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         /* Currently all group scheme kernels always calculate (shift-)forces */
 265         if (flags & GMX_FORCE_FORCES)
 266         {
 267             donb_flags |= GMX_NONBONDED_DO_FORCE;
 268         }
 269         if (flags & GMX_FORCE_VIRIAL)
 270         {
 271             donb_flags |= GMX_NONBONDED_DO_SHIFTFORCE;
 272         }
 273         if (flags & GMX_FORCE_ENERGY)
 274         {
 275             donb_flags |= GMX_NONBONDED_DO_POTENTIAL;
 276         }
 277         if (flags & GMX_FORCE_DO_LR)
 278         {
 279             donb_flags |= GMX_NONBONDED_DO_LR;
 280         }
 281
 282         wallcycle_sub_start(wcycle, ewcsNONBONDED);
 283         do_nonbonded(fr, x, f, f_longrange, md, excl,
 284                      &enerd->grpp, nrnb,
 285                      lambda, dvdl_nb, -1, -1, donb_flags);
 286
 287         /* If we do foreign lambda and we have soft-core interactions
 288          * we have to recalculate the (non-linear) energies contributions.
 289          */
 290         if (fepvals->n_lambda > 0 && (flags & GMX_FORCE_DHDL) && fepvals->sc_alpha != 0)
 291         {
 292             for (i = 0; i < enerd->n_lambda; i++)
 293             {
 294                 for (j = 0; j < efptNR; j++)
 295                 {
 296                     lam_i[j] = (i == 0 ? lambda[j] : fepvals->all_lambda[j][i-1]);
 297                 }
 298                 reset_foreign_enerdata(enerd);
 299                 do_nonbonded(fr, x, f, f_longrange, md, excl,
 300                              &(enerd->foreign_grpp), nrnb,
 301                              lam_i, dvdl_dum, -1, -1,
 302                              (donb_flags & ~GMX_NONBONDED_DO_FORCE) | GMX_NONBONDED_DO_FOREIGNLAMBDA);
 303                 sum_epot(&(enerd->foreign_grpp), enerd->foreign_term);
 304                 enerd->enerpart_lambda[i] += enerd->foreign_term[F_EPOT];
 305             }
 306         }
 307         wallcycle_sub_stop(wcycle, ewcsNONBONDED);
 308         where();
 309     }
 310
 311     /* If we are doing GB, calculate bonded forces and apply corrections
 312      * to the solvation forces */
 313     /* MRS: Eventually, many need to include free energy contribution here! */
 314     if (ir->implicit_solvent)
 315     {
 316         wallcycle_sub_start(wcycle, ewcsBONDED);
 317         calc_gb_forces(cr, md, born, top, x, f, fr, idef,
 318                        ir->gb_algorithm, ir->sa_algorithm, nrnb, &pbc, graph, enerd);
 319         wallcycle_sub_stop(wcycle, ewcsBONDED);
 320     }
 321
 322 #ifdef GMX_MPI
 323     if (TAKETIME)
 324     {
 325         t1          = MPI_Wtime();
 326         fr->t_fnbf += t1-t0;
 327     }
 328 #endif
 329
 330     if (fepvals->sc_alpha != 0)
 331     {
 332         enerd->dvdl_nonlin[efptVDW] += dvdl_nb[efptVDW];
 333     }
 334     else
 335     {
 336         enerd->dvdl_lin[efptVDW] += dvdl_nb[efptVDW];
 337     }
 338
 339     if (fepvals->sc_alpha != 0)
 340
 341     /* even though coulomb part is linear, we already added it, beacuse we
 342        need to go through the vdw calculation anyway */
 343     {
 344         enerd->dvdl_nonlin[efptCOUL] += dvdl_nb[efptCOUL];
 345     }
 346     else
 347     {
 348         enerd->dvdl_lin[efptCOUL] += dvdl_nb[efptCOUL];
 349     }
 350
 351     if (bSepDVDL)
 352     {
 353         real V_short_range    = 0;
 354         real dvdl_short_range = 0;
 355
 356         for (i = 0; i < enerd->grpp.nener; i++)
 357         {
 358             V_short_range +=
 359                 (fr->bBHAM ?
