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 #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/legacyheaders/coulomb.h"
  48 #include "gromacs/legacyheaders/domdec.h"
  49 #include "gromacs/legacyheaders/gmx_omp_nthreads.h"
  50 #include "gromacs/legacyheaders/macros.h"
  51 #include "gromacs/legacyheaders/mdrun.h"
  52 #include "gromacs/legacyheaders/names.h"
  53 #include "gromacs/legacyheaders/network.h"
  54 #include "gromacs/legacyheaders/nonbonded.h"
  55 #include "gromacs/legacyheaders/nrnb.h"
  56 #include "gromacs/legacyheaders/ns.h"
  57 #include "gromacs/legacyheaders/pme.h"
  58 #include "gromacs/legacyheaders/qmmm.h"
  59 #include "gromacs/legacyheaders/txtdump.h"
  60 #include "gromacs/legacyheaders/typedefs.h"
  61 #include "gromacs/legacyheaders/types/commrec.h"
  62 #include "gromacs/listed-forces/bonded.h"
  63 #include "gromacs/math/vec.h"
  64 #include "gromacs/pbcutil/ishift.h"
  65 #include "gromacs/pbcutil/mshift.h"
  66 #include "gromacs/pbcutil/pbc.h"
  67 #include "gromacs/timing/wallcycle.h"
  68 #include "gromacs/utility/fatalerror.h"
  69 #include "gromacs/utility/smalloc.h"
  70
  71 void ns(FILE              *fp,
  72         t_forcerec        *fr,
  73         matrix             box,
  74         gmx_groups_t      *groups,
  75         gmx_localtop_t    *top,
  76         t_mdatoms         *md,
  77         t_commrec         *cr,
  78         t_nrnb            *nrnb,
  79         gmx_bool           bFillGrid,
  80         gmx_bool           bDoLongRangeNS)
  81 {
  82     char   *ptr;
  83     int     nsearch;
  84
  85
  86     if (!fr->ns.nblist_initialized)
  87     {
  88         init_neighbor_list(fp, fr, md->homenr);
  89     }
  90
  91     if (fr->bTwinRange)
  92     {
  93         fr->nlr = 0;
  94     }
  95
  96     nsearch = search_neighbours(fp, fr, box, top, groups, cr, nrnb, md,
  97                                 bFillGrid, bDoLongRangeNS);
  98     if (debug)
  99     {
 100         fprintf(debug, "nsearch = %d\n", nsearch);
 101     }
 102
 103     /* Check whether we have to do dynamic load balancing */
 104     /*if ((nsb->nstDlb > 0) && (mod(step,nsb->nstDlb) == 0))
 105        count_nb(cr,nsb,&(top->blocks[ebCGS]),nns,fr->nlr,
 106        &(top->idef),opts->ngener);
 107      */
 108     if (fr->ns.dump_nl > 0)
 109     {
 110         dump_nblist(fp, cr, fr, fr->ns.dump_nl);
 111     }
 112 }
 113
 114 static void reduce_thread_forces(int n, rvec *f,
 115                                  tensor vir_q, tensor vir_lj,
 116                                  real *Vcorr_q, real *Vcorr_lj,
 117                                  real *dvdl_q, real *dvdl_lj,
 118                                  int nthreads, f_thread_t *f_t)
 119 {
 120     int t, i;
 121
 122     /* This reduction can run over any number of threads */
 123 #pragma omp parallel for num_threads(gmx_omp_nthreads_get(emntBonded)) private(t) schedule(static)
 124     for (i = 0; i < n; i++)
 125     {
 126         for (t = 1; t < nthreads; t++)
 127         {
 128             rvec_inc(f[i], f_t[t].f[i]);
 129         }
 130     }
 131     for (t = 1; t < nthreads; t++)
 132     {
 133         *Vcorr_q  += f_t[t].Vcorr_q;
 134         *Vcorr_lj += f_t[t].Vcorr_lj;
 135         *dvdl_q   += f_t[t].dvdl[efptCOUL];
 136         *dvdl_lj  += f_t[t].dvdl[efptVDW];
 137         m_add(vir_q, f_t[t].vir_q, vir_q);
 138         m_add(vir_lj, f_t[t].vir_lj, vir_lj);
 139     }
 140 }
 141
 142 void do_force_lowlevel(t_forcerec *fr,      t_inputrec *ir,
 143                        t_idef     *idef,    t_commrec  *cr,
 144                        t_nrnb     *nrnb,    gmx_wallcycle_t wcycle,
 145                        t_mdatoms  *md,
 146                        rvec       x[],      history_t  *hist,
 147                        rvec       f[],
 148                        rvec       f_longrange[],
 149                        gmx_enerdata_t *enerd,
 150                        t_fcdata   *fcd,
 151                        gmx_localtop_t *top,
