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 <assert.h>
  42 #include <math.h>
  43 #include <string.h>
  44
  45 #include "typedefs.h"
  46 #include "macros.h"
  47 #include "force.h"
  48 #include "nonbonded.h"
  49 #include "names.h"
  50 #include "network.h"
  51 #include "pbc.h"
  52 #include "ns.h"
  53 #include "nrnb.h"
  54 #include "bondf.h"
  55 #include "mshift.h"
  56 #include "txtdump.h"
  57 #include "coulomb.h"
  58 #include "pme.h"
  59 #include "mdrun.h"
  60 #include "domdec.h"
  61 #include "qmmm.h"
  62 #include "gmx_omp_nthreads.h"
  63
  64 #include "gromacs/legacyheaders/types/commrec.h"
  65 #include "gromacs/math/vec.h"
  66 #include "gromacs/timing/wallcycle.h"
  67 #include "gromacs/utility/fatalerror.h"
  68 #include "gromacs/utility/smalloc.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
 121     /* This reduction can run over any number of threads */
 122 #pragma omp parallel for num_threads(gmx_omp_nthreads_get(emntBonded)) private(t) schedule(static)
 123     for (i = 0; i < n; i++)
 124     {
 125         for (t = 1; t < nthreads; t++)
 126         {
 127             rvec_inc(f[i], f_t[t].f[i]);
 128         }
 129     }
 130     for (t = 1; t < nthreads; t++)
 131     {
 132         *Vcorr_q  += f_t[t].Vcorr_q;
 133         *Vcorr_lj += f_t[t].Vcorr_lj;
 134         *dvdl_q   += f_t[t].dvdl[efptCOUL];
 135         *dvdl_lj  += f_t[t].dvdl[efptVDW];
 136         m_add(vir_q, f_t[t].vir_q, vir_q);
 137         m_add(vir_lj, f_t[t].vir_lj, vir_lj);
 138     }
 139 }
 140
 141 void gmx_print_sepdvdl(FILE *fplog, const char *s, real v, real dvdlambda)
 142 {
 143     fprintf(fplog, "  %-30s V %12.5e  dVdl %12.5e\n", s, v, dvdlambda);
 144 }
 145
 146 void do_force_lowlevel(FILE       *fplog,   gmx_int64_t step,
 147                        t_forcerec *fr,      t_inputrec *ir,
 148                        t_idef     *idef,    t_commrec  *cr,
 149                        t_nrnb     *nrnb,    gmx_wallcycle_t wcycle,
 150                        t_mdatoms  *md,
 151                        rvec       x[],      history_t  *hist,
 152                        rvec       f[],
 153                        rvec       f_longrange[],
 154                        gmx_enerdata_t *enerd,
 155                        t_fcdata   *fcd,
 156                        gmx_localtop_t *top,
 157                        gmx_genborn_t *born,
 158                        t_atomtypes *atype,
 159                        gmx_bool       bBornRadii,
 160                        matrix     box,
 161                        t_lambda   *fepvals,
 162                        real       *lambda,
 163                        t_graph    *graph,
 164                        t_blocka   *excl,
 165                        rvec       mu_tot[],
 166                        int        flags,
 167                        float      *cycles_pme)
 168 {
 169     int         i, j;
 170     int         donb_flags;
 171     gmx_bool    bDoEpot, bSepDVDL, bSB;
 172     int         pme_flags;
 173     matrix      boxs;
 174     rvec        box_size;
 175     t_pbc       pbc;
 176     char        buf[22];
 177     double      clam_i, vlam_i;
 178     real        dvdl_dum[efptNR], dvdl_nb[efptNR], lam_i[efptNR];
 179     real        dvdl_q, dvdl_lj;
 180
 181 #ifdef GMX_MPI
 182     double  t0 = 0.0, t1, t2, t3; /* time measurement for coarse load balancing */
 183 #endif
 184
 185 #define PRINT_SEPDVDL(s, v, dvdlambda) if (bSepDVDL) { gmx_print_sepdvdl(fplog, s, v, dvdlambda); }
 186
 187     set_pbc(&pbc, fr->ePBC, box);
 188
 189     /* reset free energy components */
 190     for (i = 0; i < efptNR; i++)
 191     {
 192         dvdl_nb[i]  = 0;
 193         dvdl_dum[i] = 0;
 194     }
 195
 196     /* Reset box */
 197     for (i = 0; (i < DIM); i++)
 198     {
 199         box_size[i] = box[i][i];
 200     }
 201
 202     bSepDVDL = (fr->bSepDVDL && do_per_step(step, ir->nstlog));
 203     debug_gmx();
 204
 205     /* do QMMM first if requested */
 206     if (fr->bQMMM)
 207     {
 208         enerd->term[F_EQM] = calculate_QMMM(cr, x, f, fr);
 209     }
 210
 211     if (bSepDVDL)
 212     {
 213         fprintf(fplog, "Step %s: non-bonded V and dVdl for node %d:\n",
 214                 gmx_step_str(step, buf), cr->nodeid);
 215     }
 216
 217     /* Call the short range functions all in one go. */
 218
 219 #ifdef GMX_MPI
 220     /*#define TAKETIME ((cr->npmenodes) && (fr->timesteps < 12))*/
 221 #define TAKETIME FALSE
 222     if (TAKETIME)
 223     {
 224         MPI_Barrier(cr->mpi_comm_mygroup);
 225         t0 = MPI_Wtime();
 226     }
 227 #endif
 228
 229     if (ir->nwall)
 230     {
 231         /* foreign lambda component for walls */