 360                  enerd->grpp.ener[egBHAMSR][i] :
 361                  enerd->grpp.ener[egLJSR][i])
 362                 + enerd->grpp.ener[egCOULSR][i] + enerd->grpp.ener[egGB][i];
 363         }
 364         dvdl_short_range = dvdl_nb[efptVDW] + dvdl_nb[efptCOUL];
 365         PRINT_SEPDVDL("VdW and Coulomb SR particle-p.",
 366                       V_short_range,
 367                       dvdl_short_range);
 368     }
 369     debug_gmx();
 370
 371
 372     if (debug)
 373     {
 374         pr_rvecs(debug, 0, "fshift after SR", fr->fshift, SHIFTS);
 375     }
 376
 377     /* Shift the coordinates. Must be done before bonded forces and PPPM,
 378      * but is also necessary for SHAKE and update, therefore it can NOT
 379      * go when no bonded forces have to be evaluated.
 380      */
 381
 382     /* Here sometimes we would not need to shift with NBFonly,
 383      * but we do so anyhow for consistency of the returned coordinates.
 384      */
 385     if (graph)
 386     {
 387         shift_self(graph, box, x);
 388         if (TRICLINIC(box))
 389         {
 390             inc_nrnb(nrnb, eNR_SHIFTX, 2*graph->nnodes);
 391         }
 392         else
 393         {
 394             inc_nrnb(nrnb, eNR_SHIFTX, graph->nnodes);
 395         }
 396     }
 397     /* Check whether we need to do bondeds or correct for exclusions */
 398     if (fr->bMolPBC &&
 399         ((flags & GMX_FORCE_BONDED)
 400          || EEL_RF(fr->eeltype) || EEL_FULL(fr->eeltype) || EVDW_PME(fr->vdwtype)))
 401     {
 402         /* Since all atoms are in the rectangular or triclinic unit-cell,
 403          * only single box vector shifts (2 in x) are required.
 404          */
 405         set_pbc_dd(&pbc, fr->ePBC, cr->dd, TRUE, box);
 406     }
 407     debug_gmx();
 408
 409     if (flags & GMX_FORCE_BONDED)
 410     {
 411         wallcycle_sub_start(wcycle, ewcsBONDED);
 412         calc_bonds(fplog, cr->ms,
 413                    idef, x, hist, f, fr, &pbc, graph, enerd, nrnb, lambda, md, fcd,
 414                    DOMAINDECOMP(cr) ? cr->dd->gatindex : NULL, atype, born,
 415                    flags,
 416                    fr->bSepDVDL && do_per_step(step, ir->nstlog), step);
 417
 418         /* Check if we have to determine energy differences
 419          * at foreign lambda's.
 420          */
 421         if (fepvals->n_lambda > 0 && (flags & GMX_FORCE_DHDL) &&
 422             idef->ilsort != ilsortNO_FE)
 423         {
 424             if (idef->ilsort != ilsortFE_SORTED)
 425             {
 426                 gmx_incons("The bonded interactions are not sorted for free energy");
 427             }
 428             for (i = 0; i < enerd->n_lambda; i++)
 429             {
 430                 reset_foreign_enerdata(enerd);
 431                 for (j = 0; j < efptNR; j++)
 432                 {
 433                     lam_i[j] = (i == 0 ? lambda[j] : fepvals->all_lambda[j][i-1]);
 434                 }
 435                 calc_bonds_lambda(fplog, idef, x, fr, &pbc, graph, &(enerd->foreign_grpp), enerd->foreign_term, nrnb, lam_i, md,
 436                                   fcd, DOMAINDECOMP(cr) ? cr->dd->gatindex : NULL);
 437                 sum_epot(&(enerd->foreign_grpp), enerd->foreign_term);
 438                 enerd->enerpart_lambda[i] += enerd->foreign_term[F_EPOT];
 439             }
 440         }
 441         debug_gmx();
 442
 443         wallcycle_sub_stop(wcycle, ewcsBONDED);
 444     }
 445
 446     where();
 447
 448     *cycles_pme = 0;
 449     if (EEL_FULL(fr->eeltype) || EVDW_PME(fr->vdwtype))
 450     {
 451         real Vlr             = 0, Vcorr = 0;
 452         real dvdl_long_range = 0;
 453         int  status          = 0;
 454
 455         bSB = (ir->nwall == 2);
 456         if (bSB)
 457         {
 458             copy_mat(box, boxs);
 459             svmul(ir->wall_ewald_zfac, boxs[ZZ], boxs[ZZ]);
 460             box_size[ZZ] *= ir->wall_ewald_zfac;
 461         }
 462     }
 463
 464     /* Do long-range electrostatics and/or LJ-PME, including related short-range
 465      * corrections.