 152                        gmx_genborn_t *born,
 153                        gmx_bool       bBornRadii,
 154                        matrix     box,
 155                        t_lambda   *fepvals,
 156                        real       *lambda,
 157                        t_graph    *graph,
 158                        t_blocka   *excl,
 159                        rvec       mu_tot[],
 160                        int        flags,
 161                        float      *cycles_pme)
 162 {
 163     int         i, j;
 164     int         donb_flags;
 165     gmx_bool    bDoEpot, bSB;
 166     int         pme_flags;
 167     matrix      boxs;
 168     rvec        box_size;
 169     t_pbc       pbc;
 170     char        buf[22];
 171     double      clam_i, vlam_i;
 172     real        dvdl_dum[efptNR], dvdl_nb[efptNR], lam_i[efptNR];
 173     real        dvdl_q, dvdl_lj;
 174
 175 #ifdef GMX_MPI
 176     double  t0 = 0.0, t1, t2, t3; /* time measurement for coarse load balancing */
 177 #endif
 178
 179     set_pbc(&pbc, fr->ePBC, box);
 180
 181     /* reset free energy components */
 182     for (i = 0; i < efptNR; i++)
 183     {
 184         dvdl_nb[i]  = 0;
 185         dvdl_dum[i] = 0;
 186     }
 187
 188     /* Reset box */
 189     for (i = 0; (i < DIM); i++)
 190     {
 191         box_size[i] = box[i][i];
 192     }
 193
 194     debug_gmx();
 195
 196     /* do QMMM first if requested */
 197     if (fr->bQMMM)
 198     {
 199         enerd->term[F_EQM] = calculate_QMMM(cr, x, f, fr);
 200     }
 201
 202     /* Call the short range functions all in one go. */
 203
 204 #ifdef GMX_MPI
 205     /*#define TAKETIME ((cr->npmenodes) && (fr->timesteps < 12))*/
 206 #define TAKETIME FALSE
 207     if (TAKETIME)
 208     {
 209         MPI_Barrier(cr->mpi_comm_mygroup);
 210         t0 = MPI_Wtime();
 211     }
 212 #endif
 213
 214     if (ir->nwall)
 215     {
 216         /* foreign lambda component for walls */
 217         real dvdl_walls = do_walls(ir, fr, box, md, x, f, lambda[efptVDW],
 218                                    enerd->grpp.ener[egLJSR], nrnb);
 219         enerd->dvdl_lin[efptVDW] += dvdl_walls;
 220     }
 221
 222     /* If doing GB, reset dvda and calculate the Born radii */
 223     if (ir->implicit_solvent)
 224     {
 225         wallcycle_sub_start(wcycle, ewcsNONBONDED);
 226
 227         for (i = 0; i < born->nr; i++)
 228         {
 229             fr->dvda[i] = 0;
 230         }
 231
 232         if (bBornRadii)
 233         {
 234             calc_gb_rad(cr, fr, ir, top, x, &(fr->gblist), born, md, nrnb);
 235         }
 236
 237         wallcycle_sub_stop(wcycle, ewcsNONBONDED);
 238     }
 239
 240     where();
 241     /* We only do non-bonded calculation with group scheme here, the verlet
 242      * calls are done from do_force_cutsVERLET(). */
 243     if (fr->cutoff_scheme == ecutsGROUP && (flags & GMX_FORCE_NONBONDED))
 244     {
 245         donb_flags = 0;
 246         /* Add short-range interactions */
 247         donb_flags |= GMX_NONBONDED_DO_SR;
 248
 249         /* Currently all group scheme kernels always calculate (shift-)forces */
 250         if (flags & GMX_FORCE_FORCES)
 251         {
 252             donb_flags |= GMX_NONBONDED_DO_FORCE;
 253         }
 254         if (flags & GMX_FORCE_VIRIAL)
 255         {
 256             donb_flags |= GMX_NONBONDED_DO_SHIFTFORCE;
 257         }
 258         if (flags & GMX_FORCE_ENERGY)
 259         {
 260             donb_flags |= GMX_NONBONDED_DO_POTENTIAL;
 261         }
 262         if (flags & GMX_FORCE_DO_LR)
 263         {
 264             donb_flags |= GMX_NONBONDED_DO_LR;
 265         }
 266
 267         wallcycle_sub_start(wcycle, ewcsNONBONDED);
 268         do_nonbonded(fr, x, f, f_longrange, md, excl,
 269                      &enerd->grpp, nrnb,
 270                      lambda, dvdl_nb, -1, -1, donb_flags);
 271
 272         /* If we do foreign lambda and we have soft-core interactions
 273          * we have to recalculate the (non-linear) energies contributions.