 232         real dvdl_walls = do_walls(ir, fr, box, md, x, f, lambda[efptVDW],
 233                                    enerd->grpp.ener[egLJSR], nrnb);
 234         PRINT_SEPDVDL("Walls", 0.0, dvdl_walls);
 235         enerd->dvdl_lin[efptVDW] += dvdl_walls;
 236     }
 237
 238     /* If doing GB, reset dvda and calculate the Born radii */
 239     if (ir->implicit_solvent)
 240     {
 241         wallcycle_sub_start(wcycle, ewcsNONBONDED);
 242
 243         for (i = 0; i < born->nr; i++)
 244         {
 245             fr->dvda[i] = 0;
 246         }
 247
 248         if (bBornRadii)
 249         {
 250             calc_gb_rad(cr, fr, ir, top, x, &(fr->gblist), born, md, nrnb);
 251         }
 252
 253         wallcycle_sub_stop(wcycle, ewcsNONBONDED);
 254     }
 255
 256     where();
 257     /* We only do non-bonded calculation with group scheme here, the verlet
 258      * calls are done from do_force_cutsVERLET(). */
 259     if (fr->cutoff_scheme == ecutsGROUP && (flags & GMX_FORCE_NONBONDED))
 260     {
 261         donb_flags = 0;
 262         /* Add short-range interactions */
 263         donb_flags |= GMX_NONBONDED_DO_SR;
 264
 265         /* Currently all group scheme kernels always calculate (shift-)forces */
 266         if (flags & GMX_FORCE_FORCES)
 267         {
 268             donb_flags |= GMX_NONBONDED_DO_FORCE;
 269         }
 270         if (flags & GMX_FORCE_VIRIAL)
 271         {
 272             donb_flags |= GMX_NONBONDED_DO_SHIFTFORCE;
 273         }
 274         if (flags & GMX_FORCE_ENERGY)
 275         {
 276             donb_flags |= GMX_NONBONDED_DO_POTENTIAL;
 277         }
 278         if (flags & GMX_FORCE_DO_LR)
 279         {
 280             donb_flags |= GMX_NONBONDED_DO_LR;
 281         }
 282
 283         wallcycle_sub_start(wcycle, ewcsNONBONDED);
 284         do_nonbonded(fr, x, f, f_longrange, md, excl,
 285                      &enerd->grpp, nrnb,
 286                      lambda, dvdl_nb, -1, -1, donb_flags);
 287
 288         /* If we do foreign lambda and we have soft-core interactions
 289          * we have to recalculate the (non-linear) energies contributions.
 290          */
 291         if (fepvals->n_lambda > 0 && (flags & GMX_FORCE_DHDL) && fepvals->sc_alpha != 0)
 292         {
 293             for (i = 0; i < enerd->n_lambda; i++)
 294             {
 295                 for (j = 0; j < efptNR; j++)
 296                 {
 297                     lam_i[j] = (i == 0 ? lambda[j] : fepvals->all_lambda[j][i-1]);
 298                 }
 299                 reset_foreign_enerdata(enerd);
 300                 do_nonbonded(fr, x, f, f_longrange, md, excl,
 301                              &(enerd->foreign_grpp), nrnb,
 302                              lam_i, dvdl_dum, -1, -1,
 303                              (donb_flags & ~GMX_NONBONDED_DO_FORCE) | GMX_NONBONDED_DO_FOREIGNLAMBDA);
 304                 sum_epot(&(enerd->foreign_grpp), enerd->foreign_term);
 305                 enerd->enerpart_lambda[i] += enerd->foreign_term[F_EPOT];
 306             }
 307         }
 308         wallcycle_sub_stop(wcycle, ewcsNONBONDED);
 309         where();
 310     }
 311
 312     /* If we are doing GB, calculate bonded forces and apply corrections
 313      * to the solvation forces */
 314     /* MRS: Eventually, many need to include free energy contribution here! */
 315     if (ir->implicit_solvent)
 316     {
 317         wallcycle_sub_start(wcycle, ewcsBONDED);
 318         calc_gb_forces(cr, md, born, top, x, f, fr, idef,
 319                        ir->gb_algorithm, ir->sa_algorithm, nrnb, &pbc, graph, enerd);
 320         wallcycle_sub_stop(wcycle, ewcsBONDED);
 321     }
 322
 323 #ifdef GMX_MPI
 324     if (TAKETIME)
 325     {
 326         t1          = MPI_Wtime();
 327         fr->t_fnbf += t1-t0;
 328     }
 329 #endif
 330
 331     if (fepvals->sc_alpha != 0)
 332     {
 333         enerd->dvdl_nonlin[efptVDW] += dvdl_nb[efptVDW];
 334     }
 335     else
 336     {
 337         enerd->dvdl_lin[efptVDW] += dvdl_nb[efptVDW];
 338     }
 339
 340     if (fepvals->sc_alpha != 0)
 341
 342     /* even though coulomb part is linear, we already added it, beacuse we
 343        need to go through the vdw calculation anyway */
 344     {
 345         enerd->dvdl_nonlin[efptCOUL] += dvdl_nb[efptCOUL];
 346     }
 347     else
 348     {
 349         enerd->dvdl_lin[efptCOUL] += dvdl_nb[efptCOUL];
 350     }
 351
 352     if (bSepDVDL)
 353     {
 354         real V_short_range    = 0;
 355         real dvdl_short_range = 0;
 356
 357         for (i = 0; i < enerd->grpp.nener; i++)
 358         {
 359             V_short_range +=
 360                 (fr->bBHAM ?
