src/gromacs/mdlib/force.cpp

   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) 2013,2014,2015,2016,2017,2018, 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 #include "gmxpre.h"
  38
  39 #include "force.h"
  40
  41 #include "config.h"
  42
  43 #include <assert.h>
  44 #include <string.h>
  45
  46 #include <cmath>
  47
  48 #include "gromacs/domdec/domdec.h"
  49 #include "gromacs/domdec/domdec_struct.h"
  50 #include "gromacs/ewald/ewald.h"
  51 #include "gromacs/ewald/long-range-correction.h"
  52 #include "gromacs/ewald/pme.h"
  53 #include "gromacs/gmxlib/network.h"
  54 #include "gromacs/gmxlib/nrnb.h"
  55 #include "gromacs/gmxlib/nonbonded/nonbonded.h"
  56 #include "gromacs/listed-forces/listed-forces.h"
  57 #include "gromacs/math/vec.h"
  58 #include "gromacs/math/vecdump.h"
  59 #include "gromacs/mdlib/forcerec-threading.h"
  60 #include "gromacs/mdlib/genborn.h"
  61 #include "gromacs/mdlib/mdrun.h"
  62 #include "gromacs/mdlib/ns.h"
  63 #include "gromacs/mdlib/qmmm.h"
  64 #include "gromacs/mdtypes/commrec.h"
  65 #include "gromacs/mdtypes/forceoutput.h"
  66 #include "gromacs/mdtypes/inputrec.h"
  67 #include "gromacs/mdtypes/md_enums.h"
  68 #include "gromacs/pbcutil/ishift.h"
  69 #include "gromacs/pbcutil/mshift.h"
  70 #include "gromacs/pbcutil/pbc.h"
  71 #include "gromacs/timing/wallcycle.h"
  72 #include "gromacs/utility/cstringutil.h"
  73 #include "gromacs/utility/exceptions.h"
  74 #include "gromacs/utility/fatalerror.h"
  75 #include "gromacs/utility/smalloc.h"
  76
  77 void ns(FILE              *fp,
  78         t_forcerec        *fr,
  79         matrix             box,
  80         gmx_groups_t      *groups,
  81         gmx_localtop_t    *top,
  82         t_mdatoms         *md,
  83         t_commrec         *cr,
  84         t_nrnb            *nrnb,
  85         gmx_bool           bFillGrid)
  86 {
  87     int     nsearch;
  88
  89
  90     if (!fr->ns->nblist_initialized)
  91     {
  92         init_neighbor_list(fp, fr, md->homenr);
  93     }
  94
  95     nsearch = search_neighbours(fp, fr, box, top, groups, cr, nrnb, md,
  96                                 bFillGrid);
  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 clearEwaldThreadOutput(ewald_corr_thread_t *ewc_t)
 114 {
 115     ewc_t->Vcorr_q        = 0;
 116     ewc_t->Vcorr_lj       = 0;
 117     ewc_t->dvdl[efptCOUL] = 0;
 118     ewc_t->dvdl[efptVDW]  = 0;
 119     clear_mat(ewc_t->vir_q);
 120     clear_mat(ewc_t->vir_lj);
 121 }
 122
 123 static void reduceEwaldThreadOuput(int nthreads, ewald_corr_thread_t *ewc_t)
 124 {
 125     ewald_corr_thread_t &dest = ewc_t[0];
 126
 127     for (int t = 1; t < nthreads; t++)
 128     {
 129         dest.Vcorr_q        += ewc_t[t].Vcorr_q;
 130         dest.Vcorr_lj       += ewc_t[t].Vcorr_lj;
 131         dest.dvdl[efptCOUL] += ewc_t[t].dvdl[efptCOUL];
 132         dest.dvdl[efptVDW]  += ewc_t[t].dvdl[efptVDW];
 133         m_add(dest.vir_q,  ewc_t[t].vir_q,  dest.vir_q);
 134         m_add(dest.vir_lj, ewc_t[t].vir_lj, dest.vir_lj);
 135     }
 136 }
 137
 138 void do_force_lowlevel(t_forcerec *fr,      t_inputrec *ir,
 139                        t_idef     *idef,    t_commrec  *cr,
 140                        t_nrnb     *nrnb,    gmx_wallcycle_t wcycle,
 141                        t_mdatoms  *md,
 142                        rvec       x[],      history_t  *hist,
 143                        rvec      *forceForUseWithShiftForces,
 144                        gmx::ForceWithVirial *forceWithVirial,
 145                        gmx_enerdata_t *enerd,
 146                        t_fcdata   *fcd,
