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 "gromacs/utility/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 #include "gmx_fatal.h"
  69
  70 void ns(FILE              *fp,
  71         t_forcerec        *fr,
  72         matrix             box,
  73         gmx_groups_t      *groups,
  74         gmx_localtop_t    *top,
  75         t_mdatoms         *md,
  76         t_commrec         *cr,
  77         t_nrnb            *nrnb,
  78         gmx_bool           bFillGrid,
  79         gmx_bool           bDoLongRangeNS)
  80 {
  81     char   *ptr;
  82     int     nsearch;
  83
  84
  85     if (!fr->ns.nblist_initialized)
  86     {
  87         init_neighbor_list(fp, fr, md->homenr);
  88     }
  89
  90     if (fr->bTwinRange)
  91     {
  92         fr->nlr = 0;
  93     }
  94
  95     nsearch = search_neighbours(fp, fr, box, top, groups, cr, nrnb, md,
  96                                 bFillGrid, bDoLongRangeNS);
  97     if (debug)
  98     {
  99         fprintf(debug, "nsearch = %d\n", nsearch);
 100     }
 101
 102     /* Check whether we have to do dynamic load balancing */
 103     /*if ((nsb->nstDlb > 0) && (mod(step,nsb->nstDlb) == 0))
 104        count_nb(cr,nsb,&(top->blocks[ebCGS]),nns,fr->nlr,
 105        &(top->idef),opts->ngener);
 106      */
 107     if (fr->ns.dump_nl > 0)
 108     {
 109         dump_nblist(fp, cr, fr, fr->ns.dump_nl);
 110     }
 111 }
 112
 113 static void reduce_thread_forces(int n, rvec *f,
 114                                  tensor vir_q, tensor vir_lj,
 115                                  real *Vcorr_q, real *Vcorr_lj,
 116                                  real *dvdl_q, real *dvdl_lj,
 117                                  int nthreads, f_thread_t *f_t)
 118 {
 119     int t, i;
 120     int nthreads_loop gmx_unused;
 121
 122     /* This reduction can run over any number of threads */
 123     nthreads_loop = gmx_omp_nthreads_get(emntBonded);
 124 #pragma omp parallel for num_threads(nthreads_loop) private(t) schedule(static)
 125     for (i = 0; i < n; i++)
 126     {
 127         for (t = 1; t < nthreads; t++)
 128         {
 129             rvec_inc(f[i], f_t[t].f[i]);
 130         }
 131     }
 132     for (t = 1; t < nthreads; t++)
 133     {
 134         *Vcorr_q  += f_t[t].Vcorr_q;
 135         *Vcorr_lj += f_t[t].Vcorr_lj;
 136         *dvdl_q   += f_t[t].dvdl[efptCOUL];
 137         *dvdl_lj  += f_t[t].dvdl[efptVDW];
 138         m_add(vir_q, f_t[t].vir_q, vir_q);
 139         m_add(vir_lj, f_t[t].vir_lj, vir_lj);
 140     }
 141 }
 142
 143 void gmx_print_sepdvdl(FILE *fplog, const char *s, real v, real dvdlambda)
 144 {
 145     fprintf(fplog, "  %-30s V %12.5e  dVdl %12.5e\n", s, v, dvdlambda);
 146 }
 147
 148 void do_force_lowlevel(FILE       *fplog,   gmx_int64_t step,
 149                        t_forcerec *fr,      t_inputrec *ir,
 150                        t_idef     *idef,    t_commrec  *cr,
 151                        t_nrnb     *nrnb,    gmx_wallcycle_t wcycle,
 152                        t_mdatoms  *md,
 153                        rvec       x[],      history_t  *hist,
 154                        rvec       f[],
 155                        rvec       f_longrange[],
 156                        gmx_enerdata_t *enerd,
 157                        t_fcdata   *fcd,
 158                        gmx_localtop_t *top,
 159                        gmx_genborn_t *born,
 160                        t_atomtypes *atype,
 161                        gmx_bool       bBornRadii,
 162                        matrix     box,
 163                        t_lambda   *fepvals,
 164                        real       *lambda,
 165                        t_graph    *graph,
 166                        t_blocka   *excl,
 167                        rvec       mu_tot[],
 168                        int        flags,
 169                        float      *cycles_pme)
 170 {
 171     int         i, j;
 172     int         donb_flags;
