_SEPDVDL("Ewald excl. corr. LJ", Vcorr_lj, dvdl_long_range_correction_lj);
 573             enerd->dvdl_lin[efptCOUL] += dvdl_long_range_correction_q;
 574             enerd->dvdl_lin[efptVDW]  += dvdl_long_range_correction_lj;
 575         }
 576
 577         if ((EEL_PME(fr->eeltype) || EVDW_PME(fr->vdwtype)))
 578         {
 579             if (cr->duty & DUTY_PME)
 580             {
 581                 /* Do reciprocal PME for Coulomb and/or LJ. */
 582                 assert(fr->n_tpi >= 0);
 583                 if (fr->n_tpi == 0 || (flags & GMX_FORCE_STATECHANGED))
 584                 {
 585                     pme_flags = GMX_PME_SPREAD_Q | GMX_PME_SOLVE;
 586                     if (EEL_PME(fr->eeltype))
 587                     {
 588                         pme_flags     |= GMX_PME_DO_COULOMB;
 589                     }
 590                     if (EVDW_PME(fr->vdwtype))
 591                     {
 592                         pme_flags |= GMX_PME_DO_LJ;
 593                         if (fr->ljpme_combination_rule == eljpmeLB)
 594                         {
 595                             /*Lorentz-Berthelot Comb. Rules in LJ-PME*/
 596                             pme_flags |= GMX_PME_LJ_LB;
 597                         }
 598                     }
 599                     if (flags & GMX_FORCE_FORCES)
 600                     {
 601                         pme_flags |= GMX_PME_CALC_F;
 602                     }
 603                     if (flags & GMX_FORCE_VIRIAL)
 604                     {
 605                         pme_flags |= GMX_PME_CALC_ENER_VIR;
 606                     }
 607                     if (fr->n_tpi > 0)
 608                     {
 609                         /* We don't calculate f, but we do want the potential */
 610                         pme_flags |= GMX_PME_CALC_POT;
 611                     }
 612                     wallcycle_start(wcycle, ewcPMEMESH);
 613                     status = gmx_pme_do(fr->pmedata,
 614                                         0, md->homenr - fr->n_tpi,
 615                                         x, fr->f_novirsum,
 616                                         md->chargeA, md->chargeB,
 617                                         md->sqrt_c6A, md->sqrt_c6B,
 618                                         md->sigmaA, md->sigmaB,
 619                                         bSB ? boxs : box, cr,
 620                                         DOMAINDECOMP(cr) ? dd_pme_maxshift_x(cr->dd) : 0,
 621                                         DOMAINDECOMP(cr) ? dd_pme_maxshift_y(cr->dd) : 0,
 622                                         nrnb, wcycle,
 623                                         fr->vir_el_recip, fr->ewaldcoeff_q,
 624                                         fr->vir_lj_recip, fr->ewaldcoeff_lj,
 625                                         &Vlr_q, &Vlr_lj,
 626                                         lambda[efptCOUL], lambda[efptVDW],
 627                                         &dvdl_long_range_q, &dvdl_long_range_lj, pme_flags);
 628                     *cycles_pme = wallcycle_stop(wcycle, ewcPMEMESH);
 629                     if (status != 0)
 630                     {
 631                         gmx_fatal(FARGS, "Error %d in reciprocal PME routine", status);
 632                     }
 633                     /* We should try to do as little computation after
 634                      * this as possible, because parallel PME synchronizes
 635                      * the nodes, so we want all load imbalance of the
 636                      * rest of the force calculation to be before the PME
 637                      * call.  DD load balancing is done on the whole time
 638                      * of the force call (without PME).
 639                      */
 640                 }
 641                 if (fr->n_tpi > 0)
 642                 {
 643                     if (EVDW_PME(ir->vdwtype))
 644                     {
 645
 646                         gmx_fatal(FARGS, "Test particle insertion not implemented with LJ-PME");
 647                     }
 648                     /* Determine the PME grid energy of the test molecule
 649                      * with the PME grid potential of the other charges.