 147                        gmx_localtop_t *top,
 148                        gmx_genborn_t *born,
 149                        gmx_bool       bBornRadii,
 150                        matrix     box,
 151                        t_lambda   *fepvals,
 152                        real       *lambda,
 153                        t_graph    *graph,
 154                        t_blocka   *excl,
 155                        rvec       mu_tot[],
 156                        int        flags,
 157                        float     *cycles_pme)
 158 {
 159     int         i, j;
 160     int         donb_flags;
 161     int         pme_flags;
 162     t_pbc       pbc;
 163     real        dvdl_dum[efptNR], dvdl_nb[efptNR];
 164
 165 #if GMX_MPI
 166     double  t0 = 0.0, t1, t2, t3; /* time measurement for coarse load balancing */
 167 #endif
 168
 169     set_pbc(&pbc, fr->ePBC, box);
 170
 171     /* reset free energy components */
 172     for (i = 0; i < efptNR; i++)
 173     {
 174         dvdl_nb[i]  = 0;
 175         dvdl_dum[i] = 0;
 176     }
 177
 178     /* do QMMM first if requested */
 179     if (fr->bQMMM)
 180     {
 181         enerd->term[F_EQM] = calculate_QMMM(cr, forceForUseWithShiftForces, fr);
 182     }
 183
 184     /* Call the short range functions all in one go. */
 185
 186 #if GMX_MPI
 187     /*#define TAKETIME ((cr->npmenodes) && (fr->timesteps < 12))*/
 188 #define TAKETIME FALSE
 189     if (TAKETIME)
 190     {
 191         MPI_Barrier(cr->mpi_comm_mygroup);
 192         t0 = MPI_Wtime();
 193     }
 194 #endif
 195
 196     if (ir->nwall)
 197     {
 198         /* foreign lambda component for walls */
 199         real dvdl_walls = do_walls(ir, fr, box, md, x, forceForUseWithShiftForces, lambda[efptVDW],
 200                                    enerd->grpp.ener[egLJSR], nrnb);
 201         enerd->dvdl_lin[efptVDW] += dvdl_walls;
 202     }
 203
 204     /* If doing GB, reset dvda and calculate the Born radii */
 205     if (ir->implicit_solvent)
 206     {
 207         wallcycle_sub_start(wcycle, ewcsNONBONDED);
 208
 209         for (i = 0; i < born->nr; i++)
 210         {
 211             fr->dvda[i] = 0;
 212         }
 213
 214         if (bBornRadii)
 215         {
 216             calc_gb_rad(cr, fr, ir, top, x, fr->gblist, born, md, nrnb);
 217         }
 218
 219         wallcycle_sub_stop(wcycle, ewcsNONBONDED);
 220     }
 221
 222     where();
 223     /* We only do non-bonded calculation with group scheme here, the verlet
 224      * calls are done from do_force_cutsVERLET(). */
 225     if (fr->cutoff_scheme == ecutsGROUP && (flags & GMX_FORCE_NONBONDED))
 226     {
 227         donb_flags = 0;
 228         /* Add short-range interactions */
 229         donb_flags |= GMX_NONBONDED_DO_SR;
 230
 231         /* Currently all group scheme kernels always calculate (shift-)forces */
 232         if (flags & GMX_FORCE_FORCES)
 233         {
 234             donb_flags |= GMX_NONBONDED_DO_FORCE;
 235         }
 236         if (flags & GMX_FORCE_VIRIAL)
 237         {
 238             donb_flags |= GMX_NONBONDED_DO_SHIFTFORCE;
 239         }
 240         if (flags & GMX_FORCE_ENERGY)
 241         {
 242             donb_flags |= GMX_NONBONDED_DO_POTENTIAL;
 243         }
 244
 245         wallcycle_sub_start(wcycle, ewcsNONBONDED);
 246         do_nonbonded(fr, x, forceForUseWithShiftForces, md, excl,
 247                      &enerd->grpp, nrnb,
 248                      lambda, dvdl_nb, -1, -1, donb_flags);
 249
 250         /* If we do foreign lambda and we have soft-core interactions
 251          * we have to recalculate the (non-linear) energies contributions.