 361                  enerd->grpp.ener[egBHAMSR][i] :
 362                  enerd->grpp.ener[egLJSR][i])
 363                 + enerd->grpp.ener[egCOULSR][i] + enerd->grpp.ener[egGB][i];
 364         }
 365         dvdl_short_range = dvdl_nb[efptVDW] + dvdl_nb[efptCOUL];
 366         PRINT_SEPDVDL("VdW and Coulomb SR particle-p.",
 367                       V_short_range,
 368                       dvdl_short_range);
 369     }
 370     debug_gmx();
 371
 372
 373     if (debug)
 374     {
 375         pr_rvecs(debug, 0, "fshift after SR", fr->fshift, SHIFTS);
 376     }
 377
 378     /* Shift the coordinates. Must be done before bonded forces and PPPM,
 379      * but is also necessary for SHAKE and update, therefore it can NOT
 380      * go when no bonded forces have to be evaluated.
 381      */
 382
 383     /* Here sometimes we would not need to shift with NBFonly,
 384      * but we do so anyhow for consistency of the returned coordinates.
 385      */
 386     if (graph)
 387     {
 388         shift_self(graph, box, x);
 389         if (TRICLINIC(box))
 390         {
 391             inc_nrnb(nrnb, eNR_SHIFTX, 2*graph->nnodes);
 392         }
 393         else
 394         {
 395             inc_nrnb(nrnb, eNR_SHIFTX, graph->nnodes);
 396         }
 397     }
 398     /* Check whether we need to do bondeds or correct for exclusions */
 399     if (fr->bMolPBC &&
 400         ((flags & GMX_FORCE_BONDED)
 401          || EEL_RF(fr->eeltype) || EEL_FULL(fr->eeltype) || EVDW_PME(fr->vdwtype)))
 402     {
 403         /* Since all atoms are in the rectangular or triclinic unit-cell,
 404          * only single box vector shifts (2 in x) are required.
 405          */
 406         set_pbc_dd(&pbc, fr->ePBC, cr->dd, TRUE, box);
 407     }
 408     debug_gmx();
 409
 410     if (flags & GMX_FORCE_BONDED)
 411     {
 412         wallcycle_sub_start(wcycle, ewcsBONDED);
 413         calc_bonds(fplog, cr->ms,
 414                    idef, x, hist, f, fr, &pbc, graph, enerd, nrnb, lambda, md, fcd,
 415                    DOMAINDECOMP(cr) ? cr->dd->gatindex : NULL, atype, born,
 416                    flags,
 417                    fr->bSepDVDL && do_per_step(step, ir->nstlog), step);
 418
 419         /* Check if we have to determine energy differences
 420          * at foreign lambda's.
 421          */
 422         if (fepvals->n_lambda > 0 && (flags & GMX_FORCE_DHDL) &&
 423             idef->ilsort != ilsortNO_FE)
 424         {
 425             if (idef->ilsort != ilsortFE_SORTED)
 426             {
 427                 gmx_incons("The bonded interactions are not sorted for free energy");
 428             }
 429             for (i = 0; i < enerd->n_lambda; i++)
 430             {
 431                 reset_foreign_enerdata(enerd);
 432                 for (j = 0; j < efptNR; j++)
 433                 {
 434                     lam_i[j] = (i == 0 ? lambda[j] : fepvals->all_lambda[j][i-1]);
 435                 }
 436                 calc_bonds_lambda(fplog, idef, x, fr, &pbc, graph, &(enerd->foreign_grpp), enerd->foreign_term, nrnb, lam_i, md,
 437                                   fcd, DOMAINDECOMP(cr) ? cr->dd->gatindex : NULL);
 438                 sum_epot(&(enerd->foreign_grpp), enerd->foreign_term);
 439                 enerd->enerpart_lambda[i] += enerd->foreign_term[F_EPOT];
 440             }
 441         }
 442         debug_gmx();
 443
 444         wallcycle_sub_stop(wcycle, ewcsBONDED);
 445     }
 446
 447     where();
 448
 449     *cycles_pme = 0;
 450     clear_mat(fr->vir_el_recip);
 451     clear_mat(fr->vir_lj_recip);
 452
 453     /* Do long-range electrostatics and/or LJ-PME, including related short-range
 454      * corrections.