 274          */
 275         if (fepvals->n_lambda > 0 && (flags & GMX_FORCE_DHDL) && fepvals->sc_alpha != 0)
 276         {
 277             for (i = 0; i < enerd->n_lambda; i++)
 278             {
 279                 for (j = 0; j < efptNR; j++)
 280                 {
 281                     lam_i[j] = (i == 0 ? lambda[j] : fepvals->all_lambda[j][i-1]);
 282                 }
 283                 reset_foreign_enerdata(enerd);
 284                 do_nonbonded(fr, x, f, f_longrange, md, excl,
 285                              &(enerd->foreign_grpp), nrnb,
 286                              lam_i, dvdl_dum, -1, -1,
 287                              (donb_flags & ~GMX_NONBONDED_DO_FORCE) | GMX_NONBONDED_DO_FOREIGNLAMBDA);
 288                 sum_epot(&(enerd->foreign_grpp), enerd->foreign_term);
 289                 enerd->enerpart_lambda[i] += enerd->foreign_term[F_EPOT];
 290             }
 291         }
 292         wallcycle_sub_stop(wcycle, ewcsNONBONDED);
 293         where();
 294     }
 295
 296     /* If we are doing GB, calculate bonded forces and apply corrections
 297      * to the solvation forces */
 298     /* MRS: Eventually, many need to include free energy contribution here! */
 299     if (ir->implicit_solvent)
 300     {
 301         wallcycle_sub_start(wcycle, ewcsLISTED);
 302         calc_gb_forces(cr, md, born, top, x, f, fr, idef,
 303                        ir->gb_algorithm, ir->sa_algorithm, nrnb, &pbc, graph, enerd);
 304         wallcycle_sub_stop(wcycle, ewcsLISTED);
 305     }
 306
 307 #ifdef GMX_MPI
 308     if (TAKETIME)
 309     {
 310         t1          = MPI_Wtime();
 311         fr->t_fnbf += t1-t0;
 312     }
 313 #endif
 314
 315     if (fepvals->sc_alpha != 0)
 316     {
 317         enerd->dvdl_nonlin[efptVDW] += dvdl_nb[efptVDW];
 318     }
 319     else
 320     {
 321         enerd->dvdl_lin[efptVDW] += dvdl_nb[efptVDW];
 322     }
 323
 324     if (fepvals->sc_alpha != 0)
 325
 326     /* even though coulomb part is linear, we already added it, beacuse we
 327        need to go through the vdw calculation anyway */
 328     {
 329         enerd->dvdl_nonlin[efptCOUL] += dvdl_nb[efptCOUL];
 330     }
 331     else
 332     {
 333         enerd->dvdl_lin[efptCOUL] += dvdl_nb[efptCOUL];
 334     }
 335
 336     debug_gmx();
 337
 338
 339     if (debug)
 340     {
 341         pr_rvecs(debug, 0, "fshift after SR", fr->fshift, SHIFTS);
 342     }
 343
 344     /* Shift the coordinates. Must be done before listed forces and PPPM,
 345      * but is also necessary for SHAKE and update, therefore it can NOT
 346      * go when no listed forces have to be evaluated.
 347      */
 348
 349     /* Here sometimes we would not need to shift with NBFonly,
 350      * but we do so anyhow for consistency of the returned coordinates.
 351      */
 352     if (graph)
 353     {
 354         shift_self(graph, box, x);
 355         if (TRICLINIC(box))
 356         {
 357             inc_nrnb(nrnb, eNR_SHIFTX, 2*graph->nnodes);
 358         }
 359         else
 360         {
 361             inc_nrnb(nrnb, eNR_SHIFTX, graph->nnodes);
 362         }
 363     }
 364     /* Check whether we need to do listed interactions or correct for exclusions */
 365     if (fr->bMolPBC &&
 366         ((flags & GMX_FORCE_LISTED)
 367          || EEL_RF(fr->eeltype) || EEL_FULL(fr->eeltype) || EVDW_PME(fr->vdwtype)))
 368     {
 369         /* Since all atoms are in the rectangular or triclinic unit-cell,
 370          * only single box vector shifts (2 in x) are required.
 371          */
 372         set_pbc_dd(&pbc, fr->ePBC, cr->dd, TRUE, box);
 373     }
 374     debug_gmx();
 375
 376     if (flags & GMX_FORCE_LISTED)
 377     {
 378         wallcycle_sub_start(wcycle, ewcsLISTED);
 379         calc_bonds(cr->ms,
 380                    idef, (const rvec *) x, hist, f, fr, &pbc, graph, enerd, nrnb, lambda, md, fcd,
 381                    DOMAINDECOMP(cr) ? cr->dd->gatindex : NULL,
 382                    flags);
 383
 384         /* Check if we have to determine energy differences
 385          * at foreign lambda's.
 386          */
 387         if (fepvals->n_lambda > 0 && (flags & GMX_FORCE_DHDL) &&
 388             idef->ilsort != ilsortNO_FE)
 389         {
 390             if (idef->ilsort != ilsortFE_SORTED)
 391             {
 392                 gmx_incons("The bonded interactions are not sorted for free energy");
 393             }
 394             for (i = 0; i < enerd->n_lambda; i++)
 395             {
 396                 reset_foreign_enerdata(enerd);
 397                 for (j = 0; j < efptNR; j++)
 398                 {
 399                     lam_i[j] = (i == 0 ? lambda[j] : fepvals->all_lambda[j][i-1]);
 400                 }
 401                 calc_bonds_lambda(idef, (const rvec *) x, fr, &pbc, graph, &(enerd->foreign_grpp), enerd->foreign_term, nrnb, lam_i, md,
 402                                   fcd, DOMAINDECOMP(cr) ? cr->dd->gatindex : NULL);
 403                 sum_epot(&(enerd->foreign_grpp), enerd->foreign_term);
 404                 enerd->enerpart_lambda[i] += enerd->foreign_term[F_EPOT];
 405             }
 406         }
 407         debug_gmx();
 408
 409         wallcycle_sub_stop(wcycle, ewcsLISTED);
 410     }
 411
 412     where();
 413
 414     *cycles_pme = 0;
 415     clear_mat(fr->vir_el_recip);
 416     clear_mat(fr->vir_lj_recip);
 417
 418     /* Do long-range electrostatics and/or LJ-PME, including related short-range
 419      * corrections.