 252          */
 253         if (fepvals->n_lambda > 0 && (flags & GMX_FORCE_DHDL) && fepvals->sc_alpha != 0)
 254         {
 255             for (i = 0; i < enerd->n_lambda; i++)
 256             {
 257                 real lam_i[efptNR];
 258
 259                 for (j = 0; j < efptNR; j++)
 260                 {
 261                     lam_i[j] = (i == 0 ? lambda[j] : fepvals->all_lambda[j][i-1]);
 262                 }
 263                 reset_foreign_enerdata(enerd);
 264                 do_nonbonded(fr, x, forceForUseWithShiftForces, md, excl,
 265                              &(enerd->foreign_grpp), nrnb,
 266                              lam_i, dvdl_dum, -1, -1,
 267                              (donb_flags & ~GMX_NONBONDED_DO_FORCE) | GMX_NONBONDED_DO_FOREIGNLAMBDA);
 268                 sum_epot(&(enerd->foreign_grpp), enerd->foreign_term);
 269                 enerd->enerpart_lambda[i] += enerd->foreign_term[F_EPOT];
 270             }
 271         }
 272         wallcycle_sub_stop(wcycle, ewcsNONBONDED);
 273         where();
 274     }
 275
 276     /* If we are doing GB, calculate bonded forces and apply corrections
 277      * to the solvation forces */
 278     /* MRS: Eventually, many need to include free energy contribution here! */
 279     if (ir->implicit_solvent)
 280     {
 281         wallcycle_sub_start(wcycle, ewcsLISTED);
 282         calc_gb_forces(cr, md, born, top, x, forceForUseWithShiftForces, fr, idef,
 283                        ir->gb_algorithm, ir->sa_algorithm, nrnb, &pbc, graph, enerd);
 284         wallcycle_sub_stop(wcycle, ewcsLISTED);
 285     }
 286
 287 #if GMX_MPI
 288     if (TAKETIME)
 289     {
 290         t1          = MPI_Wtime();
 291         fr->t_fnbf += t1-t0;
 292     }
 293 #endif
 294
 295     if (fepvals->sc_alpha != 0)
 296     {
 297         enerd->dvdl_nonlin[efptVDW] += dvdl_nb[efptVDW];
 298     }
 299     else
 300     {
 301         enerd->dvdl_lin[efptVDW] += dvdl_nb[efptVDW];
 302     }
 303
 304     if (fepvals->sc_alpha != 0)
 305
 306     /* even though coulomb part is linear, we already added it, beacuse we
 307        need to go through the vdw calculation anyway */
 308     {
 309         enerd->dvdl_nonlin[efptCOUL] += dvdl_nb[efptCOUL];
 310     }
 311     else
 312     {
 313         enerd->dvdl_lin[efptCOUL] += dvdl_nb[efptCOUL];
 314     }
 315
 316     if (debug)
 317     {
 318         pr_rvecs(debug, 0, "fshift after SR", fr->fshift, SHIFTS);
 319     }
 320
 321     /* Shift the coordinates. Must be done before listed forces and PPPM,
 322      * but is also necessary for SHAKE and update, therefore it can NOT
 323      * go when no listed forces have to be evaluated.
 324      *
 325      * The shifting and PBC code is deliberately not timed, since with
 326      * the Verlet scheme it only takes non-zero time with triclinic
 327      * boxes, and even then the time is around a factor of 100 less
 328      * than the next smallest counter.
 329      */
 330
 331
 332     /* Here sometimes we would not need to shift with NBFonly,
 333      * but we do so anyhow for consistency of the returned coordinates.
 334      */
 335     if (graph)
 336     {
 337         shift_self(graph, box, x);
 338         if (TRICLINIC(box))
 339         {
 340             inc_nrnb(nrnb, eNR_SHIFTX, 2*graph->nnodes);
 341         }
 342         else
 343         {
 344             inc_nrnb(nrnb, eNR_SHIFTX, graph->nnodes);
 345         }
 346     }
 347     /* Check whether we need to do listed interactions or correct for exclusions */
 348     if (fr->bMolPBC &&
 349         ((flags & GMX_FORCE_LISTED)
 350          || EEL_RF(fr->ic->eeltype) || EEL_FULL(fr->ic->eeltype) || EVDW_PME(fr->ic->vdwtype)))
 351     {
 352         /* TODO There are no electrostatics methods that require this
 353            transformation, when using the Verlet scheme, so update the
 354            above conditional. */
 355         /* Since all atoms are in the rectangular or triclinic unit-cell,
 356          * only single box vector shifts (2 in x) are required.