 466      */
 467
 468     clear_mat(fr->vir_el_recip);
 469     clear_mat(fr->vir_lj_recip);
 470
 471     if (EEL_FULL(fr->eeltype) || EVDW_PME(fr->vdwtype))
 472     {
 473         real Vlr_q             = 0, Vlr_lj = 0, Vcorr_q = 0, Vcorr_lj = 0;
 474         real dvdl_long_range_q = 0, dvdl_long_range_lj = 0;
 475         int  status            = 0;
 476
 477         if (EEL_PME_EWALD(fr->eeltype) || EVDW_PME(fr->vdwtype))
 478         {
 479             real dvdl_long_range_correction_q   = 0;
 480             real dvdl_long_range_correction_lj  = 0;
 481             /* With the Verlet scheme exclusion forces are calculated
 482              * in the non-bonded kernel.
 483              */
 484             /* The TPI molecule does not have exclusions with the rest
 485              * of the system and no intra-molecular PME grid
 486              * contributions will be calculated in
 487              * gmx_pme_calc_energy.
 488              */
 489             if ((ir->cutoff_scheme == ecutsGROUP && fr->n_tpi == 0) ||
 490                 ir->ewald_geometry != eewg3D ||
 491                 ir->epsilon_surface != 0)
 492             {
 493                 int nthreads, t;
 494
 495                 wallcycle_sub_start(wcycle, ewcsEWALD_CORRECTION);
 496
 497                 if (fr->n_tpi > 0)
 498                 {
 499                     gmx_fatal(FARGS, "TPI with PME currently only works in a 3D geometry with tin-foil boundary conditions");
 500                 }
 501
 502                 nthreads = gmx_omp_nthreads_get(emntBonded);
 503 #pragma omp parallel for num_threads(nthreads) schedule(static)
 504                 for (t = 0; t < nthreads; t++)
 505                 {
 506                     int     s, e, i;
 507                     rvec   *fnv;
 508                     tensor *vir_q, *vir_lj;
 509                     real   *Vcorrt_q, *Vcorrt_lj, *dvdlt_q, *dvdlt_lj;
 510                     if (t == 0)
 511                     {
 512                         fnv       = fr->f_novirsum;
 513                         vir_q     = &fr->vir_el_recip;
 514                         vir_lj    = &fr->vir_lj_recip;
 515                         Vcorrt_q  = &Vcorr_q;
 516                         Vcorrt_lj = &Vcorr_lj;
 517                         dvdlt_q   = &dvdl_long_range_correction_q;
 518                         dvdlt_lj  = &dvdl_long_range_correction_lj;
 519                     }
 520                     else
 521                     {
 522                         fnv       = fr->f_t[t].f;
 523                         vir_q     = &fr->f_t[t].vir_q;
 524                         vir_lj    = &fr->f_t[t].vir_lj;
 525                         Vcorrt_q  = &fr->f_t[t].Vcorr_q;
 526                         Vcorrt_lj = &fr->f_t[t].Vcorr_lj;
 527                         dvdlt_q   = &fr->f_t[t].dvdl[efptCOUL];
 528                         dvdlt_lj  = &fr->f_t[t].dvdl[efptVDW];
 529                         for (i = 0; i < fr->natoms_force; i++)
 530                         {
 531                             clear_rvec(fnv[i]);
 532                         }
 533                         clear_mat(*vir_q);
 534                         clear_mat(*vir_lj);
 535                     }
 536                     *dvdlt_q  = 0;
 537                     *dvdlt_lj = 0;
 538                     ewald_LRcorrection(fr->excl_load[t], fr->excl_load[t+1],
 539                                        cr, t, fr,
 540                                        md->chargeA,
 541                                        md->nChargePerturbed ? md->chargeB : NULL,
 542                                        md->sqrt_c6A,
 543                                        md->nChargePerturbed ? md->sqrt_c6B : NULL,
 544                                        md->sigmaA,
 545                                        md->nChargePerturbed ? md->sigmaB : NULL,
 546                                        md->sigma3A,
 547                                        md->nChargePerturbed ? md->sigma3B : NULL,
 548                                        ir->cutoff_scheme != ecutsVERLET,
 549                                        excl, x, bSB ? boxs : box, mu_tot,
 550                                        ir->ewald_geometry,
 551                                        ir->epsilon_surface,
 552                                        fnv, *vir_q, *vir_lj,
 553                                        Vcorrt_q, Vcorrt_lj,
 554                                        lambda[efptCOUL], lambda[efptVDW],
 555                                        dvdlt_q, dvdlt_lj);
 556                 }
 557                 if (nthreads > 1)
 558                 {
 559                     reduce_thread_forces(fr->natoms_force, fr->f_novirsum,