 420      */
 421     if (EEL_FULL(fr->eeltype) || EVDW_PME(fr->vdwtype))
 422     {
 423         real Vlr               = 0, Vcorr = 0;
 424         real dvdl_long_range   = 0;
 425         int  status            = 0;
 426         real Vlr_q             = 0, Vlr_lj = 0, Vcorr_q = 0, Vcorr_lj = 0;
 427         real dvdl_long_range_q = 0, dvdl_long_range_lj = 0;
 428
 429         bSB = (ir->nwall == 2);
 430         if (bSB)
 431         {
 432             copy_mat(box, boxs);
 433             svmul(ir->wall_ewald_zfac, boxs[ZZ], boxs[ZZ]);
 434             box_size[ZZ] *= ir->wall_ewald_zfac;
 435         }
 436
 437         if (EEL_PME_EWALD(fr->eeltype) || EVDW_PME(fr->vdwtype))
 438         {
 439             real dvdl_long_range_correction_q   = 0;
 440             real dvdl_long_range_correction_lj  = 0;
 441             /* With the Verlet scheme exclusion forces are calculated
 442              * in the non-bonded kernel.
 443              */
 444             /* The TPI molecule does not have exclusions with the rest
 445              * of the system and no intra-molecular PME grid
 446              * contributions will be calculated in
 447              * gmx_pme_calc_energy.
 448              */
 449             if ((ir->cutoff_scheme == ecutsGROUP && fr->n_tpi == 0) ||
 450                 ir->ewald_geometry != eewg3D ||
 451                 ir->epsilon_surface != 0)
 452             {
 453                 int nthreads, t;
 454
 455                 wallcycle_sub_start(wcycle, ewcsEWALD_CORRECTION);
 456
 457                 if (fr->n_tpi > 0)
 458                 {
 459                     gmx_fatal(FARGS, "TPI with PME currently only works in a 3D geometry with tin-foil boundary conditions");
 460                 }
 461
 462                 nthreads = gmx_omp_nthreads_get(emntBonded);
 463 #pragma omp parallel for num_threads(nthreads) schedule(static)
 464                 for (t = 0; t < nthreads; t++)
 465                 {
 466                     int     s, e, i;
 467                     rvec   *fnv;
 468                     tensor *vir_q, *vir_lj;
 469                     real   *Vcorrt_q, *Vcorrt_lj, *dvdlt_q, *dvdlt_lj;
 470                     if (t == 0)
 471                     {
 472                         fnv       = fr->f_novirsum;
 473                         vir_q     = &fr->vir_el_recip;
 474                         vir_lj    = &fr->vir_lj_recip;
 475                         Vcorrt_q  = &Vcorr_q;
 476                         Vcorrt_lj = &Vcorr_lj;
 477                         dvdlt_q   = &dvdl_long_range_correction_q;
 478                         dvdlt_lj  = &dvdl_long_range_correction_lj;
 479                     }
 480                     else
 481                     {
 482                         fnv       = fr->f_t[t].f;
 483                         vir_q     = &fr->f_t[t].vir_q;
 484                         vir_lj    = &fr->f_t[t].vir_lj;
 485                         Vcorrt_q  = &fr->f_t[t].Vcorr_q;
 486                         Vcorrt_lj = &fr->f_t[t].Vcorr_lj;
 487                         dvdlt_q   = &fr->f_t[t].dvdl[efptCOUL];
 488                         dvdlt_lj  = &fr->f_t[t].dvdl[efptVDW];
 489                         for (i = 0; i < fr->natoms_force; i++)
 490                         {
 491                             clear_rvec(fnv[i]);
 492                         }
 493                         clear_mat(*vir_q);
 494                         clear_mat(*vir_lj);
 495                     }
 496                     *dvdlt_q  = 0;
 497                     *dvdlt_lj = 0;
 498
 499                     ewald_LRcorrection(fr->excl_load[t], fr->excl_load[t+1],
 500                                        cr, t, fr,
 501                                        md->chargeA,
 502                                        md->nChargePerturbed ? md->chargeB : NULL,
 503                                        md->sqrt_c6A,
 504                                        md->nTypePerturbed ? md->sqrt_c6B : NULL,
 505                                        md->sigmaA,
 506                                        md->nTypePerturbed ? md->sigmaB : NULL,
 507                                        md->sigma3A,
 508                                        md->nTypePerturbed ? md->sigma3B : NULL,
 509                                        ir->cutoff_scheme != ecutsVERLET,
 510                                        excl, x, bSB ? boxs : box, mu_tot,
 511                                        ir->ewald_geometry,
 512                                        ir->epsilon_surface,
 513                                        fnv, *vir_q, *vir_lj,
 514                                        Vcorrt_q, Vcorrt_lj,
 515                                        lambda[efptCOUL], lambda[efptVDW],
 516                                        dvdlt_q, dvdlt_lj);
 517                 }