 357          */
 358         set_pbc_dd(&pbc, fr->ePBC, DOMAINDECOMP(cr) ? cr->dd->nc : nullptr,
 359                    TRUE, box);
 360     }
 361
 362     do_force_listed(wcycle, box, ir->fepvals, cr,
 363                     idef, (const rvec *) x, hist,
 364                     forceForUseWithShiftForces, forceWithVirial,
 365                     fr, &pbc, graph, enerd, nrnb, lambda, md, fcd,
 366                     DOMAINDECOMP(cr) ? cr->dd->gatindex : nullptr,
 367                     flags);
 368
 369     where();
 370
 371     *cycles_pme = 0;
 372
 373     /* Do long-range electrostatics and/or LJ-PME, including related short-range
 374      * corrections.
 375      */
 376     if (EEL_FULL(fr->ic->eeltype) || EVDW_PME(fr->ic->vdwtype))
 377     {
 378         int  status            = 0;
 379         real Vlr_q             = 0, Vlr_lj = 0;
 380
 381         /* We reduce all virial, dV/dlambda and energy contributions, except
 382          * for the reciprocal energies (Vlr_q, Vlr_lj) into the same struct.
 383          */
 384         ewald_corr_thread_t &ewaldOutput = fr->ewc_t[0];
 385         clearEwaldThreadOutput(&ewaldOutput);
 386
 387         if (EEL_PME_EWALD(fr->ic->eeltype) || EVDW_PME(fr->ic->vdwtype))
 388         {
 389             /* With the Verlet scheme exclusion forces are calculated
 390              * in the non-bonded kernel.
 391              */
 392             /* The TPI molecule does not have exclusions with the rest
 393              * of the system and no intra-molecular PME grid
 394              * contributions will be calculated in
 395              * gmx_pme_calc_energy.
 396              */
 397             if ((ir->cutoff_scheme == ecutsGROUP && fr->n_tpi == 0) ||
 398                 ir->ewald_geometry != eewg3D ||
 399                 ir->epsilon_surface != 0)
 400             {
 401                 int nthreads, t;
 402
 403                 wallcycle_sub_start(wcycle, ewcsEWALD_CORRECTION);
 404
 405                 if (fr->n_tpi > 0)
 406                 {
 407                     gmx_fatal(FARGS, "TPI with PME currently only works in a 3D geometry with tin-foil boundary conditions");
 408                 }
 409
 410                 nthreads = fr->nthread_ewc;
 411 #pragma omp parallel for num_threads(nthreads) schedule(static)
 412                 for (t = 0; t < nthreads; t++)
 413                 {
 414                     try
 415                     {
 416                         ewald_corr_thread_t &ewc_t = fr->ewc_t[t];
 417                         if (t > 0)
 418                         {
 419                             clearEwaldThreadOutput(&ewc_t);
 420                         }
 421
 422                         /* Threading is only supported with the Verlet cut-off
 423                          * scheme and then only single particle forces (no
 424                          * exclusion forces) are calculated, so we can store
 425                          * the forces in the normal, single forceWithVirial->force_ array.
 426                          */
 427                         ewald_LRcorrection(md->homenr, cr, nthreads, t, fr, ir,
 428                                            md->chargeA, md->chargeB,
 429                                            md->sqrt_c6A, md->sqrt_c6B,
 430                                            md->sigmaA, md->sigmaB,
 431                                            md->sigma3A, md->sigma3B,
 432                                            md->nChargePerturbed || md->nTypePerturbed,
 433                                            ir->cutoff_scheme != ecutsVERLET,
 434                                            excl, x, box, mu_tot,
 435                                            ir->ewald_geometry,
 436                                            ir->epsilon_surface,
 437                                            as_rvec_array(forceWithVirial->force_.data()),
 438                                            ewc_t.vir_q, ewc_t.vir_lj,
 439                                            &ewc_t.Vcorr_q, &ewc_t.Vcorr_lj,
 440                                            lambda[efptCOUL], lambda[efptVDW],
 441                                            &ewc_t.dvdl[efptCOUL], &ewc_t.dvdl[efptVDW]);
 442                     }
 443                     GMX_CATCH_ALL_AND_EXIT_WITH_FATAL_ERROR;
 444                 }
 445                 if (nthreads > 1)
 446                 {
 447                     reduceEwaldThreadOuput(nthreads, fr->ewc_t);
 448                 }
 449                 wallcycle_sub_stop(wcycle, ewcsEWALD_CORRECTION);
