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 "typedefs.h"
  45 #include "macros.h"
  46 #include "gromacs/utility/smalloc.h"
  47 #include "macros.h"
  48 #include "physics.h"
  49 #include "force.h"
  50 #include "nonbonded.h"
  51 #include "names.h"
  52 #include "network.h"
  53 #include "pbc.h"
  54 #include "ns.h"
  55 #include "nrnb.h"
  56 #include "bondf.h"
  57 #include "mshift.h"
  58 #include "txtdump.h"
  59 #include "coulomb.h"
  60 #include "pme.h"
  61 #include "mdrun.h"
  62 #include "domdec.h"
  63 #include "qmmm.h"
  64 #include "gmx_omp_nthreads.h"
  65
  66 #include "gromacs/timing/wallcycle.h"
  67 #include "gromacs/utility/fatalerror.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
 539                     ewald_LRcorrection(fr->excl_load[t], fr->excl_load[t+1],
 540                                        cr, t, fr,
 541                                        md->chargeA,
 542                                        md->nChargePerturbed ? md->chargeB : NULL,
 543                                        md->sqrt_c6A,
 544                                        md->nTypePerturbed ? md->sqrt_c6B : NULL,
 545                                        md->sigmaA,
 546                                        md->nTypePerturbed ? md->sigmaB : NULL,
 547                                        md->sigma3A,
 548                                        md->nTypePerturbed ? md->sigma3B : NULL,
 549                                        ir->cutoff_scheme != ecutsVERLET,
 550                                        excl, x, bSB ? boxs : box, mu_tot,
 551                                        ir->ewald_geometry,
 552                                        ir->epsilon_surface,
 553                                        fnv, *vir_q, *vir_lj,
 554                                        Vcorrt_q, Vcorrt_lj,
 555                                        lambda[efptCOUL], lambda[efptVDW],
 556                                        dvdlt_q, dvdlt_lj);
 557                 }
 558                 if (nthreads > 1)
 559                 {
 560                     reduce_thread_forces(fr->natoms_force, fr->f_novirsum,
 561                                          fr->vir_el_recip, fr->vir_lj_recip,
 562                                          &Vcorr_q, &Vcorr_lj,
 563                                          &dvdl_long_range_correction_q,
 564                                          &dvdl_long_range_correction_lj,
 565                                          nthreads, fr->f_t);
 566                 }
 567                 wallcycle_sub_stop(wcycle, ewcsEWALD_CORRECTION);
 568             }
 569
 570             if (EEL_PME_EWALD(fr->eeltype) && fr->n_tpi == 0)
 571             {
 572                 Vcorr_q += ewald_charge_correction(cr, fr, lambda[efptCOUL], box,
 573                                                    &dvdl_long_range_correction_q,
 574                                                    fr->vir_el_recip);
 575             }
 576
 577             PRINT_SEPDVDL("Ewald excl./charge/dip. corr.", Vcorr_q, dvdl_long_range_correction_q);
 578             PRINT_SEPDVDL("Ewald excl. corr. LJ", Vcorr_lj, dvdl_long_range_correction_lj);
 579             enerd->dvdl_lin[efptCOUL] += dvdl_long_range_correction_q;
 580             enerd->dvdl_lin[efptVDW]  += dvdl_long_range_correction_lj;
 581         }
 582
 583         if ((EEL_PME(fr->eeltype) || EVDW_PME(fr->vdwtype)))
 584         {
 585             if (cr->duty & DUTY_PME)
 586             {
 587                 /* Do reciprocal PME for Coulomb and/or LJ. */
 588                 assert(fr->n_tpi >= 0);
 589                 if (fr->n_tpi == 0 || (flags & GMX_FORCE_STATECHANGED))
 590                 {
 591                     pme_flags = GMX_PME_SPREAD | GMX_PME_SOLVE;
 592                     if (EEL_PME(fr->eeltype))
 593                     {
 594                         pme_flags     |= GMX_PME_DO_COULOMB;
 595                     }
 596                     if (EVDW_PME(fr->vdwtype))
 597                     {
 598                         pme_flags |= GMX_PME_DO_LJ;
 599                     }
 600                     if (flags & GMX_FORCE_FORCES)
 601                     {
 602                         pme_flags |= GMX_PME_CALC_F;
 603                     }
 604                     if (flags & GMX_FORCE_VIRIAL)
 605                     {
 606                         pme_flags |= GMX_PME_CALC_ENER_VIR;
 607                     }
 608                     if (fr->n_tpi > 0)
 609                     {
 610                         /* We don't calculate f, but we do want the potential */
 611                         pme_flags |= GMX_PME_CALC_POT;