 173     gmx_bool    bDoEpot, bSepDVDL, bSB;
 174     int         pme_flags;
 175     matrix      boxs;
 176     rvec        box_size;
 177     t_pbc       pbc;
 178     char        buf[22];
 179     double      clam_i, vlam_i;
 180     real        dvdl_dum[efptNR], dvdl_nb[efptNR], lam_i[efptNR];
 181     real        dvdl_q, dvdl_lj;
 182
 183 #ifdef GMX_MPI
 184     double  t0 = 0.0, t1, t2, t3; /* time measurement for coarse load balancing */
 185 #endif
 186
 187 #define PRINT_SEPDVDL(s, v, dvdlambda) if (bSepDVDL) { gmx_print_sepdvdl(fplog, s, v, dvdlambda); }
 188
 189     set_pbc(&pbc, fr->ePBC, box);
 190
 191     /* reset free energy components */
 192     for (i = 0; i < efptNR; i++)
 193     {
 194         dvdl_nb[i]  = 0;
 195         dvdl_dum[i] = 0;
 196     }
 197
 198     /* Reset box */
 199     for (i = 0; (i < DIM); i++)
 200     {
 201         box_size[i] = box[i][i];
 202     }
 203
 204     bSepDVDL = (fr->bSepDVDL && do_per_step(step, ir->nstlog));
 205     debug_gmx();
 206
 207     /* do QMMM first if requested */
 208     if (fr->bQMMM)
 209     {
 210         enerd->term[F_EQM] = calculate_QMMM(cr, x, f, fr);
 211     }
 212
 213     if (bSepDVDL)
 214     {
 215         fprintf(fplog, "Step %s: non-bonded V and dVdl for rank %d:\n",
 216                 gmx_step_str(step, buf), cr->nodeid);
 217     }
 218
 219     /* Call the short range functions all in one go. */
 220
 221 #ifdef GMX_MPI
 222     /*#define TAKETIME ((cr->npmenodes) && (fr->timesteps < 12))*/
 223 #define TAKETIME FALSE
 224     if (TAKETIME)
 225     {
 226         MPI_Barrier(cr->mpi_comm_mygroup);
 227         t0 = MPI_Wtime();
 228     }
 229 #endif
 230
 231     if (ir->nwall)
 232     {
 233         /* foreign lambda component for walls */
 234         real dvdl_walls = do_walls(ir, fr, box, md, x, f, lambda[efptVDW],
 235                                    enerd->grpp.ener[egLJSR], nrnb);
 236         PRINT_SEPDVDL("Walls", 0.0, dvdl_walls);
 237         enerd->dvdl_lin[efptVDW] += dvdl_walls;
 238     }
 239
 240     /* If doing GB, reset dvda and calculate the Born radii */
 241     if (ir->implicit_solvent)
 242     {
 243         wallcycle_sub_start(wcycle, ewcsNONBONDED);
 244
 245         for (i = 0; i < born->nr; i++)
 246         {
 247             fr->dvda[i] = 0;
 248         }
 249
 250         if (bBornRadii)
 251         {
 252             calc_gb_rad(cr, fr, ir, top, x, &(fr->gblist), born, md, nrnb);
 253         }
 254
 255         wallcycle_sub_stop(wcycle, ewcsNONBONDED);
 256     }
 257
 258     where();
 259     /* We only do non-bonded calculation with group scheme here, the verlet
 260      * calls are done from do_force_cutsVERLET(). */
 261     if (fr->cutoff_scheme == ecutsGROUP && (flags & GMX_FORCE_NONBONDED))
 262     {
 263         donb_flags = 0;
 264         /* Add short-range interactions */
 265         donb_flags |= GMX_NONBONDED_DO_SR;
 266
 267         /* Currently all group scheme kernels always calculate (shift-)forces */
 268         if (flags & GMX_FORCE_FORCES)
 269         {
 270             donb_flags |= GMX_NONBONDED_DO_FORCE;
 271         }
 272         if (flags & GMX_FORCE_VIRIAL)
 273         {
 274             donb_flags |= GMX_NONBONDED_DO_SHIFTFORCE;
 275         }
 276         if (flags & GMX_FORCE_ENERGY)
 277         {
 278             donb_flags |= GMX_NONBONDED_DO_POTENTIAL;
 279         }
 280         if (flags & GMX_FORCE_DO_LR)
 281         {
 282             donb_flags |= GMX_NONBONDED_DO_LR;
 283         }
 284
 285         wallcycle_sub_start(wcycle, ewcsNONBONDED);
 286         do_nonbonded(fr, x, f, f_longrange, md, excl,
 287                      &enerd->grpp, nrnb,
 288                      lambda, dvdl_nb, -1, -1, donb_flags);
 289
 290         /* If we do foreign lambda and we have soft-core interactions
 291          * we have to recalculate the (non-linear) energies contributions.