 450             }
 451
 452             if (EEL_PME_EWALD(fr->ic->eeltype) && fr->n_tpi == 0)
 453             {
 454                 /* This is not in a subcounter because it takes a
 455                    negligible and constant-sized amount of time */
 456                 ewaldOutput.Vcorr_q +=
 457                     ewald_charge_correction(cr, fr, lambda[efptCOUL], box,
 458                                             &ewaldOutput.dvdl[efptCOUL],
 459                                             ewaldOutput.vir_q);
 460             }
 461
 462             if ((EEL_PME(fr->ic->eeltype) || EVDW_PME(fr->ic->vdwtype)) &&
 463                 thisRankHasDuty(cr, DUTY_PME) && (pme_run_mode(fr->pmedata) == PmeRunMode::CPU))
 464             {
 465                 /* Do reciprocal PME for Coulomb and/or LJ. */
 466                 assert(fr->n_tpi >= 0);
 467                 if (fr->n_tpi == 0 || (flags & GMX_FORCE_STATECHANGED))
 468                 {
 469                     pme_flags = GMX_PME_SPREAD | GMX_PME_SOLVE;
 470
 471                     if (flags & GMX_FORCE_FORCES)
 472                     {
 473                         pme_flags |= GMX_PME_CALC_F;
 474                     }
 475                     if (flags & GMX_FORCE_VIRIAL)
 476                     {
 477                         pme_flags |= GMX_PME_CALC_ENER_VIR;
 478                     }
 479                     if (fr->n_tpi > 0)
 480                     {
 481                         /* We don't calculate f, but we do want the potential */
 482                         pme_flags |= GMX_PME_CALC_POT;
 483                     }
 484
 485                     /* With domain decomposition we close the CPU side load
 486                      * balancing region here, because PME does global
 487                      * communication that acts as a global barrier.
 488                      */
 489                     if (DOMAINDECOMP(cr))
 490                     {
 491                         ddCloseBalanceRegionCpu(cr->dd);
 492                     }
 493
 494                     wallcycle_start(wcycle, ewcPMEMESH);
 495                     status = gmx_pme_do(fr->pmedata,
 496                                         0, md->homenr - fr->n_tpi,
 497                                         x,
 498                                         as_rvec_array(forceWithVirial->force_.data()),
 499                                         md->chargeA, md->chargeB,
 500                                         md->sqrt_c6A, md->sqrt_c6B,
 501                                         md->sigmaA, md->sigmaB,
 502                                         box, cr,
 503                                         DOMAINDECOMP(cr) ? dd_pme_maxshift_x(cr->dd) : 0,
 504                                         DOMAINDECOMP(cr) ? dd_pme_maxshift_y(cr->dd) : 0,
 505                                         nrnb, wcycle,
 506                                         ewaldOutput.vir_q, ewaldOutput.vir_lj,
 507                                         &Vlr_q, &Vlr_lj,
 508                                         lambda[efptCOUL], lambda[efptVDW],
 509                                         &ewaldOutput.dvdl[efptCOUL],
 510                                         &ewaldOutput.dvdl[efptVDW],
 511                                         pme_flags);
 512                     *cycles_pme = wallcycle_stop(wcycle, ewcPMEMESH);
 513                     if (status != 0)
 514                     {
 515                         gmx_fatal(FARGS, "Error %d in reciprocal PME routine", status);
 516                     }
 517
 518                     /* We should try to do as little computation after
 519                      * this as possible, because parallel PME synchronizes
 520                      * the nodes, so we want all load imbalance of the
 521                      * rest of the force calculation to be before the PME
 522                      * call.  DD load balancing is done on the whole time
 523                      * of the force call (without PME).
 524                      */
 525                 }
 526                 if (fr->n_tpi > 0)
 527                 {
 528                     if (EVDW_PME(ir->vdwtype))
 529                     {
 530
 531                         gmx_fatal(FARGS, "Test particle insertion not implemented with LJ-PME");
 532                     }
 533                     /* Determine the PME grid energy of the test molecule
 534                      * with the PME grid potential of the other charges.