 518                 if (nthreads > 1)
 519                 {
 520                     reduce_thread_forces(fr->natoms_force, fr->f_novirsum,
 521                                          fr->vir_el_recip, fr->vir_lj_recip,
 522                                          &Vcorr_q, &Vcorr_lj,
 523                                          &dvdl_long_range_correction_q,
 524                                          &dvdl_long_range_correction_lj,
 525                                          nthreads, fr->f_t);
 526                 }
 527                 wallcycle_sub_stop(wcycle, ewcsEWALD_CORRECTION);
 528             }
 529
 530             if (EEL_PME_EWALD(fr->eeltype) && fr->n_tpi == 0)
 531             {
 532                 Vcorr_q += ewald_charge_correction(cr, fr, lambda[efptCOUL], box,
 533                                                    &dvdl_long_range_correction_q,
 534                                                    fr->vir_el_recip);
 535             }
 536
 537             enerd->dvdl_lin[efptCOUL] += dvdl_long_range_correction_q;
 538             enerd->dvdl_lin[efptVDW]  += dvdl_long_range_correction_lj;
 539
 540             if ((EEL_PME(fr->eeltype) || EVDW_PME(fr->vdwtype)) && (cr->duty & DUTY_PME))
 541             {
 542                 /* Do reciprocal PME for Coulomb and/or LJ. */
 543                 assert(fr->n_tpi >= 0);
 544                 if (fr->n_tpi == 0 || (flags & GMX_FORCE_STATECHANGED))
 545                 {
 546                     pme_flags = GMX_PME_SPREAD | GMX_PME_SOLVE;
 547                     if (EEL_PME(fr->eeltype))
 548                     {
 549                         pme_flags     |= GMX_PME_DO_COULOMB;
 550                     }
 551                     if (EVDW_PME(fr->vdwtype))
 552                     {
 553                         pme_flags |= GMX_PME_DO_LJ;
 554                     }
 555                     if (flags & GMX_FORCE_FORCES)
 556                     {
 557                         pme_flags |= GMX_PME_CALC_F;
 558                     }
 559                     if (flags & GMX_FORCE_VIRIAL)
 560                     {
 561                         pme_flags |= GMX_PME_CALC_ENER_VIR;
 562                     }
 563                     if (fr->n_tpi > 0)
 564                     {
 565                         /* We don't calculate f, but we do want the potential */
 566                         pme_flags |= GMX_PME_CALC_POT;
 567                     }
 568                     wallcycle_start(wcycle, ewcPMEMESH);
 569                     status = gmx_pme_do(fr->pmedata,
 570                                         0, md->homenr - fr->n_tpi,
 571                                         x, fr->f_novirsum,
 572                                         md->chargeA, md->chargeB,
 573                                         md->sqrt_c6A, md->sqrt_c6B,
 574                                         md->sigmaA, md->sigmaB,
 575                                         bSB ? boxs : box, cr,
 576                                         DOMAINDECOMP(cr) ? dd_pme_maxshift_x(cr->dd) : 0,
 577                                         DOMAINDECOMP(cr) ? dd_pme_maxshift_y(cr->dd) : 0,
 578                                         nrnb, wcycle,
 579                                         fr->vir_el_recip, fr->ewaldcoeff_q,
 580                                         fr->vir_lj_recip, fr->ewaldcoeff_lj,
 581                                         &Vlr_q, &Vlr_lj,
 582                                         lambda[efptCOUL], lambda[efptVDW],
 583                                         &dvdl_long_range_q, &dvdl_long_range_lj, pme_flags);
 584                     *cycles_pme = wallcycle_stop(wcycle, ewcPMEMESH);
 585                     if (status != 0)
 586                     {
 587                         gmx_fatal(FARGS, "Error %d in reciprocal PME routine", status);
 588                     }
 589                     /* We should try to do as little computation after
 590                      * this as possible, because parallel PME synchronizes
 591                      * the nodes, so we want all load imbalance of the
 592                      * rest of the force calculation to be before the PME
 593                      * call.  DD load balancing is done on the whole time
 594                      * of the force call (without PME).
 595                      */
 596                 }
 597                 if (fr->n_tpi > 0)
 598                 {
 599                     if (EVDW_PME(ir->vdwtype))
 600                     {
 601
 602                         gmx_fatal(FARGS, "Test particle insertion not implemented with LJ-PME");
 603                     }
 604                     /* Determine the PME grid energy of the test molecule
 605                      * with the PME grid potential of the other charges.