 650                      */
 651                     gmx_pme_calc_energy(fr->pmedata, fr->n_tpi,
 652                                         x + md->homenr - fr->n_tpi,
 653                                         md->chargeA + md->homenr - fr->n_tpi,
 654                                         &Vlr_q);
 655                 }
 656                 PRINT_SEPDVDL("PME mesh", Vlr_q + Vlr_lj, dvdl_long_range_q+dvdl_long_range_lj);
 657             }
 658         }
 659
 660         if (!EEL_PME(fr->eeltype) && EEL_EWALD(fr->eeltype))
 661         {
 662             Vlr_q = do_ewald(ir, x, fr->f_novirsum,
 663                              md->chargeA, md->chargeB,
 664                              box_size, cr, md->homenr,
 665                              fr->vir_el_recip, fr->ewaldcoeff_q,
 666                              lambda[efptCOUL], &dvdl_long_range_q, fr->ewald_table);
 667             PRINT_SEPDVDL("Ewald long-range", Vlr_q, dvdl_long_range_q);
 668         }
 669
 670         /* Note that with separate PME nodes we get the real energies later */
 671         enerd->dvdl_lin[efptCOUL] += dvdl_long_range_q;
 672         enerd->dvdl_lin[efptVDW]  += dvdl_long_range_lj;
 673         enerd->term[F_COUL_RECIP]  = Vlr_q + Vcorr_q;
 674         enerd->term[F_LJ_RECIP]    = Vlr_lj + Vcorr_lj;
 675         if (debug)
 676         {
 677             fprintf(debug, "Vlr_q = %g, Vcorr_q = %g, Vlr_corr_q = %g\n",
 678                     Vlr_q, Vcorr_q, enerd->term[F_COUL_RECIP]);
 679             pr_rvecs(debug, 0, "vir_el_recip after corr", fr->vir_el_recip, DIM);
 680             pr_rvecs(debug, 0, "fshift after LR Corrections", fr->fshift, SHIFTS);
 681             fprintf(debug, "Vlr_lj: %g, Vcorr_lj = %g, Vlr_corr_lj = %g\n",
 682                     Vlr_lj, Vcorr_lj, enerd->term[F_LJ_RECIP]);
 683             pr_rvecs(debug, 0, "vir_lj_recip after corr", fr->vir_lj_recip, DIM);
 684         }
 685     }
 686     else
 687     {
 688         /* Is there a reaction-field exclusion correction needed? */
 689         if (EEL_RF(fr->eeltype) && eelRF_NEC != fr->eeltype)
 690         {
 691             /* With the Verlet scheme, exclusion forces are calculated
 692              * in the non-bonded kernel.