 455      */
 456     if (EEL_FULL(fr->eeltype) || EVDW_PME(fr->vdwtype))
 457     {
 458         real Vlr               = 0, Vcorr = 0;
 459         real dvdl_long_range   = 0;
 460         int  status            = 0;
 461         real Vlr_q             = 0, Vlr_lj = 0, Vcorr_q = 0, Vcorr_lj = 0;
 462         real dvdl_long_range_q = 0, dvdl_long_range_lj = 0;
 463
 464         bSB = (ir->nwall == 2);
 465         if (bSB)
 466         {
 467             copy_mat(box, boxs);
 468             svmul(ir->wall_ewald_zfac, boxs[ZZ], boxs[ZZ]);
 469             box_size[ZZ] *= ir->wall_ewald_zfac;
 470         }
 471
 472         if (EEL_PME_EWALD(fr->eeltype) || EVDW_PME(fr->vdwtype))
 473         {
 474             real dvdl_long_range_correction_q   = 0;
 475             real dvdl_long_range_correction_lj  = 0;
 476             /* With the Verlet scheme exclusion forces are calculated
 477              * in the non-bonded kernel.
 478              */
 479             /* The TPI molecule does not have exclusions with the rest
 480              * of the system and no intra-molecular PME grid
 481              * contributions will be calculated in
 482              * gmx_pme_calc_energy.
 483              */
 484             if ((ir->cutoff_scheme == ecutsGROUP && fr->n_tpi == 0) ||
 485                 ir->ewald_geometry != eewg3D ||
 486                 ir->epsilon_surface != 0)
 487             {
 488                 int nthreads, t;
 489
 490                 wallcycle_sub_start(wcycle, ewcsEWALD_CORRECTION);
 491
 492                 if (fr->n_tpi > 0)
 493                 {
 494                     gmx_fatal(FARGS, "TPI with PME currently only works in a 3D geometry with tin-foil boundary conditions");
 495                 }
 496
 497                 nthreads = gmx_omp_nthreads_get(emntBonded);
 498 #pragma omp parallel for num_threads(nthreads) schedule(static)
 499                 for (t = 0; t < nthreads; t++)
 500                 {
 501                     int     s, e, i;
 502                     rvec   *fnv;
 503                     tensor *vir_q, *vir_lj;
 504                     real   *Vcorrt_q, *Vcorrt_lj, *dvdlt_q, *dvdlt_lj;
 505                     if (t == 0)
 506                     {
 507                         fnv       = fr->f_novirsum;
 508                         vir_q     = &fr->vir_el_recip;
 509                         vir_lj    = &fr->vir_lj_recip;
 510                         Vcorrt_q  = &Vcorr_q;
 511                         Vcorrt_lj = &Vcorr_lj;
 512                         dvdlt_q   = &dvdl_long_range_correction_q;
 513                         dvdlt_lj  = &dvdl_long_range_correction_lj;
 514                     }
 515                     else
 516                     {
 517                         fnv       = fr->f_t[t].f;
 518                         vir_q     = &fr->f_t[t].vir_q;
 519                         vir_lj    = &fr->f_t[t].vir_lj;
 520                         Vcorrt_q  = &fr->f_t[t].Vcorr_q;
 521                         Vcorrt_lj = &fr->f_t[t].Vcorr_lj;
 522                         dvdlt_q   = &fr->f_t[t].dvdl[efptCOUL];
 523                         dvdlt_lj  = &fr->f_t[t].dvdl[efptVDW];
 524                         for (i = 0; i < fr->natoms_force; i++)
 525                         {
 526                             clear_rvec(fnv[i]);
 527                         }
 528                         clear_mat(*vir_q);
 529                         clear_mat(*vir_lj);
 530                     }
 531                     *dvdlt_q  = 0;
 532                     *dvdlt_lj = 0;
 533
 534                     ewald_LRcorrection(fr->excl_load[t], fr->excl_load[t+1],
 535                                        cr, t, fr,
 536                                        md->chargeA,
 537                                        md->nChargePerturbed ? md->chargeB : NULL,
 538                                        md->sqrt_c6A,
 539                                        md->nTypePerturbed ? md->sqrt_c6B : NULL,
 540                                        md->sigmaA,
 541                                        md->nTypePerturbed ? md->sigmaB : NULL,
 542                                        md->sigma3A,
 543                                        md->nTypePerturbed ? md->sigma3B : NULL,
 544                                        ir->cutoff_scheme != ecutsVERLET,
 545                                        excl, x, bSB ? boxs : box, mu_tot,
 546                                        ir->ewald_geometry,
 547                                        ir->epsilon_surface,
 548                                        fnv, *vir_q, *vir_lj,
 549                                        Vcorrt_q, Vcorrt_lj,
 550                                        lambda[efptCOUL], lambda[efptVDW],
 551                                        dvdlt_q, dvdlt_lj);
 552                 }
 553                 if (nthreads > 1)
 554                 {