 606                      */
 607                     gmx_pme_calc_energy(fr->pmedata, fr->n_tpi,
 608                                         x + md->homenr - fr->n_tpi,
 609                                         md->chargeA + md->homenr - fr->n_tpi,
 610                                         &Vlr_q);
 611                 }
 612             }
 613         }
 614
 615         if (!EEL_PME(fr->eeltype) && EEL_PME_EWALD(fr->eeltype))
 616         {
 617             Vlr_q = do_ewald(ir, x, fr->f_novirsum,
 618                              md->chargeA, md->chargeB,
 619                              box_size, cr, md->homenr,
 620                              fr->vir_el_recip, fr->ewaldcoeff_q,
 621                              lambda[efptCOUL], &dvdl_long_range_q, fr->ewald_table);
 622         }
 623
 624         /* Note that with separate PME nodes we get the real energies later */
 625         enerd->dvdl_lin[efptCOUL] += dvdl_long_range_q;
 626         enerd->dvdl_lin[efptVDW]  += dvdl_long_range_lj;
 627         enerd->term[F_COUL_RECIP]  = Vlr_q + Vcorr_q;
 628         enerd->term[F_LJ_RECIP]    = Vlr_lj + Vcorr_lj;
 629         if (debug)
 630         {
 631             fprintf(debug, "Vlr_q = %g, Vcorr_q = %g, Vlr_corr_q = %g\n",
 632                     Vlr_q, Vcorr_q, enerd->term[F_COUL_RECIP]);
 633             pr_rvecs(debug, 0, "vir_el_recip after corr", fr->vir_el_recip, DIM);
 634             pr_rvecs(debug, 0, "fshift after LR Corrections", fr->fshift, SHIFTS);
 635             fprintf(debug, "Vlr_lj: %g, Vcorr_lj = %g, Vlr_corr_lj = %g\n",
 636                     Vlr_lj, Vcorr_lj, enerd->term[F_LJ_RECIP]);
 637             pr_rvecs(debug, 0, "vir_lj_recip after corr", fr->vir_lj_recip, DIM);
 638         }
 639     }
 640     else
 641     {
 642         /* Is there a reaction-field exclusion correction needed? */
 643         if (EEL_RF(fr->eeltype) && eelRF_NEC != fr->eeltype)
 644         {
 645             /* With the Verlet scheme, exclusion forces are calculated
 646              * in the non-bonded kernel.
 647              */
 648             if (ir->cutoff_scheme != ecutsVERLET)
 649             {
 650                 real dvdl_rf_excl      = 0;
 651                 enerd->term[F_RF_EXCL] =
 652                     RF_excl_correction(fr, graph, md, excl, x, f,
 653                                        fr->fshift, &pbc, lambda[efptCOUL], &dvdl_rf_excl);
 654
 655                 enerd->dvdl_lin[efptCOUL] += dvdl_rf_excl;
 656             }
 657         }
 658     }
 659     where();
 660     debug_gmx();
 661
 662     if (debug)
 663     {
 664         print_nrnb(debug, nrnb);
 665     }
 666     debug_gmx();
 667
 668 #ifdef GMX_MPI
 669     if (TAKETIME)
 670     {
 671         t2 = MPI_Wtime();
 672         MPI_Barrier(cr->mpi_comm_mygroup);
 673         t3          = MPI_Wtime();
 674         fr->t_wait += t3-t2;
 675         if (fr->timesteps == 11)
 676         {
 677             fprintf(stderr, "* PP load balancing info: rank %d, step %s, rel wait time=%3.0f%% , load string value: %7.2f\n",
 678                     cr->nodeid, gmx_step_str(fr->timesteps, buf),
 679                     100*fr->t_wait/(fr->t_wait+fr->t_fnbf),
 680                     (fr->t_fnbf+fr->t_wait)/fr->t_fnbf);
 681         }
 682         fr->timesteps++;
 683     }
 684 #endif
 685
 686     if (debug)
 687     {
 688         pr_rvecs(debug, 0, "fshift after bondeds", fr->fshift, SHIFTS);
 689     }
 690
 691 }
 692
 693 void init_enerdata(int ngener, int n_lambda, gmx_enerdata_t *enerd)
 694 {
 695     int i, n2;
 696
 697     for (i = 0; i < F_NRE; i++)
 698     {
 699         enerd->term[i]         = 0;
 700         enerd->foreign_term[i] = 0;
 701     }
 702
 703
 704     for (i = 0; i < efptNR; i++)
 705     {
 706         enerd->dvdl_lin[i]     = 0;
 707         enerd->dvdl_nonlin[i]  = 0;
 708     }
 709
 710     n2 = ngener*ngener;
 711     if (debug)
 712     {
 713         fprintf(debug, "Creating %d sized group matrix for energies\n", n2);
 714     }
 715     enerd->grpp.nener         = n2;
 716     enerd->foreign_grpp.nener = n2;
 717     for (i = 0; (i < egNR); i++)
 718     {
 719         snew(enerd->grpp.ener[i], n2);
 720         snew(enerd->foreign_grpp.ener[i], n2);