 560                                          fr->vir_el_recip, fr->vir_lj_recip,
 561                                          &Vcorr_q, &Vcorr_lj,
 562                                          &dvdl_long_range_correction_q,
 563                                          &dvdl_long_range_correction_lj,
 564                                          nthreads, fr->f_t);
 565                 }
 566                 wallcycle_sub_stop(wcycle, ewcsEWALD_CORRECTION);
 567             }
 568
 569             if (EEL_PME_EWALD(fr->eeltype) && fr->n_tpi == 0)
 570             {
 571                 Vcorr_q += ewald_charge_correction(cr, fr, lambda[efptCOUL], box,
 572                                                    &dvdl_long_range_correction_q,
 573                                                    fr->vir_el_recip);
 574             }
 575
 576             PRINT_SEPDVDL("Ewald excl./charge/dip. corr.", Vcorr_q, dvdl_long_range_correction_q);
 577             PRINT_SEPDVDL("Ewald excl. corr. LJ", Vcorr_lj, dvdl_long_range_correction_lj);
 578             enerd->dvdl_lin[efptCOUL] += dvdl_long_range_correction_q;
 579             enerd->dvdl_lin[efptVDW]  += dvdl_long_range_correction_lj;
 580         }
 581
 582         if ((EEL_PME(fr->eeltype) || EVDW_PME(fr->vdwtype)))
 583         {
 584             if (cr->duty & DUTY_PME)
 585             {
 586                 /* Do reciprocal PME for Coulomb and/or LJ. */
 587                 assert(fr->n_tpi >= 0);
 588                 if (fr->n_tpi == 0 || (flags & GMX_FORCE_STATECHANGED))
 589                 {
 590                     pme_flags = GMX_PME_SPREAD_Q | GMX_PME_SOLVE;
 591                     if (EEL_PME(fr->eeltype))
 592                     {
 593                         pme_flags     |= GMX_PME_DO_COULOMB;
 594                     }
 595                     if (EVDW_PME(fr->vdwtype))
 596                     {
 597                         pme_flags |= GMX_PME_DO_LJ;
 598                         if (fr->ljpme_combination_rule == eljpmeLB)
 599                         {
 600                             /*Lorentz-Berthelot Comb. Rules in LJ-PME*/
 601                             pme_flags |= GMX_PME_LJ_LB;
 602                         }
 603                     }
 604                     if (flags & GMX_FORCE_FORCES)
 605                     {
 606                         pme_flags |= GMX_PME_CALC_F;
 607                     }
 608                     if (flags & GMX_FORCE_VIRIAL)
 609                     {
 610                         pme_flags |= GMX_PME_CALC_ENER_VIR;
 611                     }
 612                     if (fr->n_tpi > 0)
 613                     {
 614                         /* We don't calculate f, but we do want the potential */
 615                         pme_flags |= GMX_PME_CALC_POT;
 616                     }
 617                     wallcycle_start(wcycle, ewcPMEMESH);
 618                     status = gmx_pme_do(fr->pmedata,
 619                                         0, md->homenr - fr->n_tpi,
 620                                         x, fr->f_novirsum,
 621                                         md->chargeA, md->chargeB,
 622                                         md->sqrt_c6A, md->sqrt_c6B,
 623                                         md->sigmaA, md->sigmaB,
 624                                         bSB ? boxs : box, cr,
 625                                         DOMAINDECOMP(cr) ? dd_pme_maxshift_x(cr->dd) : 0,
 626                                         DOMAINDECOMP(cr) ? dd_pme_maxshift_y(cr->dd) : 0,
 627                                         nrnb, wcycle,
 628                                         fr->vir_el_recip, fr->ewaldcoeff_q,
 629                                         fr->vir_lj_recip, fr->ewaldcoeff_lj,
 630                                         &Vlr_q, &Vlr_lj,
 631                                         lambda[efptCOUL], lambda[efptVDW],
 632                                         &dvdl_long_range_q, &dvdl_long_range_lj, pme_flags);
 633                     *cycles_pme = wallcycle_stop(wcycle, ewcPMEMESH);
 634                     if (status != 0)
 635                     {
 636                         gmx_fatal(FARGS, "Error %d in reciprocal PME routine", status);
 637                     }
 638                     /* We should try to do as little computation after
 639                      * this as possible, because parallel PME synchronizes
 640                      * the nodes, so we want all load imbalance of the
 641                      * rest of the force calculation to be before the PME
 642                      * call.  DD load balancing is done on the whole time
 643                      * of the force call (without PME).