 292          */
 293         if (fepvals->n_lambda > 0 && (flags & GMX_FORCE_DHDL) && fepvals->sc_alpha != 0)
 294         {
 295             for (i = 0; i < enerd->n_lambda; i++)
 296             {
 297                 for (j = 0; j < efptNR; j++)
 298                 {
 299                     lam_i[j] = (i == 0 ? lambda[j] : fepvals->all_lambda[j][i-1]);
 300                 }
 301                 reset_foreign_enerdata(enerd);
 302                 do_nonbonded(fr, x, f, f_longrange, md, excl,
 303                              &(enerd->foreign_grpp), nrnb,
 304                              lam_i, dvdl_dum, -1, -1,
 305                              (donb_flags & ~GMX_NONBONDED_DO_FORCE) | GMX_NONBONDED_DO_FOREIGNLAMBDA);
 306                 sum_epot(&(enerd->foreign_grpp), enerd->foreign_term);
 307                 enerd->enerpart_lambda[i] += enerd->foreign_term[F_EPOT];
 308             }
 309         }
 310         wallcycle_sub_stop(wcycle, ewcsNONBONDED);
 311         where();
 312     }
 313
 314     /* If we are doing GB, calculate bonded forces and apply corrections
 315      * to the solvation forces */
 316     /* MRS: Eventually, many need to include free energy contribution here! */
 317     if (ir->implicit_solvent)
 318     {
 319         wallcycle_sub_start(wcycle, ewcsBONDED);
 320         calc_gb_forces(cr, md, born, top, x, f, fr, idef,
 321                        ir->gb_algorithm, ir->sa_algorithm, nrnb, &pbc, graph, enerd);
 322         wallcycle_sub_stop(wcycle, ewcsBONDED);
 323     }
 324
 325 #ifdef GMX_MPI
 326     if (TAKETIME)
 327     {
 328         t1          = MPI_Wtime();
 329         fr->t_fnbf += t1-t0;
 330     }
 331 #endif
 332
 333     if (fepvals->sc_alpha != 0)
 334     {
 335         enerd->dvdl_nonlin[efptVDW] += dvdl_nb[efptVDW];
 336     }
 337     else
 338     {
 339         enerd->dvdl_lin[efptVDW] += dvdl_nb[efptVDW];
 340     }
 341
 342     if (fepvals->sc_alpha != 0)
 343
 344     /* even though coulomb part is linear, we already added it, beacuse we
 345        need to go through the vdw calculation anyway */
 346     {
 347         enerd->dvdl_nonlin[efptCOUL] += dvdl_nb[efptCOUL];
 348     }
 349     else
 350     {
 351         enerd->dvdl_lin[efptCOUL] += dvdl_nb[efptCOUL];
 352     }
 353
 354     if (bSepDVDL)
 355     {
 356         real V_short_range    = 0;
 357         real dvdl_short_range = 0;
 358
 359         for (i = 0; i < enerd->grpp.nener; i++)
 360         {
 361             V_short_range +=
 362                 (fr->bBHAM ?
 363                  enerd->grpp.ener[egBHAMSR][i] :
 364                  enerd->grpp.ener[egLJSR][i])
 365                 + enerd->grpp.ener[egCOULSR][i] + enerd->grpp.ener[egGB][i];
 366         }
 367         dvdl_short_range = dvdl_nb[efptVDW] + dvdl_nb[efptCOUL];
 368         PRINT_SEPDVDL("VdW and Coulomb SR particle-p.",
 369                       V_short_range,
 370                       dvdl_short_range);
 371     }
 372     debug_gmx();
 373
 374
 375     if (debug)
 376     {
 377         pr_rvecs(debug, 0, "fshift after SR", fr->fshift, SHIFTS);
 378     }
 379
 380     /* Shift the coordinates. Must be done before bonded forces and PPPM,
 381      * but is also necessary for SHAKE and update, therefore it can NOT
 382      * go when no bonded forces have to be evaluated.
 383      */
 384
 385     /* Here sometimes we would not need to shift with NBFonly,
 386      * but we do so anyhow for consistency of the returned coordinates.
 387      */
 388     if (graph)
 389     {
 390         shift_self(graph, box, x);
 391         if (TRICLINIC(box))
 392         {
 393             inc_nrnb(nrnb, eNR_SHIFTX, 2*graph->nnodes);
 394         }
 395         else
 396         {
 397             inc_nrnb(nrnb, eNR_SHIFTX, graph->nnodes);
 398         }
 399     }
 400     /* Check whether we need to do bondeds or correct for exclusions */
 401     if (fr->bMolPBC &&
 402         ((flags & GMX_FORCE_BONDED)
 403          || EEL_RF(fr->eeltype) || EEL_FULL(fr->eeltype) || EVDW_PME(fr->vdwtype)))
 404     {
 405         /* Since all atoms are in the rectangular or triclinic unit-cell,
 406          * only single box vector shifts (2 in x) are required.
 407          */
 408         set_pbc_dd(&pbc, fr->ePBC, cr->dd, TRUE, box);
 409     }
 410     debug_gmx();
 411
 412     if (flags & GMX_FORCE_BONDED)
 413     {
 414         wallcycle_sub_start(wcycle, ewcsBONDED);
 415         calc_bonds(fplog, cr->ms,
 416                    idef, x, hist, f, fr, &pbc, graph, enerd, nrnb, lambda, md, fcd,
 417                    DOMAINDECOMP(cr) ? cr->dd->gatindex : NULL, atype, born,
 418                    flags,
 419                    fr->bSepDVDL && do_per_step(step, ir->nstlog), step);
 420
 421         /* Check if we have to determine energy differences
 422          * at foreign lambda's.