 721     }
 722
 723     if (n_lambda)
 724     {
 725         enerd->n_lambda = 1 + n_lambda;
 726         snew(enerd->enerpart_lambda, enerd->n_lambda);
 727     }
 728     else
 729     {
 730         enerd->n_lambda = 0;
 731     }
 732 }
 733
 734 void destroy_enerdata(gmx_enerdata_t *enerd)
 735 {
 736     int i;
 737
 738     for (i = 0; (i < egNR); i++)
 739     {
 740         sfree(enerd->grpp.ener[i]);
 741     }
 742
 743     for (i = 0; (i < egNR); i++)
 744     {
 745         sfree(enerd->foreign_grpp.ener[i]);
 746     }
 747
 748     if (enerd->n_lambda)
 749     {
 750         sfree(enerd->enerpart_lambda);
 751     }
 752 }
 753
 754 static real sum_v(int n, real v[])
 755 {
 756     real t;
 757     int  i;
 758
 759     t = 0.0;
 760     for (i = 0; (i < n); i++)
 761     {
 762         t = t + v[i];
 763     }
 764
 765     return t;
 766 }
 767
 768 void sum_epot(gmx_grppairener_t *grpp, real *epot)
 769 {
 770     int i;
 771
 772     /* Accumulate energies */
 773     epot[F_COUL_SR]  = sum_v(grpp->nener, grpp->ener[egCOULSR]);
 774     epot[F_LJ]       = sum_v(grpp->nener, grpp->ener[egLJSR]);
 775     epot[F_LJ14]     = sum_v(grpp->nener, grpp->ener[egLJ14]);
 776     epot[F_COUL14]   = sum_v(grpp->nener, grpp->ener[egCOUL14]);
 777     epot[F_COUL_LR]  = sum_v(grpp->nener, grpp->ener[egCOULLR]);
 778     epot[F_LJ_LR]    = sum_v(grpp->nener, grpp->ener[egLJLR]);
 779     /* We have already added 1-2,1-3, and 1-4 terms to F_GBPOL */
 780     epot[F_GBPOL]   += sum_v(grpp->nener, grpp->ener[egGB]);
 781
 782 /* lattice part of LR doesnt belong to any group
 783  * and has been added earlier
 784  */
 785     epot[F_BHAM]     = sum_v(grpp->nener, grpp->ener[egBHAMSR]);
 786     epot[F_BHAM_LR]  = sum_v(grpp->nener, grpp->ener[egBHAMLR]);
 787
 788     epot[F_EPOT] = 0;
 789     for (i = 0; (i < F_EPOT); i++)
 790     {
 791         if (i != F_DISRESVIOL && i != F_ORIRESDEV)
 792         {
 793             epot[F_EPOT] += epot[i];
 794         }
 795     }
 796 }
 797
 798 void sum_dhdl(gmx_enerdata_t *enerd, real *lambda, t_lambda *fepvals)
 799 {
 800     int    i, j, index;
 801     double dlam;
 802
 803     enerd->dvdl_lin[efptVDW] += enerd->term[F_DVDL_VDW];  /* include dispersion correction */
 804     enerd->term[F_DVDL]       = 0.0;
 805     for (i = 0; i < efptNR; i++)
 806     {
 807         if (fepvals->separate_dvdl[i])
 808         {
 809             /* could this be done more readably/compactly? */
 810             switch (i)
 811             {
 812                 case (efptMASS):
 813                     index = F_DKDL;
 814                     break;
 815                 case (efptCOUL):
 816                     index = F_DVDL_COUL;
 817                     break;
 818                 case (efptVDW):
 819                     index = F_DVDL_VDW;
 820                     break;
 821                 case (efptBONDED):
 822                     index = F_DVDL_BONDED;
 823                     break;
 824                 case (efptRESTRAINT):
 825                     index = F_DVDL_RESTRAINT;
 826                     break;
 827                 default:
 828                     index = F_DVDL;
 829                     break;
 830             }
 831             enerd->term[index] = enerd->dvdl_lin[i] + enerd->dvdl_nonlin[i];
 832             if (debug)
 833             {
 834                 fprintf(debug, "dvdl-%s[%2d]: %f: non-linear %f + linear %f\n",
 835                         efpt_names[i], i, enerd->term[index], enerd->dvdl_nonlin[i], enerd->dvdl_lin[i]);
 836             }
 837         }
 838         else
 839         {
 840             enerd->term[F_DVDL] += enerd->dvdl_lin[i] + enerd->dvdl_nonlin[i];
 841             if (debug)
 842             {
 843                 fprintf(debug, "dvd-%sl[%2d]: %f: non-linear %f + linear %f\n",
 844                         efpt_names[0], i, enerd->term[F_DVDL], enerd->dvdl_nonlin[i], enerd->dvdl_lin[i]);
 845             }
 846         }
 847     }
 848
 849     /* Notes on the foreign lambda free energy difference evaluation:
 850      * Adding the potential and ekin terms that depend linearly on lambda
 851      * as delta lam * dvdl to the energy differences is exact.