 644                      */
 645                 }
 646                 if (fr->n_tpi > 0)
 647                 {
 648                     if (EVDW_PME(ir->vdwtype))
 649                     {
 650
 651                         gmx_fatal(FARGS, "Test particle insertion not implemented with LJ-PME");
 652                     }
 653                     /* Determine the PME grid energy of the test molecule
 654                      * with the PME grid potential of the other charges.
 655                      */
 656                     gmx_pme_calc_energy(fr->pmedata, fr->n_tpi,
 657                                         x + md->homenr - fr->n_tpi,
 658                                         md->chargeA + md->homenr - fr->n_tpi,
 659                                         &Vlr_q);
 660                 }
 661                 PRINT_SEPDVDL("PME mesh", Vlr_q + Vlr_lj, dvdl_long_range_q+dvdl_long_range_lj);
 662             }
 663         }
 664
 665         if (!EEL_PME(fr->eeltype) && EEL_PME_EWALD(fr->eeltype))
 666         {
 667             Vlr_q = do_ewald(ir, x, fr->f_novirsum,
 668                              md->chargeA, md->chargeB,
 669                              box_size, cr, md->homenr,
 670                              fr->vir_el_recip, fr->ewaldcoeff_q,
 671                              lambda[efptCOUL], &dvdl_long_range_q, fr->ewald_table);
 672             PRINT_SEPDVDL("Ewald long-range", Vlr_q, dvdl_long_range_q);
 673         }
 674
 675         /* Note that with separate PME nodes we get the real energies later */
 676         enerd->dvdl_lin[efptCOUL] += dvdl_long_range_q;
 677         enerd->dvdl_lin[efptVDW]  += dvdl_long_range_lj;
 678         enerd->term[F_COUL_RECIP]  = Vlr_q + Vcorr_q;
 679         enerd->term[F_LJ_RECIP]    = Vlr_lj + Vcorr_lj;
 680         if (debug)
 681         {
 682             fprintf(debug, "Vlr_q = %g, Vcorr_q = %g, Vlr_corr_q = %g\n",
 683                     Vlr_q, Vcorr_q, enerd->term[F_COUL_RECIP]);
 684             pr_rvecs(debug, 0, "vir_el_recip after corr", fr->vir_el_recip, DIM);
 685             pr_rvecs(debug, 0, "fshift after LR Corrections", fr->fshift, SHIFTS);
 686             fprintf(debug, "Vlr_lj: %g, Vcorr_lj = %g, Vlr_corr_lj = %g\n",
 687                     Vlr_lj, Vcorr_lj, enerd->term[F_LJ_RECIP]);
 688             pr_rvecs(debug, 0, "vir_lj_recip after corr", fr->vir_lj_recip, DIM);
 689         }
 690     }
 691     else
 692     {
 693         /* Is there a reaction-field exclusion correction needed? */
 694         if (EEL_RF(fr->eeltype) && eelRF_NEC != fr->eeltype)
 695         {
 696             /* With the Verlet scheme, exclusion forces are calculated
 697              * in the non-bonded kernel.