 535                      */
 536                     gmx_pme_calc_energy(fr->pmedata, fr->n_tpi,
 537                                         x + md->homenr - fr->n_tpi,
 538                                         md->chargeA + md->homenr - fr->n_tpi,
 539                                         &Vlr_q);
 540                 }
 541             }
 542         }
 543
 544         if (!EEL_PME(fr->ic->eeltype) && EEL_PME_EWALD(fr->ic->eeltype))
 545         {
 546             Vlr_q = do_ewald(ir, x, as_rvec_array(forceWithVirial->force_.data()),
 547                              md->chargeA, md->chargeB,
 548                              box, cr, md->homenr,
 549                              ewaldOutput.vir_q, fr->ic->ewaldcoeff_q,
 550                              lambda[efptCOUL], &ewaldOutput.dvdl[efptCOUL],
 551                              fr->ewald_table);
 552         }
 553
 554         /* Note that with separate PME nodes we get the real energies later */
 555         forceWithVirial->addVirialContribution(ewaldOutput.vir_q);
 556         forceWithVirial->addVirialContribution(ewaldOutput.vir_lj);
 557         enerd->dvdl_lin[efptCOUL] += ewaldOutput.dvdl[efptCOUL];
 558         enerd->dvdl_lin[efptVDW]  += ewaldOutput.dvdl[efptVDW];
 559         enerd->term[F_COUL_RECIP]  = Vlr_q + ewaldOutput.Vcorr_q;
 560         enerd->term[F_LJ_RECIP]    = Vlr_lj + ewaldOutput.Vcorr_lj;
 561
 562         if (debug)
 563         {
 564             fprintf(debug, "Vlr_q = %g, Vcorr_q = %g, Vlr_corr_q = %g\n",
 565                     Vlr_q, ewaldOutput.Vcorr_q, enerd->term[F_COUL_RECIP]);
 566             pr_rvecs(debug, 0, "vir_el_recip after corr", ewaldOutput.vir_q, DIM);
 567             pr_rvecs(debug, 0, "fshift after LR Corrections", fr->fshift, SHIFTS);
 568             fprintf(debug, "Vlr_lj: %g, Vcorr_lj = %g, Vlr_corr_lj = %g\n",
 569                     Vlr_lj, ewaldOutput.Vcorr_lj, enerd->term[F_LJ_RECIP]);
 570             pr_rvecs(debug, 0, "vir_lj_recip after corr", ewaldOutput.vir_lj, DIM);
 571         }
 572     }
 573     else
 574     {
 575         /* Is there a reaction-field exclusion correction needed?
 576          * With the Verlet scheme, exclusion forces are calculated
 577          * in the non-bonded kernel.
 578          */
 579         if (ir->cutoff_scheme != ecutsVERLET && EEL_RF(fr->ic->eeltype))
 580         {
 581             real dvdl_rf_excl      = 0;
 582             enerd->term[F_RF_EXCL] =
 583                 RF_excl_correction(fr, graph, md, excl, x, forceForUseWithShiftForces,
 584                                    fr->fshift, &pbc, lambda[efptCOUL], &dvdl_rf_excl);
 585
 586             enerd->dvdl_lin[efptCOUL] += dvdl_rf_excl;
 587         }
 588     }
 589     where();
 590
 591     if (debug)
 592     {
 593         print_nrnb(debug, nrnb);
 594     }
 595
 596 #if GMX_MPI
 597     if (TAKETIME)
 598     {
 599         t2 = MPI_Wtime();
 600         MPI_Barrier(cr->mpi_comm_mygroup);
 601         t3          = MPI_Wtime();
 602         fr->t_wait += t3-t2;
 603         if (fr->timesteps == 11)
 604         {
 605             char buf[22];
 606             fprintf(stderr, "* PP load balancing info: rank %d, step %s, rel wait time=%3.0f%% , load string value: %7.2f\n",
 607                     cr->nodeid, gmx_step_str(fr->timesteps, buf),
 608                     100*fr->t_wait/(fr->t_wait+fr->t_fnbf),
 609                     (fr->t_fnbf+fr->t_wait)/fr->t_fnbf);
 610         }
 611         fr->timesteps++;
 612     }
 613 #endif
 614
 615     if (debug)
 616     {
 617         pr_rvecs(debug, 0, "fshift after bondeds", fr->fshift, SHIFTS);
 618     }
 619
 620 }
 621
 622 void init_enerdata(int ngener, int n_lambda, gmx_enerdata_t *enerd)
 623 {
 624     int i, n2;
 625
 626     for (i = 0; i < F_NRE; i++)
 627     {
 628         enerd->term[i]         = 0;
 629         enerd->foreign_term[i] = 0;
 630     }
 631
 632
 633     for (i = 0; i < efptNR; i++)
 634     {
 635         enerd->dvdl_lin[i]     = 0;
 636         enerd->dvdl_nonlin[i]  = 0;
 637     }
 638
 639     n2 = ngener*ngener;
 640     if (debug)
 641     {
 642         fprintf(debug, "Creating %d sized group matrix for energies\n", n2);
 643     }
 644     enerd->grpp.nener         = n2;
 645     enerd->foreign_grpp.nener = n2;