 693              */
 694             if (ir->cutoff_scheme != ecutsVERLET)
 695             {
 696                 real dvdl_rf_excl      = 0;
 697                 enerd->term[F_RF_EXCL] =
 698                     RF_excl_correction(fr, graph, md, excl, x, f,
 699                                        fr->fshift, &pbc, lambda[efptCOUL], &dvdl_rf_excl);
 700
 701                 enerd->dvdl_lin[efptCOUL] += dvdl_rf_excl;
 702                 PRINT_SEPDVDL("RF exclusion correction",
 703                               enerd->term[F_RF_EXCL], dvdl_rf_excl);
 704             }
 705         }
 706     }
 707     where();
 708     debug_gmx();
 709
 710     if (debug)
 711     {
 712         print_nrnb(debug, nrnb);
 713     }
 714     debug_gmx();
 715
 716 #ifdef GMX_MPI
 717     if (TAKETIME)
 718     {
 719         t2 = MPI_Wtime();
 720         MPI_Barrier(cr->mpi_comm_mygroup);
 721         t3          = MPI_Wtime();
 722         fr->t_wait += t3-t2;
 723         if (fr->timesteps == 11)
 724         {
 725             fprintf(stderr, "* PP load balancing info: node %d, step %s, rel wait time=%3.0f%% , load string value: %7.2f\n",
 726                     cr->nodeid, gmx_step_str(fr->timesteps, buf),
 727                     100*fr->t_wait/(fr->t_wait+fr->t_fnbf),
 728                     (fr->t_fnbf+fr->t_wait)/fr->t_fnbf);
 729         }
 730         fr->timesteps++;
 731     }
 732 #endif
 733
 734     if (debug)
 735     {
 736         pr_rvecs(debug, 0, "fshift after bondeds", fr->fshift, SHIFTS);
 737     }
 738
 739 }
 740
 741 void init_enerdata(int ngener, int n_lambda, gmx_enerdata_t *enerd)
 742 {
 743     int i, n2;
 744
 745     for (i = 0; i < F_NRE; i++)
 746     {
 747         enerd->term[i]         = 0;
 748         enerd->foreign_term[i] = 0;
 749     }
 750
 751
 752     for (i = 0; i < efptNR; i++)
 753     {
 754         enerd->dvdl_lin[i]     = 0;
 755         enerd->dvdl_nonlin[i]  = 0;
 756     }
 757
 758     n2 = ngener*ngener;
 759     if (debug)
 760     {
 761         fprintf(debug, "Creating %d sized group matrix for energies\n", n2);
 762     }
 763     enerd->grpp.nener         = n2;
 764     enerd->foreign_grpp.nener = n2;
 765     for (i = 0; (i < egNR); i++)
 766     {
 767         snew(enerd->grpp.ener[i], n2);
 768         snew(enerd->foreign_grpp.ener[i], n2);
 769     }
 770
 771     if (n_lambda)
 772     {
 773         enerd->n_lambda = 1 + n_lambda;
 774         snew(enerd->enerpart_lambda, enerd->n_lambda);
 775     }
 776     else
 777     {
 778         enerd->n_lambda = 0;
 779     }
 780 }
 781
 782 void destroy_enerdata(gmx_enerdata_t *enerd)
 783 {
 784     int i;
 785
 786     for (i = 0; (i < egNR); i++)
 787     {
 788         sfree(enerd->grpp.ener[i]);
 789     }
 790
 791     for (i = 0; (i < egNR); i++)
 792     {
 793         sfree(enerd->foreign_grpp.ener[i]);
 794     }
 795
 796     if (enerd->n_lambda)
 797     {
 798         sfree(enerd->enerpart_lambda);
 799     }
 800 }
 801
 802 static real sum_v(int n, real v[])
 803 {
 804     real t;
 805     int  i;
 806
 807     t = 0.0;
 808     for (i = 0; (i < n); i++)
 809     {
 810         t = t + v[i];
 811     }
 812
 813     return t;
 814 }
 815
 816 void sum_epot(gmx_grppairener_t *grpp, real *epot)
 817 {
 818     int i;
 819
 820     /* Accumulate energies */
 821     epot[F_COUL_SR]  = sum_v(grpp->nener, grpp->ener[egCOULSR]);
 822     epot[F_LJ]       = sum_v(grpp->nener, grpp->ener[egLJSR]);
 823     epot[F_LJ14]     = sum_v(grpp->nener, grpp->ener[egLJ14]);
 824     epot[F_COUL14]   = sum_v(grpp->nener, grpp->ener[egCOUL14]);
 825     epot[F_COUL_LR]  = sum_v(grpp->nener, grpp->ener[egCOULLR]);
 826     epot[F_LJ_LR]    = sum_v(grpp->nener, grpp->ener[egLJLR]);