 612                     }
 613                     wallcycle_start(wcycle, ewcPMEMESH);
 614                     status = gmx_pme_do(fr->pmedata,
 615                                         0, md->homenr - fr->n_tpi,
 616                                         x, fr->f_novirsum,
 617                                         md->chargeA, md->chargeB,
 618                                         md->sqrt_c6A, md->sqrt_c6B,
 619                                         md->sigmaA, md->sigmaB,
 620                                         bSB ? boxs : box, cr,
 621                                         DOMAINDECOMP(cr) ? dd_pme_maxshift_x(cr->dd) : 0,
 622                                         DOMAINDECOMP(cr) ? dd_pme_maxshift_y(cr->dd) : 0,
 623                                         nrnb, wcycle,
 624                                         fr->vir_el_recip, fr->ewaldcoeff_q,
 625                                         fr->vir_lj_recip, fr->ewaldcoeff_lj,
 626                                         &Vlr_q, &Vlr_lj,
 627                                         lambda[efptCOUL], lambda[efptVDW],
 628                                         &dvdl_long_range_q, &dvdl_long_range_lj, pme_flags);
 629                     *cycles_pme = wallcycle_stop(wcycle, ewcPMEMESH);
 630                     if (status != 0)
 631                     {
 632                         gmx_fatal(FARGS, "Error %d in reciprocal PME routine", status);
 633                     }
 634                     /* We should try to do as little computation after
 635                      * this as possible, because parallel PME synchronizes
 636                      * the nodes, so we want all load imbalance of the
 637                      * rest of the force calculation to be before the PME
 638                      * call.  DD load balancing is done on the whole time
 639                      * of the force call (without PME).
 640                      */
 641                 }
 642                 if (fr->n_tpi > 0)
 643                 {
 644                     if (EVDW_PME(ir->vdwtype))
 645                     {
 646
 647                         gmx_fatal(FARGS, "Test particle insertion not implemented with LJ-PME");
 648                     }
 649                     /* Determine the PME grid energy of the test molecule
 650                      * with the PME grid potential of the other charges.
 651                      */
 652                     gmx_pme_calc_energy(fr->pmedata, fr->n_tpi,
 653                                         x + md->homenr - fr->n_tpi,
 654                                         md->chargeA + md->homenr - fr->n_tpi,
 655                                         &Vlr_q);
 656                 }
 657                 PRINT_SEPDVDL("PME mesh", Vlr_q + Vlr_lj, dvdl_long_range_q+dvdl_long_range_lj);
 658             }
 659         }
 660
 661         if (!EEL_PME(fr->eeltype) && EEL_PME_EWALD(fr->eeltype))
 662         {
 663             Vlr_q = do_ewald(ir, x, fr->f_novirsum,
 664                              md->chargeA, md->chargeB,
 665                              box_size, cr, md->homenr,
 666                              fr->vir_el_recip, fr->ewaldcoeff_q,
 667                              lambda[efptCOUL], &dvdl_long_range_q, fr->ewald_table);
 668             PRINT_SEPDVDL("Ewald long-range", Vlr_q, dvdl_long_range_q);
 669         }
 670
 671         /* Note that with separate PME nodes we get the real energies later */
 672         enerd->dvdl_lin[efptCOUL] += dvdl_long_range_q;
 673         enerd->dvdl_lin[efptVDW]  += dvdl_long_range_lj;
 674         enerd->term[F_COUL_RECIP]  = Vlr_q + Vcorr_q;
 675         enerd->term[F_LJ_RECIP]    = Vlr_lj + Vcorr_lj;
 676         if (debug)
 677         {
 678             fprintf(debug, "Vlr_q = %g, Vcorr_q = %g, Vlr_corr_q = %g\n",
 679                     Vlr_q, Vcorr_q, enerd->term[F_COUL_RECIP]);
 680             pr_rvecs(debug, 0, "vir_el_recip after corr", fr->vir_el_recip, DIM);
 681             pr_rvecs(debug, 0, "fshift after LR Corrections", fr->fshift, SHIFTS);
 682             fprintf(debug, "Vlr_lj: %g, Vcorr_lj = %g, Vlr_corr_lj = %g\n",
 683                     Vlr_lj, Vcorr_lj, enerd->term[F_LJ_RECIP]);
 684             pr_rvecs(debug, 0, "vir_lj_recip after corr", fr->vir_lj_recip, DIM);
 685         }
 686     }
 687     else
 688     {
 689         /* Is there a reaction-field exclusion correction needed? */
 690         if (EEL_RF(fr->eeltype) && eelRF_NEC != fr->eeltype)
 691         {
 692             /* With the Verlet scheme, exclusion forces are calculated
 693              * in the non-bonded kernel.