 423          */
 424         if (fepvals->n_lambda > 0 && (flags & GMX_FORCE_DHDL) &&
 425             idef->ilsort != ilsortNO_FE)
 426         {
 427             if (idef->ilsort != ilsortFE_SORTED)
 428             {
 429                 gmx_incons("The bonded interactions are not sorted for free energy");
 430             }
 431             for (i = 0; i < enerd->n_lambda; i++)
 432             {
 433                 reset_foreign_enerdata(enerd);
 434                 for (j = 0; j < efptNR; j++)
 435                 {
 436                     lam_i[j] = (i == 0 ? lambda[j] : fepvals->all_lambda[j][i-1]);
 437                 }
 438                 calc_bonds_lambda(fplog, idef, x, fr, &pbc, graph, &(enerd->foreign_grpp), enerd->foreign_term, nrnb, lam_i, md,
 439                                   fcd, DOMAINDECOMP(cr) ? cr->dd->gatindex : NULL);
 440                 sum_epot(&(enerd->foreign_grpp), enerd->foreign_term);
 441                 enerd->enerpart_lambda[i] += enerd->foreign_term[F_EPOT];
 442             }
 443         }
 444         debug_gmx();
 445
 446         wallcycle_sub_stop(wcycle, ewcsBONDED);
 447     }
 448
 449     where();
 450
 451     *cycles_pme = 0;
 452     if (EEL_FULL(fr->eeltype) || EVDW_PME(fr->vdwtype))
 453     {
 454         real Vlr             = 0, Vcorr = 0;
 455         real dvdl_long_range = 0;
 456         int  status          = 0;
 457
 458         bSB = (ir->nwall == 2);
 459         if (bSB)
 460         {
 461             copy_mat(box, boxs);
 462             svmul(ir->wall_ewald_zfac, boxs[ZZ], boxs[ZZ]);
 463             box_size[ZZ] *= ir->wall_ewald_zfac;
 464         }
 465     }
 466
 467     /* Do long-range electrostatics and/or LJ-PME, including related short-range
 468      * corrections.
 469      */
 470
 471     clear_mat(fr->vir_el_recip);
 472     clear_mat(fr->vir_lj_recip);
 473
 474     if (EEL_FULL(fr->eeltype) || EVDW_PME(fr->vdwtype))
 475     {
 476         real Vlr_q             = 0, Vlr_lj = 0, Vcorr_q = 0, Vcorr_lj = 0;
 477         real dvdl_long_range_q = 0, dvdl_long_range_lj = 0;
 478         int  status            = 0;
 479
 480         if (EEL_PME_EWALD(fr->eeltype) || EVDW_PME(fr->vdwtype))
 481         {
 482             real dvdl_long_range_correction_q   = 0;
 483             real dvdl_long_range_correction_lj  = 0;
 484             /* With the Verlet scheme exclusion forces are calculated
 485              * in the non-bonded kernel.
 486              */
 487             /* The TPI molecule does not have exclusions with the rest
 488              * of the system and no intra-molecular PME grid
 489              * contributions will be calculated in
 490              * gmx_pme_calc_energy.
 491              */
 492             if ((ir->cutoff_scheme == ecutsGROUP && fr->n_tpi == 0) ||
 493                 ir->ewald_geometry != eewg3D ||
 494                 ir->epsilon_surface != 0)
 495             {
 496                 int nthreads, t;
 497
 498                 wallcycle_sub_start(wcycle, ewcsEWALD_CORRECTION);
 499
 500                 if (fr->n_tpi > 0)
 501                 {
 502                     gmx_fatal(FARGS, "TPI with PME currently only works in a 3D geometry with tin-foil boundary conditions");
 503                 }
 504
 505                 nthreads = gmx_omp_nthreads_get(emntBonded);
 506 #pragma omp parallel for num_threads(nthreads) schedule(static)
 507                 for (t = 0; t < nthreads; t++)
 508                 {
 509                     int     s, e, i;
 510                     rvec   *fnv;
 511                     tensor *vir_q, *vir_lj;
 512                     real   *Vcorrt_q, *Vcorrt_lj, *dvdlt_q, *dvdlt_lj;
 513                     if (t == 0)
 514                     {
 515                         fnv       = fr->f_novirsum;
 516                         vir_q     = &fr->vir_el_recip;
 517                         vir_lj    = &fr->vir_lj_recip;
 518                         Vcorrt_q  = &Vcorr_q;
 519                         Vcorrt_lj = &Vcorr_lj;
 520                         dvdlt_q   = &dvdl_long_range_correction_q;
 521                         dvdlt_lj  = &dvdl_long_range_correction_lj;
 522                     }
 523                     else
 524                     {
 525                         fnv       = fr->f_t[t].f;
 526                         vir_q     = &fr->f_t[t].vir_q;
 527                         vir_lj    = &fr->f_t[t].vir_lj;
 528                         Vcorrt_q  = &fr->f_t[t].Vcorr_q;
 529                         Vcorrt_lj = &fr->f_t[t].Vcorr_lj;
 530                         dvdlt_q   = &fr->f_t[t].dvdl[efptCOUL];
 531                         dvdlt_lj  = &fr->f_t[t].dvdl[efptVDW];
 532                         for (i = 0; i < fr->natoms_force; i++)
 533                         {
 534                             clear_rvec(fnv[i]);
 535                         }
 536                         clear_mat(*vir_q);
 537                         clear_mat(*vir_lj);
 538                     }
 539                     *dvdlt_q  = 0;
 540                     *dvdlt_lj = 0;
 541
 542                     ewald_LRcorrection(fr->excl_load[t], fr->excl_load[t+1],
 543                                        cr, t, fr,
 544                                        md->chargeA,
 545                                        md->nChargePerturbed ? md->chargeB : NULL,
 546                                        md->sqrt_c6A,
 547                                        md->nTypePerturbed ? md->sqrt_c6B : NULL,