 827     /* We have already added 1-2,1-3, and 1-4 terms to F_GBPOL */
 828     epot[F_GBPOL]   += sum_v(grpp->nener, grpp->ener[egGB]);
 829
 830 /* lattice part of LR doesnt belong to any group
 831  * and has been added earlier
 832  */
 833     epot[F_BHAM]     = sum_v(grpp->nener, grpp->ener[egBHAMSR]);
 834     epot[F_BHAM_LR]  = sum_v(grpp->nener, grpp->ener[egBHAMLR]);
 835
 836     epot[F_EPOT] = 0;
 837     for (i = 0; (i < F_EPOT); i++)
 838     {
 839         if (i != F_DISRESVIOL && i != F_ORIRESDEV)
 840         {
 841             epot[F_EPOT] += epot[i];
 842         }
 843     }
 844 }
 845
 846 void sum_dhdl(gmx_enerdata_t *enerd, real *lambda, t_lambda *fepvals)
 847 {
 848     int    i, j, index;
 849     double dlam;
 850
 851     enerd->dvdl_lin[efptVDW] += enerd->term[F_DVDL_VDW];  /* include dispersion correction */
 852     enerd->term[F_DVDL]       = 0.0;
 853     for (i = 0; i < efptNR; i++)
 854     {
 855         if (fepvals->separate_dvdl[i])
 856         {
 857             /* could this be done more readably/compactly? */
 858             switch (i)
 859             {
 860                 case (efptMASS):
 861                     index = F_DKDL;
 862                     break;
 863                 case (efptCOUL):
 864                     index = F_DVDL_COUL;
 865                     break;
 866                 case (efptVDW):
 867                     index = F_DVDL_VDW;
 868                     break;
 869                 case (efptBONDED):
 870                     index = F_DVDL_BONDED;
 871                     break;
 872                 case (efptRESTRAINT):
 873                     index = F_DVDL_RESTRAINT;
 874                     break;
 875                 default:
 876                     index = F_DVDL;
 877                     break;
 878             }
 879             enerd->term[index] = enerd->dvdl_lin[i] + enerd->dvdl_nonlin[i];
 880             if (debug)
 881             {
 882                 fprintf(debug, "dvdl-%s[%2d]: %f: non-linear %f + linear %f\n",
 883                         efpt_names[i], i, enerd->term[index], enerd->dvdl_nonlin[i], enerd->dvdl_lin[i]);
 884             }
 885         }
 886         else
 887         {
 888             enerd->term[F_DVDL] += enerd->dvdl_lin[i] + enerd->dvdl_nonlin[i];
 889             if (debug)
 890             {
 891                 fprintf(debug, "dvd-%sl[%2d]: %f: non-linear %f + linear %f\n",
 892                         efpt_names[0], i, enerd->term[F_DVDL], enerd->dvdl_nonlin[i], enerd->dvdl_lin[i]);
 893             }
 894         }
 895     }
 896
 897     /* Notes on the foreign lambda free energy difference evaluation:
 898      * Adding the potential and ekin terms that depend linearly on lambda
 899      * as delta lam * dvdl to the energy differences is exact.
 900      * For the constraints this is not exact, but we have no other option
 901      * without literally changing the lengths and reevaluating the energies at each step.
 902      * (try to remedy this post 4.6 - MRS)
 903      * For the non-bonded LR term we assume that the soft-core (if present)
 904      * no longer affects the energy beyond the short-range cut-off,
 905      * which is a very good approximation (except for exotic settings).
 906      * (investigate how to overcome this post 4.6 - MRS)
 907      */
 908     if (fepvals->separate_dvdl[efptBONDED])
 909     {
 910         enerd->term[F_DVDL_BONDED] += enerd->term[F_DVDL_CONSTR];
 911     }
 912     else
 913     {
 914         enerd->term[F_DVDL] += enerd->term[F_DVDL_CONSTR];
 915     }
 916     enerd->term[F_DVDL_CONSTR] = 0;
 917
 918     for (i = 0; i < fepvals->n_lambda; i++)
 919     {
 920         /* note we are iterating over fepvals here!