 694              */
 695             if (ir->cutoff_scheme != ecutsVERLET)
 696             {
 697                 real dvdl_rf_excl      = 0;
 698                 enerd->term[F_RF_EXCL] =
 699                     RF_excl_correction(fr, graph, md, excl, x, f,
 700                                        fr->fshift, &pbc, lambda[efptCOUL], &dvdl_rf_excl);
 701
 702                 enerd->dvdl_lin[efptCOUL] += dvdl_rf_excl;
 703                 PRINT_SEPDVDL("RF exclusion correction",
 704                               enerd->term[F_RF_EXCL], dvdl_rf_excl);
 705             }
 706         }
 707     }
 708     where();
 709     debug_gmx();
 710
 711     if (debug)
 712     {
 713         print_nrnb(debug, nrnb);
 714     }
 715     debug_gmx();
 716
 717 #ifdef GMX_MPI
 718     if (TAKETIME)
 719     {
 720         t2 = MPI_Wtime();
 721         MPI_Barrier(cr->mpi_comm_mygroup);
 722         t3          = MPI_Wtime();
 723         fr->t_wait += t3-t2;
 724         if (fr->timesteps == 11)
 725         {
 726             fprintf(stderr, "* PP load balancing info: node %d, step %s, rel wait time=%3.0f%% , load string value: %7.2f\n",
 727                     cr->nodeid, gmx_step_str(fr->timesteps, buf),
 728                     100*fr->t_wait/(fr->t_wait+fr->t_fnbf),
 729                     (fr->t_fnbf+fr->t_wait)/fr->t_fnbf);
 730         }
 731         fr->timesteps++;
 732     }
 733 #endif
 734
 735     if (debug)
 736     {
 737         pr_rvecs(debug, 0, "fshift after bondeds", fr->fshift, SHIFTS);
 738     }
 739
 740 }
 741
 742 void init_enerdata(int ngener, int n_lambda, gmx_enerdata_t *enerd)
 743 {
 744     int i, n2;
 745
 746     for (i = 0; i < F_NRE; i++)
 747     {
 748         enerd->term[i]         = 0;
 749         enerd->foreign_term[i] = 0;
 750     }
 751
 752
 753     for (i = 0; i < efptNR; i++)
 754     {
 755         enerd->dvdl_lin[i]     = 0;
 756         enerd->dvdl_nonlin[i]  = 0;
 757     }
 758
 759     n2 = ngener*ngener;
 760     if (debug)
 761     {
 762         fprintf(debug, "Creating %d sized group matrix for energies\n", n2);
 763     }
 764     enerd->grpp.nener         = n2;
 765     enerd->foreign_grpp.nener = n2;
 766     for (i = 0; (i < egNR); i++)
 767     {
 768         snew(enerd->grpp.ener[i], n2);
 769         snew(enerd->foreign_grpp.ener[i], n2);
 770     }
 771
 772     if (n_lambda)
 773     {
 774         enerd->n_lambda = 1 + n_lambda;
 775         snew(enerd->enerpart_lambda, enerd->n_lambda);
 776     }
 777     else
 778     {
 779         enerd->n_lambda = 0;
 780     }
 781 }
 782
 783 void destroy_enerdata(gmx_enerdata_t *enerd)
 784 {
 785     int i;
 786
 787     for (i = 0; (i < egNR); i++)
 788     {
 789         sfree(enerd->grpp.ener[i]);
 790     }
 791
 792     for (i = 0; (i < egNR); i++)
 793     {
 794         sfree(enerd->foreign_grpp.ener[i]);
 795     }
 796
 797     if (enerd->n_lambda)
 798     {
 799         sfree(enerd->enerpart_lambda);
 800     }
 801 }
 802
 803 static real sum_v(int n, real v[])
 804 {
 805     real t;
 806     int  i;
 807
 808     t = 0.0;
 809     for (i = 0; (i < n); i++)
 810     {
 811         t = t + v[i];
 812     }
 813
 814     return t;
 815 }
 816
 817 void sum_epot(gmx_grppairener_t *grpp, real *epot)
 818 {
 819     int i;
 820
 821     /* Accumulate energies */
 822     epot[F_COUL_SR]  = sum_v(grpp->nener, grpp->ener[egCOULSR]);
 823     epot[F_LJ]       = sum_v(grpp->nener, grpp->ener[egLJSR]);
 824     epot[F_LJ14]     = sum_v(grpp->nener, grpp->ener[egLJ14]);