 555                     reduce_thread_forces(fr->natoms_force, fr->f_novirsum,
 556                                          fr->vir_el_recip, fr->vir_lj_recip,
 557                                          &Vcorr_q, &Vcorr_lj,
 558                                          &dvdl_long_range_correction_q,
 559                                          &dvdl_long_range_correction_lj,
 560                                          nthreads, fr->f_t);
 561                 }
 562                 wallcycle_sub_stop(wcycle, ewcsEWALD_CORRECTION);
 563             }
 564
 565             if (EEL_PME_EWALD(fr->eeltype) && fr->n_tpi == 0)
 566             {
 567                 Vcorr_q += ewald_charge_correction(cr, fr, lambda[efptCOUL], box,
 568                                                    &dvdl_long_range_correction_q,
 569                                                    fr->vir_el_recip);
 570             }
 571
 572             PRINT_SEPDVDL("Ewald excl./charge/dip. corr.", Vcorr_q, dvdl_long_range_correction_q);
 573             PRINT_SEPDVDL("Ewald excl. corr. LJ", Vcorr_lj, dvdl_long_range_correction_lj);
 574             enerd->dvdl_lin[efptCOUL] += dvdl_long_range_correction_q;
 575             enerd->dvdl_lin[efptVDW]  += dvdl_long_range_correction_lj;
 576
 577             if ((EEL_PME(fr->eeltype) || EVDW_PME(fr->vdwtype)) && (cr->duty & DUTY_PME))
 578             {
 579                 /* Do reciprocal PME for Coulomb and/or LJ. */
 580                 assert(fr->n_tpi >= 0);
 581                 if (fr->n_tpi == 0 || (flags & GMX_FORCE_STATECHANGED))
 582                 {
 583                     pme_flags = GMX_PME_SPREAD | GMX_PME_SOLVE;
 584                     if (EEL_PME(fr->eeltype))
 585                     {
 586                         pme_flags     |= GMX_PME_DO_COULOMB;
 587                     }
 588                     if (EVDW_PME(fr->vdwtype))
 589                     {
 590                         pme_flags |= GMX_PME_DO_LJ;
 591                     }
 592                     if (flags & GMX_FORCE_FORCES)
 593                     {
 594                         pme_flags |= GMX_PME_CALC_F;
 595                     }
 596                     if (flags & GMX_FORCE_VIRIAL)
 597                     {
 598                         pme_flags |= GMX_PME_CALC_ENER_VIR;
 599                     }
 600                     if (fr->n_tpi > 0)
 601                     {
 602                         /* We don't calculate f, but we do want the potential */
 603                         pme_flags |= GMX_PME_CALC_POT;
 604                     }
 605                     wallcycle_start(wcycle, ewcPMEMESH);
 606                     status = gmx_pme_do(fr->pmedata,
 607                                         0, md->homenr - fr->n_tpi,
 608                                         x, fr->f_novirsum,
 609                                         md->chargeA, md->chargeB,
 610                                         md->sqrt_c6A, md->sqrt_c6B,
 611                                         md->sigmaA, md->sigmaB,
 612                                         bSB ? boxs : box, cr,
 613                                         DOMAINDECOMP(cr) ? dd_pme_maxshift_x(cr->dd) : 0,
 614                                         DOMAINDECOMP(cr) ? dd_pme_maxshift_y(cr->dd) : 0,
 615                                         nrnb, wcycle,
 616                                         fr->vir_el_recip, fr->ewaldcoeff_q,
 617                                         fr->vir_lj_recip, fr->ewaldcoeff_lj,
 618                                         &Vlr_q, &Vlr_lj,
 619                                         lambda[efptCOUL], lambda[efptVDW],
 620                                         &dvdl_long_range_q, &dvdl_long_range_lj, pme_flags);
 621                     *cycles_pme = wallcycle_stop(wcycle, ewcPMEMESH);
 622                     if (status != 0)
 623                     {
 624                         gmx_fatal(FARGS, "Error %d in reciprocal PME routine", status);
 625                     }
 626                     /* We should try to do as little computation after
 627                      * this as possible, because parallel PME synchronizes
 628                      * the nodes, so we want all load imbalance of the
 629                      * rest of the force calculation to be before the PME
 630                      * call.  DD load balancing is done on the whole time
 631                      * of the force call (without PME).
 632                      */
 633                 }
 634                 if (fr->n_tpi > 0)
 635                 {
 636                     if (EVDW_PME(ir->vdwtype))
 637                     {
 638
 639                         gmx_fatal(FARGS, "Test particle insertion not implemented with LJ-PME");
 640                     }
 641                     /* Determine the PME grid energy of the test molecule
 642                      * with the PME grid potential of the other charges.