 698              */
 699             if (ir->cutoff_scheme != ecutsVERLET)
 700             {
 701                 real dvdl_rf_excl      = 0;
 702                 enerd->term[F_RF_EXCL] =
 703                     RF_excl_correction(fr, graph, md, excl, x, f,
 704                                        fr->fshift, &pbc, lambda[efptCOUL], &dvdl_rf_excl);
 705
 706                 enerd->dvdl_lin[efptCOUL] += dvdl_rf_excl;
 707                 PRINT_SEPDVDL("RF exclusion correction",
 708                               enerd->term[F_RF_EXCL], dvdl_rf_excl);
 709             }
 710         }
 711     }
 712     where();
 713     debug_gmx();
 714
 715     if (debug)
 716     {
 717         print_nrnb(debug, nrnb);
 718     }
 719     debug_gmx();
 720
 721 #ifdef GMX_MPI
 722     if (TAKETIME)
 723     {
 724         t2 = MPI_Wtime();
 725         MPI_Barrier(cr->mpi_comm_mygroup);
 726         t3          = MPI_Wtime();
 727         fr->t_wait += t3-t2;
 728         if (fr->timesteps == 11)
 729         {
 730             fprintf(stderr, "* PP load balancing info: node %d, step %s, rel wait time=%3.0f%% , load string value: %7.2f\n",
 731                     cr->nodeid, gmx_step_str(fr->timesteps, buf),
 732                     100*fr->t_wait/(fr->t_wait+fr->t_fnbf),
 733                     (fr->t_fnbf+fr->t_wait)/fr->t_fnbf);
 734         }
 735         fr->timesteps++;
 736     }
 737 #endif
 738
 739     if (debug)
 740     {
 741         pr_rvecs(debug, 0, "fshift after bondeds", fr->fshift, SHIFTS);
 742     }
 743
 744 }
 745
 746 void init_enerdata(int ngener, int n_lambda, gmx_enerdata_t *enerd)
 747 {
 748     int i, n2;
 749
 750     for (i = 0; i < F_NRE; i++)
 751     {
 752         enerd->term[i]         = 0;
 753         enerd->foreign_term[i] = 0;
 754     }
 755
 756
 757     for (i = 0; i < efptNR; i++)
 758     {
 759         enerd->dvdl_lin[i]     = 0;
 760         enerd->dvdl_nonlin[i]  = 0;
 761     }
 762
 763     n2 = ngener*ngener;
 764     if (debug)
 765     {
 766         fprintf(debug, "Creating %d sized group matrix for energies\n", n2);
 767     }
 768     enerd->grpp.nener         = n2;
 769     enerd->foreign_grpp.nener = n2;
 770     for (i = 0; (i < egNR); i++)
 771     {
 772         snew(enerd->grpp.ener[i], n2);
 773         snew(enerd->foreign_grpp.ener[i], n2);
 774     }
 775
 776     if (n_lambda)
 777     {
 778         enerd->n_lambda = 1 + n_lambda;
 779         snew(enerd->enerpart_lambda, enerd->n_lambda);
 780     }
 781     else
 782     {
 783         enerd->n_lambda = 0;
 784     }
 785 }
 786
 787 void destroy_enerdata(gmx_enerdata_t *enerd)
 788 {
 789     int i;
 790
 791     for (i = 0; (i < egNR); i++)
 792     {
 793         sfree(enerd->grpp.ener[i]);
 794     }
 795
 796     for (i = 0; (i < egNR); i++)
 797     {
 798         sfree(enerd->foreign_grpp.ener[i]);
 799     }
 800
 801     if (enerd->n_lambda)
 802     {
 803         sfree(enerd->enerpart_lambda);
 804     }
 805 }
 806
 807 static real sum_v(int n, real v[])
 808 {
 809     real t;
 810     int  i;
 811
 812     t = 0.0;
 813     for (i = 0; (i < n); i++)
 814     {
 815         t = t + v[i];
 816     }
 817
 818     return t;
 819 }
 820
 821 void sum_epot(gmx_grppairener_t *grpp, real *epot)
 822 {
 823     int i;
 824
 825     /* Accumulate energies */
 826     epot[F_COUL_SR]  = sum_v(grpp->nener, grpp->ener[egCOULSR]);
 827     epot[F_LJ]       = sum_v(grpp->nener, grpp->ener[egLJSR]);
 828     epot[F_LJ14]     = sum_v(grpp->nener, grpp->ener[egLJ14]);