 646     for (i = 0; (i < egNR); i++)
 647     {
 648         snew(enerd->grpp.ener[i], n2);
 649         snew(enerd->foreign_grpp.ener[i], n2);
 650     }
 651
 652     if (n_lambda)
 653     {
 654         enerd->n_lambda = 1 + n_lambda;
 655         snew(enerd->enerpart_lambda, enerd->n_lambda);
 656     }
 657     else
 658     {
 659         enerd->n_lambda = 0;
 660     }
 661 }
 662
 663 void destroy_enerdata(gmx_enerdata_t *enerd)
 664 {
 665     int i;
 666
 667     for (i = 0; (i < egNR); i++)
 668     {
 669         sfree(enerd->grpp.ener[i]);
 670     }
 671
 672     for (i = 0; (i < egNR); i++)
 673     {
 674         sfree(enerd->foreign_grpp.ener[i]);
 675     }
 676
 677     if (enerd->n_lambda)
 678     {
 679         sfree(enerd->enerpart_lambda);
 680     }
 681 }
 682
 683 static real sum_v(int n, real v[])
 684 {
 685     real t;
 686     int  i;
 687
 688     t = 0.0;
 689     for (i = 0; (i < n); i++)
 690     {
 691         t = t + v[i];
 692     }
 693
 694     return t;
 695 }
 696
 697 void sum_epot(gmx_grppairener_t *grpp, real *epot)
 698 {
 699     int i;
 700
 701     /* Accumulate energies */
 702     epot[F_COUL_SR]  = sum_v(grpp->nener, grpp->ener[egCOULSR]);
 703     epot[F_LJ]       = sum_v(grpp->nener, grpp->ener[egLJSR]);
 704     epot[F_LJ14]     = sum_v(grpp->nener, grpp->ener[egLJ14]);
 705     epot[F_COUL14]   = sum_v(grpp->nener, grpp->ener[egCOUL14]);
 706     /* We have already added 1-2,1-3, and 1-4 terms to F_GBPOL */
 707     epot[F_GBPOL]   += sum_v(grpp->nener, grpp->ener[egGB]);
 708
 709 /* lattice part of LR doesnt belong to any group
 710  * and has been added earlier
 711  */
 712     epot[F_BHAM]     = sum_v(grpp->nener, grpp->ener[egBHAMSR]);
 713
 714     epot[F_EPOT] = 0;
 715     for (i = 0; (i < F_EPOT); i++)
 716     {
 717         if (i != F_DISRESVIOL && i != F_ORIRESDEV)
 718         {
 719             epot[F_EPOT] += epot[i];
 720         }
 721     }
 722 }
 723
 724 void sum_dhdl(gmx_enerdata_t *enerd, gmx::ArrayRef<const real> lambda, t_lambda *fepvals)
 725 {
 726     int    index;
 727
 728     enerd->dvdl_lin[efptVDW] += enerd->term[F_DVDL_VDW];  /* include dispersion correction */
 729     enerd->term[F_DVDL]       = 0.0;
 730     for (int i = 0; i < efptNR; i++)
 731     {
 732         if (fepvals->separate_dvdl[i])
 733         {
 734             /* could this be done more readably/compactly? */
 735             switch (i)
 736             {
 737                 case (efptMASS):
 738                     index = F_DKDL;
 739                     break;
 740                 case (efptCOUL):
 741                     index = F_DVDL_COUL;
 742                     break;
 743                 case (efptVDW):
 744                     index = F_DVDL_VDW;
 745                     break;
 746                 case (efptBONDED):
 747                     index = F_DVDL_BONDED;
 748                     break;
 749                 case (efptRESTRAINT):
 750                     index = F_DVDL_RESTRAINT;
 751                     break;
 752                 default:
 753                     index = F_DVDL;
 754                     break;
 755             }
 756             enerd->term[index] = enerd->dvdl_lin[i] + enerd->dvdl_nonlin[i];
 757             if (debug)
 758             {
 759                 fprintf(debug, "dvdl-%s[%2d]: %f: non-linear %f + linear %f\n",
 760                         efpt_names[i], i, enerd->term[index], enerd->dvdl_nonlin[i], enerd->dvdl_lin[i]);
 761             }
 762         }
 763         else
 764         {
 765             enerd->term[F_DVDL] += enerd->dvdl_lin[i] + enerd->dvdl_nonlin[i];
 766             if (debug)
 767             {
 768                 fprintf(debug, "dvd-%sl[%2d]: %f: non-linear %f + linear %f\n",
 769                         efpt_names[0], i, enerd->term[F_DVDL], enerd->dvdl_nonlin[i], enerd->dvdl_lin[i]);
 770             }
 771         }
 772     }
 773
 774     if (fepvals->separate_dvdl[efptBONDED])
 775     {
 776         enerd->term[F_DVDL_BONDED] += enerd->term[F_DVDL_CONSTR];
 777     }
 778     else
 779     {
 780         enerd->term[F_DVDL] += enerd->term[F_DVDL_CONSTR];
 781     }
 782
 783     for (int i = 0; i < fepvals->n_lambda; i++)
 784     {
 785         /* note we are iterating over fepvals here!