 643                      */
 644                     gmx_pme_calc_energy(fr->pmedata, fr->n_tpi,
 645                                         x + md->homenr - fr->n_tpi,
 646                                         md->chargeA + md->homenr - fr->n_tpi,
 647                                         &Vlr_q);
 648                 }
 649                 PRINT_SEPDVDL("PME mesh", Vlr_q + Vlr_lj, dvdl_long_range_q+dvdl_long_range_lj);
 650             }
 651         }
 652
 653         if (!EEL_PME(fr->eeltype) && EEL_PME_EWALD(fr->eeltype))
 654         {
 655             Vlr_q = do_ewald(ir, x, fr->f_novirsum,
 656                              md->chargeA, md->chargeB,
 657                              box_size, cr, md->homenr,
 658                              fr->vir_el_recip, fr->ewaldcoeff_q,
 659                              lambda[efptCOUL], &dvdl_long_range_q, fr->ewald_table);
 660             PRINT_SEPDVDL("Ewald long-range", Vlr_q, dvdl_long_range_q);
 661         }
 662
 663         /* Note that with separate PME nodes we get the real energies later */
 664         enerd->dvdl_lin[efptCOUL] += dvdl_long_range_q;
 665         enerd->dvdl_lin[efptVDW]  += dvdl_long_range_lj;
 666         enerd->term[F_COUL_RECIP]  = Vlr_q + Vcorr_q;
 667         enerd->term[F_LJ_RECIP]    = Vlr_lj + Vcorr_lj;
 668         if (debug)
 669         {
 670             fprintf(debug, "Vlr_q = %g, Vcorr_q = %g, Vlr_corr_q = %g\n",
 671                     Vlr_q, Vcorr_q, enerd->term[F_COUL_RECIP]);
 672             pr_rvecs(debug, 0, "vir_el_recip after corr", fr->vir_el_recip, DIM);
 673             pr_rvecs(debug, 0, "fshift after LR Corrections", fr->fshift, SHIFTS);
 674             fprintf(debug, "Vlr_lj: %g, Vcorr_lj = %g, Vlr_corr_lj = %g\n",
 675                     Vlr_lj, Vcorr_lj, enerd->term[F_LJ_RECIP]);
 676             pr_rvecs(debug, 0, "vir_lj_recip after corr", fr->vir_lj_recip, DIM);
 677         }
 678     }
 679     else
 680     {
 681         /* Is there a reaction-field exclusion correction needed? */
 682         if (EEL_RF(fr->eeltype) && eelRF_NEC != fr->eeltype)
 683         {
 684             /* With the Verlet scheme, exclusion forces are calculated
 685              * in the non-bonded kernel.
 686              */
 687             if (ir->cutoff_scheme != ecutsVERLET)
 688             {
 689                 real dvdl_rf_excl      = 0;
 690                 enerd->term[F_RF_EXCL] =
 691                     RF_excl_correction(fr, graph, md, excl, x, f,
 692                                        fr->fshift, &pbc, lambda[efptCOUL], &dvdl_rf_excl);
 693
 694                 enerd->dvdl_lin[efptCOUL] += dvdl_rf_excl;
 695                 PRINT_SEPDVDL("RF exclusion correction",
 696                               enerd->term[F_RF_EXCL], dvdl_rf_excl);
 697             }
 698         }
 699     }
 700     where();
 701     debug_gmx();
 702
 703     if (debug)
 704     {
 705         print_nrnb(debug, nrnb);
 706     }
 707     debug_gmx();
 708
 709 #ifdef GMX_MPI
 710     if (TAKETIME)
 711     {
 712         t2 = MPI_Wtime();
 713         MPI_Barrier(cr->mpi_comm_mygroup);
 714         t3          = MPI_Wtime();
 715         fr->t_wait += t3-t2;
 716         if (fr->timesteps == 11)
 717         {
 718             fprintf(stderr, "* PP load balancing info: node %d, step %s, rel wait time=%3.0f%% , load string value: %7.2f\n",
 719                     cr->nodeid, gmx_step_str(fr->timesteps, buf),
 720                     100*fr->t_wait/(fr->t_wait+fr->t_fnbf),
 721                     (fr->t_fnbf+fr->t_wait)/fr->t_fnbf);
 722         }
 723         fr->timesteps++;
 724     }
 725 #endif
 726
 727     if (debug)
 728     {
 729         pr_rvecs(debug, 0, "fshift after bondeds", fr->fshift, SHIFTS);
 730     }
 731
 732 }
 733
 734 void init_enerdata(int ngener, int n_lambda, gmx_enerdata_t *enerd)
 735 {
 736     int i, n2;
 737
 738     for (i = 0; i < F_NRE; i++)
 739     {
 740         enerd->term[i]         = 0;
 741         enerd->foreign_term[i] = 0;
 742     }
 743
 744
 745     for (i = 0; i < efptNR; i++)
 746     {
 747         enerd->dvdl_lin[i]     = 0;
 748         enerd->dvdl_nonlin[i]  = 0;
 749     }
 750
 751     n2 = ngener*ngener;
 752     if (debug)
 753     {
 754         fprintf(debug, "Creating %d sized group matrix for energies\n", n2);
 755     }
 756     enerd->grpp.nener         = n2;