 829     epot[F_COUL14]   = sum_v(grpp->nener, grpp->ener[egCOUL14]);
 830     epot[F_COUL_LR]  = sum_v(grpp->nener, grpp->ener[egCOULLR]);
 831     epot[F_LJ_LR]    = sum_v(grpp->nener, grpp->ener[egLJLR]);
 832     /* We have already added 1-2,1-3, and 1-4 terms to F_GBPOL */
 833     epot[F_GBPOL]   += sum_v(grpp->nener, grpp->ener[egGB]);
 834
 835 /* lattice part of LR doesnt belong to any group
 836  * and has been added earlier
 837  */
 838     epot[F_BHAM]     = sum_v(grpp->nener, grpp->ener[egBHAMSR]);
 839     epot[F_BHAM_LR]  = sum_v(grpp->nener, grpp->ener[egBHAMLR]);
 840
 841     epot[F_EPOT] = 0;
 842     for (i = 0; (i < F_EPOT); i++)
 843     {
 844         if (i != F_DISRESVIOL && i != F_ORIRESDEV)
 845         {
 846             epot[F_EPOT] += epot[i];
 847         }
 848     }
 849 }
 850
 851 void sum_dhdl(gmx_enerdata_t *enerd, real *lambda, t_lambda *fepvals)
 852 {
 853     int    i, j, index;
 854     double dlam;
 855
 856     enerd->dvdl_lin[efptVDW] += enerd->term[F_DVDL_VDW];  /* include dispersion correction */
 857     enerd->term[F_DVDL]       = 0.0;
 858     for (i = 0; i < efptNR; i++)
 859     {
 860         if (fepvals->separate_dvdl[i])
 861         {
 862             /* could this be done more readably/compactly? */
 863             switch (i)
 864             {
 865                 case (efptMASS):
 866                     index = F_DKDL;
 867                     break;
 868                 case (efptCOUL):
 869                     index = F_DVDL_COUL;
 870                     break;
 871                 case (efptVDW):
 872                     index = F_DVDL_VDW;
 873                     break;
 874                 case (efptBONDED):
 875                     index = F_DVDL_BONDED;
 876                     break;
 877                 case (efptRESTRAINT):
 878                     index = F_DVDL_RESTRAINT;
 879                     break;
 880                 default:
 881                     index = F_DVDL;
 882                     break;
 883             }
 884             enerd->term[index] = enerd->dvdl_lin[i] + enerd->dvdl_nonlin[i];
 885             if (debug)
 886             {
 887                 fprintf(debug, "dvdl-%s[%2d]: %f: non-linear %f + linear %f\n",
 888                         efpt_names[i], i, enerd->term[index], enerd->dvdl_nonlin[i], enerd->dvdl_lin[i]);
 889             }
 890         }
 891         else
 892         {
 893             enerd->term[F_DVDL] += enerd->dvdl_lin[i] + enerd->dvdl_nonlin[i];
 894             if (debug)
 895             {
 896                 fprintf(debug, "dvd-%sl[%2d]: %f: non-linear %f + linear %f\n",
 897                         efpt_names[0], i, enerd->term[F_DVDL], enerd->dvdl_nonlin[i], enerd->dvdl_lin[i]);
 898             }
 899         }
 900     }
 901
 902     /* Notes on the foreign lambda free energy difference evaluation:
 903      * Adding the potential and ekin terms that depend linearly on lambda
 904      * as delta lam * dvdl to the energy differences is exact.
 905      * For the constraints this is not exact, but we have no other option
 906      * without literally changing the lengths and reevaluating the energies at each step.
 907      * (try to remedy this post 4.6 - MRS)
 908      * For the non-bonded LR term we assume that the soft-core (if present)
 909      * no longer affects the energy beyond the short-range cut-off,
 910      * which is a very good approximation (except for exotic settings).