 548                                        md->sigmaA,
 549                                        md->nTypePerturbed ? md->sigmaB : NULL,
 550                                        md->sigma3A,
 551                                        md->nTypePerturbed ? md->sigma3B : NULL,
 552                                        ir->cutoff_scheme != ecutsVERLET,
 553                                        excl, x, bSB ? boxs : box, mu_tot,
 554                                        ir->ewald_geometry,
 555                                        ir->epsilon_surface,
 556                                        fnv, *vir_q, *vir_lj,
 557                                        Vcorrt_q, Vcorrt_lj,
 558                                        lambda[efptCOUL], lambda[efptVDW],
 559                                        dvdlt_q, dvdlt_lj);
 560                 }
 561                 if (nthreads > 1)
 562                 {
 563                     reduce_thread_forces(fr->natoms_force, fr->f_novirsum,
 564                                          fr->vir_el_recip, fr->vir_lj_recip,
 565                                          &Vcorr_q, &Vcorr_lj,
 566                                          &dvdl_long_range_correction_q,
 567                                          &dvdl_long_range_correction_lj,
 568                                          nthreads, fr->f_t);
 569                 }
 570                 wallcycle_sub_stop(wcycle, ewcsEWALD_CORRECTION);
 571             }
 572
 573             if (EEL_PME_EWALD(fr->eeltype) && fr->n_tpi == 0)
 574             {
 575                 Vcorr_q += ewald_charge_correction(cr, fr, lambda[efptCOUL], box,
 576                                                    &dvdl_long_range_correction_q,
 577                                                    fr->vir_el_recip);
 578             }
 579
 580             PRINT_SEPDVDL("Ewald excl./charge/dip. corr.", Vcorr_q, dvdl_long_range_correction_q);
 581             PRINT_SEPDVDL("Ewald excl. corr. LJ", Vcorr_lj, dvdl_long_range_correction_lj);
 582             enerd->dvdl_lin[efptCOUL] += dvdl_long_range_correction_q;
 583             enerd->dvdl_lin[efptVDW]  += dvdl_long_range_correction_lj;
 584         }
 585
 586         if ((EEL_PME(fr->eeltype) || EVDW_PME(fr->vdwtype)))
 587         {
 588             if (cr->duty & DUTY_PME)
 589             {
 590                 /* Do reciprocal PME for Coulomb and/or LJ. */
 591                 assert(fr->n_tpi >= 0);
 592                 if (fr->n_tpi == 0 || (flags & GMX_FORCE_STATECHANGED))
 593                 {
 594                     pme_flags = GMX_PME_SPREAD | GMX_PME_SOLVE;
 595                     if (EEL_PME(fr->eeltype))
 596                     {
 597                         pme_flags     |= GMX_PME_DO_COULOMB;
 598                     }
 599                     if (EVDW_PME(fr->vdwtype))
 600                     {
 601                         pme_flags |= GMX_PME_DO_LJ;
 602                     }
 603                     if (flags & GMX_FORCE_FORCES)
 604                     {
 605                         pme_flags |= GMX_PME_CALC_F;
 606                     }
 607                     if (flags & GMX_FORCE_VIRIAL)
 608                     {
 609                         pme_flags |= GMX_PME_CALC_ENER_VIR;
 610                     }
 611                     if (fr->n_tpi > 0)
 612                     {
 613                         /* We don't calculate f, but we do want the potential */
 614                         pme_flags |= GMX_PME_CALC_POT;
 615                     }
 616                     wallcycle_start(wcycle, ewcPMEMESH);
 617                     status = gmx_pme_do(fr->pmedata,
 618                                         0, md->homenr - fr->n_tpi,
 619                                         x, fr->f_novirsum,
 620                                         md->chargeA, md->chargeB,
 621                                         md->sqrt_c6A, md->sqrt_c6B,
 622                                         md->sigmaA, md->sigmaB,
 623                                         bSB ? boxs : box, cr,
 624                                         DOMAINDECOMP(cr) ? dd_pme_maxshift_x(cr->dd) : 0,
 625                                         DOMAINDECOMP(cr) ? dd_pme_maxshift_y(cr->dd) : 0,
 626                                         nrnb, wcycle,
 627                                         fr->vir_el_recip, fr->ewaldcoeff_q,
 628                                         fr->vir_lj_recip, fr->ewaldcoeff_lj,
 629                                         &Vlr_q, &Vlr_lj,
 630                                         lambda[efptCOUL], lambda[efptVDW],
 631                                         &dvdl_long_range_q, &dvdl_long_range_lj, pme_flags);
 632                     *cycles_pme = wallcycle_stop(wcycle, ewcPMEMESH);
 633                     if (status != 0)
 634                     {
 635                         gmx_fatal(FARGS, "Error %d in reciprocal PME routine", status);
 636                     }
 637                     /* We should try to do as little computation after
 638                      * this as possible, because parallel PME synchronizes
 639                      * the nodes, so we want all load imbalance of the
 640                      * rest of the force calculation to be before the PME
 641                      * call.  DD load balancing is done on the whole time
 642                      * of the force call (without PME).