 757     enerd->foreign_grpp.nener = n2;
 758     for (i = 0; (i < egNR); i++)
 759     {
 760         snew(enerd->grpp.ener[i], n2);
 761         snew(enerd->foreign_grpp.ener[i], n2);
 762     }
 763
 764     if (n_lambda)
 765     {
 766         enerd->n_lambda = 1 + n_lambda;
 767         snew(enerd->enerpart_lambda, enerd->n_lambda);
 768     }
 769     else
 770     {
 771         enerd->n_lambda = 0;
 772     }
 773 }
 774
 775 void destroy_enerdata(gmx_enerdata_t *enerd)
 776 {
 777     int i;
 778
 779     for (i = 0; (i < egNR); i++)
 780     {
 781         sfree(enerd->grpp.ener[i]);
 782     }
 783
 784     for (i = 0; (i < egNR); i++)
 785     {
 786         sfree(enerd->foreign_grpp.ener[i]);
 787     }
 788
 789     if (enerd->n_lambda)
 790     {
 791         sfree(enerd->enerpart_lambda);
 792     }
 793 }
 794
 795 static real sum_v(int n, real v[])
 796 {
 797     real t;
 798     int  i;
 799
 800     t = 0.0;
 801     for (i = 0; (i < n); i++)
 802     {
 803         t = t + v[i];
 804     }
 805
 806     return t;
 807 }
 808
 809 void sum_epot(gmx_grppairener_t *grpp, real *epot)
 810 {
 811     int i;
 812
 813     /* Accumulate energies */
 814     epot[F_COUL_SR]  = sum_v(grpp->nener, grpp->ener[egCOULSR]);
 815     epot[F_LJ]       = sum_v(grpp->nener, grpp->ener[egLJSR]);
 816     epot[F_LJ14]     = sum_v(grpp->nener, grpp->ener[egLJ14]);
 817     epot[F_COUL14]   = sum_v(grpp->nener, grpp->ener[egCOUL14]);
 818     epot[F_COUL_LR]  = sum_v(grpp->nener, grpp->ener[egCOULLR]);
 819     epot[F_LJ_LR]    = sum_v(grpp->nener, grpp->ener[egLJLR]);
 820     /* We have already added 1-2,1-3, and 1-4 terms to F_GBPOL */
 821     epot[F_GBPOL]   += sum_v(grpp->nener, grpp->ener[egGB]);
 822
 823 /* lattice part of LR doesnt belong to any group
 824  * and has been added earlier
 825  */
 826     epot[F_BHAM]     = sum_v(grpp->nener, grpp->ener[egBHAMSR]);
 827     epot[F_BHAM_LR]  = sum_v(grpp->nener, grpp->ener[egBHAMLR]);
 828
 829     epot[F_EPOT] = 0;
 830     for (i = 0; (i < F_EPOT); i++)
 831     {
 832         if (i != F_DISRESVIOL && i != F_ORIRESDEV)
 833         {
 834             epot[F_EPOT] += epot[i];
 835         }
 836     }
 837 }
 838
 839 void sum_dhdl(gmx_enerdata_t *enerd, real *lambda, t_lambda *fepvals)
 840 {
 841     int    i, j, index;
 842     double dlam;
 843
 844     enerd->dvdl_lin[efptVDW] += enerd->term[F_DVDL_VDW];  /* include dispersion correction */
 845     enerd->term[F_DVDL]       = 0.0;
 846     for (i = 0; i < efptNR; i++)
 847     {
 848         if (fepvals->separate_dvdl[i])
 849         {
 850             /* could this be done more readably/compactly? */
 851             switch (i)
 852             {
 853                 case (efptMASS):
 854                     index = F_DKDL;
 855                     break;
 856                 case (efptCOUL):
 857                     index = F_DVDL_COUL;
 858                     break;
 859                 case (efptVDW):
 860                     index = F_DVDL_VDW;
 861                     break;
 862                 case (efptBONDED):
 863                     index = F_DVDL_BONDED;
 864                     break;
 865                 case (efptRESTRAINT):
 866                     index = F_DVDL_RESTRAINT;
 867                     break;
 868                 default:
 869                     index = F_DVDL;
 870                     break;
 871             }
 872             enerd->term[index] = enerd->dvdl_lin[i] + enerd->dvdl_nonlin[i];
 873             if (debug)
 874             {
 875                 fprintf(debug, "dvdl-%s[%2d]: %f: non-linear %f + linear %f\n",
 876                         efpt_names[i], i, enerd->term[index], enerd->dvdl_nonlin[i], enerd->dvdl_lin[i]);
 877             }
 878         }
 879         else
 880         {
 881             enerd->term[F_DVDL] += enerd->dvdl_lin[i] + enerd->dvdl_nonlin[i];
 882             if (debug)
 883             {
 884                 fprintf(debug, "dvd-%sl[%2d]: %f: non-linear %f + linear %f\n",
 885                         efpt_names[0], i, enerd->term[F_DVDL], enerd->dvdl_nonlin[i], enerd->dvdl_lin[i]);
 886             }
 887         }
 888     }
 889
 890     /* Notes on the foreign lambda free energy difference evaluation:
 891      * Adding the potential and ekin terms that depend linearly on lambda
 892      * as delta lam * dvdl to the energy differences is exact.
