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