 825     epot[F_COUL14]   = sum_v(grpp->nener, grpp->ener[egCOUL14]);
 826     epot[F_COUL_LR]  = sum_v(grpp->nener, grpp->ener[egCOULLR]);
 827     epot[F_LJ_LR]    = sum_v(grpp->nener, grpp->ener[egLJLR]);
 828     /* We have already added 1-2,1-3, and 1-4 terms to F_GBPOL */
 829     epot[F_GBPOL]   += sum_v(grpp->nener, grpp->ener[egGB]);
 830
 831 /* lattice part of LR doesnt belong to any group
 832  * and has been added earlier
 833  */
 834     epot[F_BHAM]     = sum_v(grpp->nener, grpp->ener[egBHAMSR]);
 835     epot[F_BHAM_LR]  = sum_v(grpp->nener, grpp->ener[egBHAMLR]);
 836
 837     epot[F_EPOT] = 0;
 838     for (i = 0; (i < F_EPOT); i++)
 839     {
 840         if (i != F_DISRESVIOL && i != F_ORIRESDEV)
 841         {
 842             epot[F_EPOT] += epot[i];
 843         }
 844     }
 845 }
 846
 847 void sum_dhdl(gmx_enerdata_t *enerd, real *lambda, t_lambda *fepvals)
 848 {
 849     int    i, j, index;
 850     double dlam;
 851
 852     enerd->dvdl_lin[efptVDW] += enerd->term[F_DVDL_VDW];  /* include dispersion correction */
 853     enerd->term[F_DVDL]       = 0.0;
 854     for (i = 0; i < efptNR; i++)
 855     {
 856         if (fepvals->separate_dvdl[i])
 857         {
 858             /* could this be done more readably/compactly? */
 859             switch (i)
 860             {
 861                 case (efptMASS):
 862                     index = F_DKDL;
 863                     break;
 864                 case (efptCOUL):
 865                     index = F_DVDL_COUL;
 866                     break;
 867                 case (efptVDW):
 868                     index = F_DVDL_VDW;
 869                     break;
 870                 case (efptBONDED):
 871                     index = F_DVDL_BONDED;
 872                     break;
 873                 case (efptRESTRAINT):
 874                     index = F_DVDL_RESTRAINT;
 875                     break;
 876                 default:
 877                     index = F_DVDL;
 878                     break;
 879             }
 880             enerd->term[index] = enerd->dvdl_lin[i] + enerd->dvdl_nonlin[i];
 881             if (debug)
 882             {
 883                 fprintf(debug, "dvdl-%s[%2d]: %f: non-linear %f + linear %f\n",
 884                         efpt_names[i], i, enerd->term[index], enerd->dvdl_nonlin[i], enerd->dvdl_lin[i]);
 885             }
 886         }
 887         else
 888         {
 889             enerd->term[F_DVDL] += enerd->dvdl_lin[i] + enerd->dvdl_nonlin[i];
 890             if (debug)
 891             {
 892                 fprintf(debug, "dvd-%sl[%2d]: %f: non-linear %f + linear %f\n",
 893                         efpt_names[0], i, enerd->term[F_DVDL], enerd->dvdl_nonlin[i], enerd->dvdl_lin[i]);
 894             }
 895         }
 896     }
 897
 898     /* Notes on the foreign lambda free energy difference evaluation:
 899      * Adding the potential and ekin terms that depend linearly on lambda
 900      * as delta lam * dvdl to the energy differences is exact.
 901      * For the constraints this is not exact, but we have no other option
 902      * without literally changing the lengths and reevaluating the energies at each step.
 903      * (try to remedy this post 4.6 - MRS)
 904      * For the non-bonded LR term we assume that the soft-core (if present)
 905      * no longer affects the energy beyond the short-range cut-off,
 906      * which is a very good approximation (except for exotic settings).