 893      * For the constraints this is not exact, but we have no other option
 894      * without literally changing the lengths and reevaluating the energies at each step.
 895      * (try to remedy this post 4.6 - MRS)
 896      * For the non-bonded LR term we assume that the soft-core (if present)
 897      * no longer affects the energy beyond the short-range cut-off,
 898      * which is a very good approximation (except for exotic settings).
 899      * (investigate how to overcome this post 4.6 - MRS)
 900      */
 901     if (fepvals->separate_dvdl[efptBONDED])
 902     {
 903         enerd->term[F_DVDL_BONDED] += enerd->term[F_DVDL_CONSTR];
 904     }
 905     else
 906     {
 907         enerd->term[F_DVDL] += enerd->term[F_DVDL_CONSTR];
 908     }
 909     enerd->term[F_DVDL_CONSTR] = 0;
 910
 911     for (i = 0; i < fepvals->n_lambda; i++)
 912     {
 913         /* note we are iterating over fepvals here!
 914            For the current lam, dlam = 0 automatically,
 915            so we don't need to add anything to the
 916            enerd->enerpart_lambda[0] */
 917
 918         /* we don't need to worry about dvdl_lin contributions to dE at
 919            current lambda, because the contributions to the current
 920            lambda are automatically zeroed */
 921
 922         for (j = 0; j < efptNR; j++)
 923         {
 924             /* Note that this loop is over all dhdl components, not just the separated ones */
 925             dlam = (fepvals->all_lambda[j][i]-lambda[j]);
 926             enerd->enerpart_lambda[i+1] += dlam*enerd->dvdl_lin[j];
 927             if (debug)
 928             {
 929                 fprintf(debug, "enerdiff lam %g: (%15s), non-linear %f linear %f*%f\n",
 930                         fepvals->all_lambda[j][i], efpt_names[j],
 931                         (enerd->enerpart_lambda[i+1] - enerd->enerpart_lambda[0]),
 932                         dlam, enerd->dvdl_lin[j]);
 933             }
 934         }
 935     }
 936 }
 937
 938
 939 void reset_foreign_enerdata(gmx_enerdata_t *enerd)
 940 {
 941     int  i, j;
 942
 943     /* First reset all foreign energy components.  Foreign energies always called on
 944        neighbor search steps */
 945     for (i = 0; (i < egNR); i++)
 946     {
 947         for (j = 0; (j < enerd->grpp.nener); j++)
 948         {
 949             enerd->foreign_grpp.ener[i][j] = 0.0;
 950         }
 951     }
 952
 953     /* potential energy components */
 954     for (i = 0; (i <= F_EPOT); i++)
 955     {
 956         enerd->foreign_term[i] = 0.0;
 957     }
 958 }
 959
 960 void reset_enerdata(t_forcerec *fr, gmx_bool bNS,
 961                     gmx_enerdata_t *enerd,
 962                     gmx_bool bMaster)
 963 {
 964     gmx_bool bKeepLR;
 965     int      i, j;
 966
 967     /* First reset all energy components, except for the long range terms
 968      * on the master at non neighbor search steps, since the long range
 969      * terms have already been summed at the last neighbor search step.
 970      */
 971     bKeepLR = (fr->bTwinRange && !bNS);
 972     for (i = 0; (i < egNR); i++)
 973     {
 974         if (!(bKeepLR && bMaster && (i == egCOULLR || i == egLJLR)))
 975         {
 976             for (j = 0; (j < enerd->grpp.nener); j++)
 977             {
 978                 enerd->grpp.ener[i][j] = 0.0;
 979             }
 980         }
 981     }
 982     for (i = 0; i < efptNR; i++)
 983     {
 984         enerd->dvdl_lin[i]    = 0.0;
 985         enerd->dvdl_nonlin[i] = 0.0;
 986     }
 987
 988     /* Normal potential energy components */
 989     for (i = 0; (i <= F_EPOT); i++)
 990     {
 991         enerd->term[i] = 0.0;
 992     }
 993     /* Initialize the dVdlambda term with the long range contribution */
 994     /* Initialize the dvdl term with the long range contribution */
 995     enerd->term[F_DVDL]            = 0.0;
 996     enerd->term[F_DVDL_COUL]       = 0.0;
 997     enerd->term[F_DVDL_VDW]        = 0.0;
 998     enerd->term[F_DVDL_BONDED]     = 0.0;
 999     enerd->term[F_DVDL_RESTRAINT]  = 0.0;
1000     enerd->term[F_DKDL]            = 0.0;
1001     if (enerd->n_lambda > 0)
1002     {
1003         for (i = 0; i < enerd->n_lambda; i++)
1004         {
1005             enerd->enerpart_lambda[i] = 0.0;
1006         }
1007     }
1008     /* reset foreign energy data - separate function since we also call it elsewhere */
1009     reset_foreign_enerdata(enerd);
1010 }