 643                      */
 644                 }
 645                 if (fr->n_tpi > 0)
 646                 {
 647                     if (EVDW_PME(ir->vdwtype))
 648                     {
 649
 650                         gmx_fatal(FARGS, "Test particle insertion not implemented with LJ-PME");
 651                     }
 652                     /* Determine the PME grid energy of the test molecule
 653                      * with the PME grid potential of the other charges.
 654                      */
 655                     gmx_pme_calc_energy(fr->pmedata, fr->n_tpi,
 656                                         x + md->homenr - fr->n_tpi,
 657                                         md->chargeA + md->homenr - fr->n_tpi,
 658                                         &Vlr_q);
 659                 }
 660                 PRINT_SEPDVDL("PME mesh", Vlr_q + Vlr_lj, dvdl_long_range_q+dvdl_long_range_lj);
 661             }
 662         }
 663
 664         if (!EEL_PME(fr->eeltype) && EEL_PME_EWALD(fr->eeltype))
 665         {
 666             Vlr_q = do_ewald(ir, x, fr->f_novirsum,
 667                              md->chargeA, md->chargeB,
 668                              box_size, cr, md->homenr,
 669                              fr->vir_el_recip, fr->ewaldcoeff_q,
 670                              lambda[efptCOUL], &dvdl_long_range_q, fr->ewald_table);
 671             PRINT_SEPDVDL("Ewald long-range", Vlr_q, dvdl_long_range_q);
 672         }
 673
 674         /* Note that with separate PME nodes we get the real energies later */
 675         enerd->dvdl_lin[efptCOUL] += dvdl_long_range_q;
 676         enerd->dvdl_lin[efptVDW]  += dvdl_long_range_lj;
 677         enerd->term[F_COUL_RECIP]  = Vlr_q + Vcorr_q;
 678         enerd->term[F_LJ_RECIP]    = Vlr_lj + Vcorr_lj;
 679         if (debug)
 680         {
 681             fprintf(debug, "Vlr_q = %g, Vcorr_q = %g, Vlr_corr_q = %g\n",
 682                     Vlr_q, Vcorr_q, enerd->term[F_COUL_RECIP]);
 683             pr_rvecs(debug, 0, "vir_el_recip after corr", fr->vir_el_recip, DIM);
 684             pr_rvecs(debug, 0, "fshift after LR Corrections", fr->fshift, SHIFTS);
 685             fprintf(debug, "Vlr_lj: %g, Vcorr_lj = %g, Vlr_corr_lj = %g\n",
 686                     Vlr_lj, Vcorr_lj, enerd->term[F_LJ_RECIP]);
 687             pr_rvecs(debug, 0, "vir_lj_recip after corr", fr->vir_lj_recip, DIM);
 688         }
 689     }
 690     else
 691     {
 692         /* Is there a reaction-field exclusion correction needed? */
 693         if (EEL_RF(fr->eeltype) && eelRF_NEC != fr->eeltype)
 694         {
 695             /* With the Verlet scheme, exclusion forces are calculated
 696              * in the non-bonded kernel.
 697              */
 698             if (ir->cutoff_scheme != ecutsVERLET)
 699             {
 700                 real dvdl_rf_excl      = 0;
 701                 enerd->term[F_RF_EXCL] =
 702                     RF_excl_correction(fr, graph, md, excl, x, f,
 703                                        fr->fshift, &pbc, lambda[efptCOUL], &dvdl_rf_excl);
 704
 705                 enerd->dvdl_lin[efptCOUL] += dvdl_rf_excl;
 706                 PRINT_SEPDVDL("RF exclusion correction",
 707                               enerd->term[F_RF_EXCL], dvdl_rf_excl);
 708             }
 709         }
 710     }
 711     where();
 712     debug_gmx();
 713
 714     if (debug)
 715     {
 716         print_nrnb(debug, nrnb);
 717     }
 718     debug_gmx();
 719
 720 #ifdef GMX_MPI
 721     if (TAKETIME)
 722     {
 723         t2 = MPI_Wtime();
 724         MPI_Barrier(cr->mpi_comm_mygroup);
 725         t3          = MPI_Wtime();
 726         fr->t_wait += t3-t2;
 727         if (fr->timesteps == 11)
 728         {
 729             fprintf(stderr, "* PP load balancing info: rank %d, step %s, rel wait time=%3.0f%% , load string value: %7.2f\n",
 730                     cr->nodeid, gmx_step_str(fr->timesteps, buf),
 731                     100*fr->t_wait/(fr->t_wait+fr->t_fnbf),
 732                     (fr->t_fnbf+fr->t_wait)/fr->t_fnbf);
 733         }
 734         fr->timesteps++;
 735     }
 736 #endif
 737
 738     if (debug)
 739     {
 740         pr_rvecs(debug, 0, "fshift after bondeds", fr->fshift, SHIFTS);
 741     }
 742
 743 }
 744
 745 void init_enerdata(int ngener, int n_lambda, gmx_enerdata_t *enerd)
 746 {
 747     int i, n2;
 748
 749     for (i = 0; i < F_NRE; i++)
 750     {
 751         enerd->term[i]         = 0;
 752         enerd->foreign_term[i] = 0;
 753     }
 754
 755
 756     for (i = 0; i < efptNR; i++)
 757     {
 758         enerd->dvdl_lin[i]     = 0;