 911      * (investigate how to overcome this post 4.6 - MRS)
 912      */
 913     if (fepvals->separate_dvdl[efptBONDED])
 914     {
 915         enerd->term[F_DVDL_BONDED] += enerd->term[F_DVDL_CONSTR];
 916     }
 917     else
 918     {
 919         enerd->term[F_DVDL] += enerd->term[F_DVDL_CONSTR];
 920     }
 921     enerd->term[F_DVDL_CONSTR] = 0;
 922
 923     for (i = 0; i < fepvals->n_lambda; i++)
 924     {
 925         /* note we are iterating over fepvals here!
 926            For the current lam, dlam = 0 automatically,
 927            so we don't need to add anything to the
 928            enerd->enerpart_lambda[0] */
 929
 930         /* we don't need to worry about dvdl_lin contributions to dE at
 931            current lambda, because the contributions to the current
 932            lambda are automatically zeroed */
 933
 934         for (j = 0; j < efptNR; j++)
 935         {
 936             /* Note that this loop is over all dhdl components, not just the separated ones */
 937             dlam = (fepvals->all_lambda[j][i]-lambda[j]);
 938             enerd->enerpart_lambda[i+1] += dlam*enerd->dvdl_lin[j];
 939             if (debug)
 940             {
 941                 fprintf(debug, "enerdiff lam %g: (%15s), non-linear %f linear %f*%f\n",
 942                         fepvals->all_lambda[j][i], efpt_names[j],
 943                         (enerd->enerpart_lambda[i+1] - enerd->enerpart_lambda[0]),
 944                         dlam, enerd->dvdl_lin[j]);
 945             }
 946         }
 947     }
 948 }
 949
 950
 951 void reset_foreign_enerdata(gmx_enerdata_t *enerd)
 952 {
 953     int  i, j;
 954
 955     /* First reset all foreign energy components.  Foreign energies always called on
 956        neighbor search steps */
 957     for (i = 0; (i < egNR); i++)
 958     {
 959         for (j = 0; (j < enerd->grpp.nener); j++)
 960         {
 961             enerd->foreign_grpp.ener[i][j] = 0.0;
 962         }
 963     }
 964
 965     /* potential energy components */
 966     for (i = 0; (i <= F_EPOT); i++)
 967     {
 968         enerd->foreign_term[i] = 0.0;
 969     }
 970 }
 971
 972 void reset_enerdata(t_forcerec *fr, gmx_bool bNS,
 973                     gmx_enerdata_t *enerd,
 974                     gmx_bool bMaster)
 975 {
 976     gmx_bool bKeepLR;
 977     int      i, j;
 978
 979     /* First reset all energy components, except for the long range terms
 980      * on the master at non neighbor search steps, since the long range
 981      * terms have already been summed at the last neighbor search step.
 982      */
 983     bKeepLR = (fr->bTwinRange && !bNS);
 984     for (i = 0; (i < egNR); i++)
 985     {
 986         if (!(bKeepLR && bMaster && (i == egCOULLR || i == egLJLR)))
 987         {
 988             for (j = 0; (j < enerd->grpp.nener); j++)
 989             {
 990                 enerd->grpp.ener[i][j] = 0.0;
 991             }
 992         }
 993     }
 994     for (i = 0; i < efptNR; i++)
 995     {
 996         enerd->dvdl_lin[i]    = 0.0;
 997         enerd->dvdl_nonlin[i] = 0.0;
 998     }
 999
1000     /* Normal potential energy components */
1001     for (i = 0; (i <= F_EPOT); i++)
1002     {
1003         enerd->term[i] = 0.0;
1004     }
1005     /* Initialize the dVdlambda term with the long range contribution */
1006     /* Initialize the dvdl term with the long range contribution */
1007     enerd->term[F_DVDL]            = 0.0;
1008     enerd->term[F_DVDL_COUL]       = 0.0;
1009     enerd->term[F_DVDL_VDW]        = 0.0;
1010     enerd->term[F_DVDL_BONDED]     = 0.0;
1011     enerd->term[F_DVDL_RESTRAINT]  = 0.0;
1012     enerd->term[F_DKDL]            = 0.0;
1013     if (enerd->n_lambda > 0)
1014     {
1015         for (i = 0; i < enerd->n_lambda; i++)
1016         {
1017             enerd->enerpart_lambda[i] = 0.0;
1018         }
1019     }
1020     /* reset foreign energy data - separate function since we also call it elsewhere */
1021     reset_foreign_enerdata(enerd);
1022 }