 852      * For the constraints this is not exact, but we have no other option
 853      * without literally changing the lengths and reevaluating the energies at each step.
 854      * (try to remedy this post 4.6 - MRS)
 855      * For the non-bonded LR term we assume that the soft-core (if present)
 856      * no longer affects the energy beyond the short-range cut-off,
 857      * which is a very good approximation (except for exotic settings).
 858      * (investigate how to overcome this post 4.6 - MRS)
 859      */
 860     if (fepvals->separate_dvdl[efptBONDED])
 861     {
 862         enerd->term[F_DVDL_BONDED] += enerd->term[F_DVDL_CONSTR];
 863     }
 864     else
 865     {
 866         enerd->term[F_DVDL] += enerd->term[F_DVDL_CONSTR];
 867     }
 868     enerd->term[F_DVDL_CONSTR] = 0;
 869
 870     for (i = 0; i < fepvals->n_lambda; i++)
 871     {
 872         /* note we are iterating over fepvals here!
 873            For the current lam, dlam = 0 automatically,
 874            so we don't need to add anything to the
 875            enerd->enerpart_lambda[0] */
 876
 877         /* we don't need to worry about dvdl_lin contributions to dE at
 878            current lambda, because the contributions to the current
 879            lambda are automatically zeroed */
 880
 881         for (j = 0; j < efptNR; j++)
 882         {
 883             /* Note that this loop is over all dhdl components, not just the separated ones */
 884             dlam = (fepvals->all_lambda[j][i]-lambda[j]);
 885             enerd->enerpart_lambda[i+1] += dlam*enerd->dvdl_lin[j];
 886             if (debug)
 887             {
 888                 fprintf(debug, "enerdiff lam %g: (%15s), non-linear %f linear %f*%f\n",
 889                         fepvals->all_lambda[j][i], efpt_names[j],
 890                         (enerd->enerpart_lambda[i+1] - enerd->enerpart_lambda[0]),
 891                         dlam, enerd->dvdl_lin[j]);
 892             }
 893         }
 894     }
 895 }
 896
 897
 898 void reset_foreign_enerdata(gmx_enerdata_t *enerd)
 899 {
 900     int  i, j;
 901
 902     /* First reset all foreign energy components.  Foreign energies always called on
 903        neighbor search steps */
 904     for (i = 0; (i < egNR); i++)
 905     {
 906         for (j = 0; (j < enerd->grpp.nener); j++)
 907         {
 908             enerd->foreign_grpp.ener[i][j] = 0.0;
 909         }
 910     }
 911
 912     /* potential energy components */
 913     for (i = 0; (i <= F_EPOT); i++)
 914     {
 915         enerd->foreign_term[i] = 0.0;
 916     }
 917 }
 918
 919 void reset_enerdata(t_forcerec *fr, gmx_bool bNS,
 920                     gmx_enerdata_t *enerd,
 921                     gmx_bool bMaster)
 922 {
 923     gmx_bool bKeepLR;
 924     int      i, j;
 925
 926     /* First reset all energy components, except for the long range terms
 927      * on the master at non neighbor search steps, since the long range
 928      * terms have already been summed at the last neighbor search step.
 929      */
 930     bKeepLR = (fr->bTwinRange && !bNS);
 931     for (i = 0; (i < egNR); i++)
 932     {
 933         if (!(bKeepLR && bMaster && (i == egCOULLR || i == egLJLR)))
 934         {
 935             for (j = 0; (j < enerd->grpp.nener); j++)
 936             {
 937                 enerd->grpp.ener[i][j] = 0.0;
 938             }
 939         }
 940     }
 941     for (i = 0; i < efptNR; i++)
 942     {
 943         enerd->dvdl_lin[i]    = 0.0;
 944         enerd->dvdl_nonlin[i] = 0.0;
 945     }
 946
 947     /* Normal potential energy components */
 948     for (i = 0; (i <= F_EPOT); i++)
 949     {
 950         enerd->term[i] = 0.0;
 951     }
 952     /* Initialize the dVdlambda term with the long range contribution */
 953     /* Initialize the dvdl term with the long range contribution */
 954     enerd->term[F_DVDL]            = 0.0;
 955     enerd->term[F_DVDL_COUL]       = 0.0;
 956     enerd->term[F_DVDL_VDW]        = 0.0;
 957     enerd->term[F_DVDL_BONDED]     = 0.0;
 958     enerd->term[F_DVDL_RESTRAINT]  = 0.0;
 959     enerd->term[F_DKDL]            = 0.0;
 960     if (enerd->n_lambda > 0)
 961     {
 962         for (i = 0; i < enerd->n_lambda; i++)
 963         {
 964             enerd->enerpart_lambda[i] = 0.0;
 965         }
 966     }
 967     /* reset foreign energy data - separate function since we also call it elsewhere */
 968     reset_foreign_enerdata(enerd);
 969 }