 759         enerd->dvdl_nonlin[i]  = 0;
 760     }
 761
 762     n2 = ngener*ngener;
 763     if (debug)
 764     {
 765         fprintf(debug, "Creating %d sized group matrix for energies\n", n2);
 766     }
 767     enerd->grpp.nener         = n2;
 768     enerd->foreign_grpp.nener = n2;
 769     for (i = 0; (i < egNR); i++)
 770     {
 771         snew(enerd->grpp.ener[i], n2);
 772         snew(enerd->foreign_grpp.ener[i], n2);
 773     }
 774
 775     if (n_lambda)
 776     {
 777         enerd->n_lambda = 1 + n_lambda;
 778         snew(enerd->enerpart_lambda, enerd->n_lambda);
 779     }
 780     else
 781     {
 782         enerd->n_lambda = 0;
 783     }
 784 }
 785
 786 void destroy_enerdata(gmx_enerdata_t *enerd)
 787 {
 788     int i;
 789
 790     for (i = 0; (i < egNR); i++)
 791     {
 792         sfree(enerd->grpp.ener[i]);
 793     }
 794
 795     for (i = 0; (i < egNR); i++)
 796     {
 797         sfree(enerd->foreign_grpp.ener[i]);
 798     }
 799
 800     if (enerd->n_lambda)
 801     {
 802         sfree(enerd->enerpart_lambda);
 803     }
 804 }
 805
 806 static real sum_v(int n, real v[])
 807 {
 808     real t;
 809     int  i;
 810
 811     t = 0.0;
 812     for (i = 0; (i < n); i++)
 813     {
 814         t = t + v[i];
 815     }
 816
 817     return t;
 818 }
 819
 820 void sum_epot(gmx_grppairener_t *grpp, real *epot)
 821 {
 822     int i;
 823
 824     /* Accumulate energies */
 825     epot[F_COUL_SR]  = sum_v(grpp->nener, grpp->ener[egCOULSR]);
 826     epot[F_LJ]       = sum_v(grpp->nener, grpp->ener[egLJSR]);
 827     epot[F_LJ14]     = sum_v(grpp->nener, grpp->ener[egLJ14]);
 828     epot[F_COUL14]   = sum_v(grpp->nener, grpp->ener[egCOUL14]);
 829     epot[F_COUL_LR]  = sum_v(grpp->nener, grpp->ener[egCOULLR]);
 830     epot[F_LJ_LR]    = sum_v(grpp->nener, grpp->ener[egLJLR]);
 831     /* We have already added 1-2,1-3, and 1-4 terms to F_GBPOL */
 832     epot[F_GBPOL]   += sum_v(grpp->nener, grpp->ener[egGB]);
 833
 834 /* lattice part of LR doesnt belong to any group
 835  * and has been added earlier
 836  */
 837     epot[F_BHAM]     = sum_v(grpp->nener, grpp->ener[egBHAMSR]);
 838     epot[F_BHAM_LR]  = sum_v(grpp->nener, grpp->ener[egBHAMLR]);
 839
 840     epot[F_EPOT] = 0;
 841     for (i = 0; (i < F_EPOT); i++)
 842     {
 843         if (i != F_DISRESVIOL && i != F_ORIRESDEV)
 844         {
 845             epot[F_EPOT] += epot[i];
 846         }
 847     }
 848 }
 849
 850 void sum_dhdl(gmx_enerdata_t *enerd, real *lambda, t_lambda *fepvals)
 851 {
 852     int    i, j, index;
 853     double dlam;
 854
 855     enerd->dvdl_lin[efptVDW] += enerd->term[F_DVDL_VDW];  /* include dispersion correction */
 856     enerd->term[F_DVDL]       = 0.0;
 857     for (i = 0; i < efptNR; i++)
 858     {
 859         if (fepvals->separate_dvdl[i])
 860         {
 861             /* could this be done more readably/compactly? */
 862             switch (i)
 863             {
 864                 case (efptMASS):
 865                     index = F_DKDL;
 866                     break;
 867                 case (efptCOUL):
 868                     index = F_DVDL_COUL;
 869                     break;
 870                 case (efptVDW):
 871                     index = F_DVDL_VDW;
 872                     break;
 873                 case (efptBONDED):
 874                     index = F_DVDL_BONDED;
 875                     break;
 876                 case (efptRESTRAINT):
 877                     index = F_DVDL_RESTRAINT;
 878                     break;
 879                 default:
 880                     index = F_DVDL;
 881                     break;
 882             }
 883             enerd->term[index] = enerd->dvdl_lin[i] + enerd->dvdl_nonlin[i];
 884             if (debug)
 885             {
 886                 fprintf(debug, "dvdl-%s[%2d]: %f: non-linear %f + linear %f\n",
 887                         efpt_names[i], i, enerd->term[index], enerd->dvdl_nonlin[i], enerd->dvdl_lin[i]);
 888             }
 889         }
 890         else
 891         {
 892             enerd->term[F_DVDL] += enerd->dvdl_lin[i] + enerd->dvdl_nonlin[i];
 893             if (debug)
 894             {
 895                 fprintf(debug, "dvd-%sl[%2d]: %f: non-linear %f + linear %f\n",
 896                         efpt_names[0], i, enerd->term[F_DVDL], enerd->dvdl_nonlin[i], enerd->dvdl_lin[i]);
 897             }
 898         }
 899     }
 900
 901     /* Notes on the foreign lambda free energy difference evaluation:
 902      * Adding the potential and ekin terms that depend linearly on lambda
 903      * as delta lam * dvdl to the energy differences is exact.