 907      * (investigate how to overcome this post 4.6 - MRS)
 908      */
 909     if (fepvals->separate_dvdl[efptBONDED])
 910     {
 911         enerd->term[F_DVDL_BONDED] += enerd->term[F_DVDL_CONSTR];
 912     }
 913     else
 914     {
 915         enerd->term[F_DVDL] += enerd->term[F_DVDL_CONSTR];
 916     }
 917     enerd->term[F_DVDL_CONSTR] = 0;
 918
 919     for (i = 0; i < fepvals->n_lambda; i++)
 920     {
 921         /* note we are iterating over fepvals here!
 922            For the current lam, dlam = 0 automatically,
 923            so we don't need to add anything to the
 924            enerd->enerpart_lambda[0] */
 925
 926         /* we don't need to worry about dvdl_lin contributions to dE at
 927            current lambda, because the contributions to the current
 928            lambda are automatically zeroed */
 929
 930         for (j = 0; j < efptNR; j++)
 931         {
 932             /* Note that this loop is over all dhdl components, not just the separated ones */
 933             dlam = (fepvals->all_lambda[j][i]-lambda[j]);
 934             enerd->enerpart_lambda[i+1] += dlam*enerd->dvdl_lin[j];
 935             if (debug)
 936             {
 937                 fprintf(debug, "enerdiff lam %g: (%15s), non-linear %f linear %f*%f\n",
 938                         fepvals->all_lambda[j][i], efpt_names[j],
 939                         (enerd->enerpart_lambda[i+1] - enerd->enerpart_lambda[0]),
 940                         dlam, enerd->dvdl_lin[j]);
 941             }
 942         }
 943     }
 944 }
 945
 946
 947 void reset_foreign_enerdata(gmx_enerdata_t *enerd)
 948 {
 949     int  i, j;
 950
 951     /* First reset all foreign energy components.  Foreign energies always called on
 952        neighbor search steps */
 953     for (i = 0; (i < egNR); i++)
 954     {
 955         for (j = 0; (j < enerd->grpp.nener); j++)
 956         {
 957             enerd->foreign_grpp.ener[i][j] = 0.0;
 958         }
 959     }
 960
 961     /* potential energy components */
 962     for (i = 0; (i <= F_EPOT); i++)
 963     {
 964         enerd->foreign_term[i] = 0.0;
 965     }
 966 }
 967
 968 void reset_enerdata(t_forcerec *fr, gmx_bool bNS,
 969                     gmx_enerdata_t *enerd,
 970                     gmx_bool bMaster)
 971 {
 972     gmx_bool bKeepLR;
 973     int      i, j;
 974
 975     /* First reset all energy components, except for the long range terms
 976      * on the master at non neighbor search steps, since the long range
 977      * terms have already been summed at the last neighbor search step.
 978      */
 979     bKeepLR = (fr->bTwinRange && !bNS);
 980     for (i = 0; (i < egNR); i++)
 981     {
 982         if (!(bKeepLR && bMaster && (i == egCOULLR || i == egLJLR)))
 983         {
 984             for (j = 0; (j < enerd->grpp.nener); j++)
 985             {
 986                 enerd->grpp.ener[i][j] = 0.0;
 987             }
 988         }
 989     }
 990     for (i = 0; i < efptNR; i++)
 991     {
 992         enerd->dvdl_lin[i]    = 0.0;
 993         enerd->dvdl_nonlin[i] = 0.0;
 994     }
 995
 996     /* Normal potential energy components */
 997     for (i = 0; (i <= F_EPOT); i++)
 998     {
 999         enerd->term[i] = 0.0;
1000     }
1001     /* Initialize the dVdlambda term with the long range contribution */
1002     /* Initialize the dvdl term with the long range contribution */
1003     enerd->term[F_DVDL]            = 0.0;
1004     enerd->term[F_DVDL_COUL]       = 0.0;
1005     enerd->term[F_DVDL_VDW]        = 0.0;
1006     enerd->term[F_DVDL_BONDED]     = 0.0;
1007     enerd->term[F_DVDL_RESTRAINT]  = 0.0;
1008     enerd->term[F_DKDL]            = 0.0;
1009     if (enerd->n_lambda > 0)
1010     {
1011         for (i = 0; i < enerd->n_lambda; i++)
1012         {
1013             enerd->enerpart_lambda[i] = 0.0;
1014         }
1015     }
1016     /* reset foreign energy data - separate function since we also call it elsewhere */
1017     reset_foreign_enerdata(enerd);
1018 }