 921            For the current lam, dlam = 0 automatically,
 922            so we don't need to add anything to the
 923            enerd->enerpart_lambda[0] */
 924
 925         /* we don't need to worry about dvdl_lin contributions to dE at
 926            current lambda, because the contributions to the current
 927            lambda are automatically zeroed */
 928
 929         for (j = 0; j < efptNR; j++)
 930         {
 931             /* Note that this loop is over all dhdl components, not just the separated ones */
 932             dlam = (fepvals->all_lambda[j][i]-lambda[j]);
 933             enerd->enerpart_lambda[i+1] += dlam*enerd->dvdl_lin[j];
 934             if (debug)
 935             {
 936                 fprintf(debug, "enerdiff lam %g: (%15s), non-linear %f linear %f*%f\n",
 937                         fepvals->all_lambda[j][i], efpt_names[j],
 938                         (enerd->enerpart_lambda[i+1] - enerd->enerpart_lambda[0]),
 939                         dlam, enerd->dvdl_lin[j]);
 940             }
 941         }
 942     }
 943 }
 944
 945
 946 void reset_foreign_enerdata(gmx_enerdata_t *enerd)
 947 {
 948     int  i, j;
 949
 950     /* First reset all foreign energy components.  Foreign energies always called on
 951        neighbor search steps */
 952     for (i = 0; (i < egNR); i++)
 953     {
 954         for (j = 0; (j < enerd->grpp.nener); j++)
 955         {
 956             enerd->foreign_grpp.ener[i][j] = 0.0;
 957         }
 958     }
 959
 960     /* potential energy components */
 961     for (i = 0; (i <= F_EPOT); i++)
 962     {
 963         enerd->foreign_term[i] = 0.0;
 964     }
 965 }
 966
 967 void reset_enerdata(t_forcerec *fr, gmx_bool bNS,
 968                     gmx_enerdata_t *enerd,
 969                     gmx_bool bMaster)
 970 {
 971     gmx_bool bKeepLR;
 972     int      i, j;
 973
 974     /* First reset all energy components, except for the long range terms
 975      * on the master at non neighbor search steps, since the long range
 976      * terms have already been summed at the last neighbor search step.
 977      */
 978     bKeepLR = (fr->bTwinRange && !bNS);
 979     for (i = 0; (i < egNR); i++)
 980     {
 981         if (!(bKeepLR && bMaster && (i == egCOULLR || i == egLJLR)))
 982         {
 983             for (j = 0; (j < enerd->grpp.nener); j++)
 984             {
 985                 enerd->grpp.ener[i][j] = 0.0;
 986             }
 987         }
 988     }
 989     for (i = 0; i < efptNR; i++)
 990     {
 991         enerd->dvdl_lin[i]    = 0.0;
 992         enerd->dvdl_nonlin[i] = 0.0;
 993     }
 994
 995     /* Normal potential energy components */
 996     for (i = 0; (i <= F_EPOT); i++)
 997     {
 998         enerd->term[i] = 0.0;
 999     }
1000     /* Initialize the dVdlambda term with the long range contribution */
1001     /* Initialize the dvdl term with the long range contribution */
1002     enerd->term[F_DVDL]            = 0.0;
1003     enerd->term[F_DVDL_COUL]       = 0.0;
1004     enerd->term[F_DVDL_VDW]        = 0.0;
1005     enerd->term[F_DVDL_BONDED]     = 0.0;
1006     enerd->term[F_DVDL_RESTRAINT]  = 0.0;
1007     enerd->term[F_DKDL]            = 0.0;
1008     if (enerd->n_lambda > 0)
1009     {
1010         for (i = 0; i < enerd->n_lambda; i++)
1011         {
1012             enerd->enerpart_lambda[i] = 0.0;
1013         }
1014     }
1015     /* reset foreign energy data - separate function since we also call it elsewhere */
1016     reset_foreign_enerdata(enerd);
1017 }