 904      * For the constraints this is not exact, but we have no other option
 905      * without literally changing the lengths and reevaluating the energies at each step.
 906      * (try to remedy this post 4.6 - MRS)
 907      * For the non-bonded LR term we assume that the soft-core (if present)
 908      * no longer affects the energy beyond the short-range cut-off,
 909      * which is a very good approximation (except for exotic settings).
 910      * (investigate how to overcome this post 4.6 - MRS)
 911      */
 912     if (fepvals->separate_dvdl[efptBONDED])
 913     {
 914         enerd->term[F_DVDL_BONDED] += enerd->term[F_DVDL_CONSTR];
 915     }
 916     else
 917     {
 918         enerd->term[F_DVDL] += enerd->term[F_DVDL_CONSTR];
 919     }
 920     enerd->term[F_DVDL_CONSTR] = 0;
 921
 922     for (i = 0; i < fepvals->n_lambda; i++)
 923     {
 924         /* note we are iterating over fepvals here!
 925            For the current lam, dlam = 0 automatically,
 926            so we don't need to add anything to the
 927            enerd->enerpart_lambda[0] */
 928
 929         /* we don't need to worry about dvdl_lin contributions to dE at
 930            current lambda, because the contributions to the current
 931            lambda are automatically zeroed */
 932
 933         for (j = 0; j < efptNR; j++)
 934         {
 935             /* Note that this loop is over all dhdl components, not just the separated ones */
 936             dlam = (fepvals->all_lambda[j][i]-lambda[j]);
 937             enerd->enerpart_lambda[i+1] += dlam*enerd->dvdl_lin[j];
 938             if (debug)
 939             {
 940                 fprintf(debug, "enerdiff lam %g: (%15s), non-linear %f linear %f*%f\n",
 941                         fepvals->all_lambda[j][i], efpt_names[j],
 942                         (enerd->enerpart_lambda[i+1] - enerd->enerpart_lambda[0]),
 943                         dlam, enerd->dvdl_lin[j]);
 944             }
 945         }
 946     }
 947 }
 948
 949
 950 void reset_foreign_enerdata(gmx_enerdata_t *enerd)
 951 {
 952     int  i, j;
 953
 954     /* First reset all foreign energy components.  Foreign energies always called on
 955        neighbor search steps */
 956     for (i = 0; (i < egNR); i++)
 957     {
 958         for (j = 0; (j < enerd->grpp.nener); j++)
 959         {
 960             enerd->foreign_grpp.ener[i][j] = 0.0;
 961         }
 962     }
 963
 964     /* potential energy components */
 965     for (i = 0; (i <= F_EPOT); i++)
 966     {
 967         enerd->foreign_term[i] = 0.0;
 968     }
 969 }
 970
 971 void reset_enerdata(t_forcerec *fr, gmx_bool bNS,
 972                     gmx_enerdata_t *enerd,
 973                     gmx_bool bMaster)
 974 {
 975     gmx_bool bKeepLR;
 976     int      i, j;
 977
 978     /* First reset all energy components, except for the long range terms
 979      * on the master at non neighbor search steps, since the long range
 980      * terms have already been summed at the last neighbor search step.
 981      */
 982     bKeepLR = (fr->bTwinRange && !bNS);
 983     for (i = 0; (i < egNR); i++)
 984     {
 985         if (!(bKeepLR && bMaster && (i == egCOULLR || i == egLJLR)))
 986         {
 987             for (j = 0; (j < enerd->grpp.nener); j++)
 988             {
 989                 enerd->grpp.ener[i][j] = 0.0;
 990             }
 991         }
 992     }
 993     for (i = 0; i < efptNR; i++)
 994     {
 995         enerd->dvdl_lin[i]    = 0.0;
 996         enerd->dvdl_nonlin[i] = 0.0;
 997     }
 998
 999     /* Normal potential energy components */
1000     for (i = 0; (i <= F_EPOT); i++)
1001     {
1002         enerd->term[i] = 0.0;
1003     }
1004     /* Initialize the dVdlambda term with the long range contribution */
1005     /* Initialize the dvdl term with the long range contribution */
1006     enerd->term[F_DVDL]            = 0.0;
1007     enerd->term[F_DVDL_COUL]       = 0.0;
1008     enerd->term[F_DVDL_VDW]        = 0.0;
1009     enerd->term[F_DVDL_BONDED]     = 0.0;
1010     enerd->term[F_DVDL_RESTRAINT]  = 0.0;
1011     enerd->term[F_DKDL]            = 0.0;
1012     if (enerd->n_lambda > 0)
1013     {
1014         for (i = 0; i < enerd->n_lambda; i++)
1015         {
1016             enerd->enerpart_lambda[i] = 0.0;
1017         }
1018     }
1019     /* reset foreign energy data - separate function since we also call it elsewhere */
1020     reset_foreign_enerdata(enerd);
1021 }