 786            For the current lam, dlam = 0 automatically,
 787            so we don't need to add anything to the
 788            enerd->enerpart_lambda[0] */
 789
 790         /* we don't need to worry about dvdl_lin contributions to dE at
 791            current lambda, because the contributions to the current
 792            lambda are automatically zeroed */
 793
 794         double &enerpart_lambda = enerd->enerpart_lambda[i + 1];
 795
 796         for (size_t j = 0; j < lambda.size(); j++)
 797         {
 798             /* Note that this loop is over all dhdl components, not just the separated ones */
 799             const double dlam  = fepvals->all_lambda[j][i] - lambda[j];
 800
 801             enerpart_lambda   += dlam*enerd->dvdl_lin[j];
 802
 803             /* Constraints can not be evaluated at foreign lambdas, so we add
 804              * a linear extrapolation. This is an approximation, but usually
 805              * quite accurate since constraints change little between lambdas.
 806              */
 807             if ((j == efptBONDED && fepvals->separate_dvdl[efptBONDED]) ||
 808                 (j == efptFEP && !fepvals->separate_dvdl[efptBONDED]))
 809             {
 810                 enerpart_lambda += dlam*enerd->term[F_DVDL_CONSTR];
 811             }
 812
 813             if (j == efptMASS)
 814             {
 815                 enerpart_lambda += dlam*enerd->term[F_DKDL];
 816             }
 817
 818             if (debug)
 819             {
 820                 fprintf(debug, "enerdiff lam %g: (%15s), non-linear %f linear %f*%f\n",
 821                         fepvals->all_lambda[j][i], efpt_names[j],
 822                         enerpart_lambda - enerd->enerpart_lambda[0],
 823                         dlam, enerd->dvdl_lin[j]);
 824             }
 825         }
 826     }
 827
 828     /* The constrain contribution is now included in other terms, so clear it */
 829     enerd->term[F_DVDL_CONSTR] = 0;
 830 }
 831
 832
 833 void reset_foreign_enerdata(gmx_enerdata_t *enerd)
 834 {
 835     int  i, j;
 836
 837     /* First reset all foreign energy components.  Foreign energies always called on
 838        neighbor search steps */
 839     for (i = 0; (i < egNR); i++)
 840     {
 841         for (j = 0; (j < enerd->grpp.nener); j++)
 842         {
 843             enerd->foreign_grpp.ener[i][j] = 0.0;
 844         }
 845     }
 846
 847     /* potential energy components */
 848     for (i = 0; (i <= F_EPOT); i++)
 849     {
 850         enerd->foreign_term[i] = 0.0;
 851     }
 852 }
 853
 854 void reset_enerdata(gmx_enerdata_t *enerd)
 855 {
 856     int      i, j;
 857
 858     /* First reset all energy components. */
 859     for (i = 0; (i < egNR); i++)
 860     {
 861         for (j = 0; (j < enerd->grpp.nener); j++)
 862         {
 863             enerd->grpp.ener[i][j] = 0.0;
 864         }
 865     }
 866     for (i = 0; i < efptNR; i++)
 867     {
 868         enerd->dvdl_lin[i]    = 0.0;
 869         enerd->dvdl_nonlin[i] = 0.0;
 870     }
 871
 872     /* Normal potential energy components */
 873     for (i = 0; (i <= F_EPOT); i++)
 874     {
 875         enerd->term[i] = 0.0;
 876     }
 877     enerd->term[F_DVDL]            = 0.0;
 878     enerd->term[F_DVDL_COUL]       = 0.0;
 879     enerd->term[F_DVDL_VDW]        = 0.0;
 880     enerd->term[F_DVDL_BONDED]     = 0.0;
 881     enerd->term[F_DVDL_RESTRAINT]  = 0.0;
 882     enerd->term[F_DKDL]            = 0.0;
 883     if (enerd->n_lambda > 0)
 884     {
 885         for (i = 0; i < enerd->n_lambda; i++)
 886         {
 887             enerd->enerpart_lambda[i] = 0.0;
 888         }
 889     }
 890     /* reset foreign energy data - separate function since we also call it elsewhere */
 891     reset_foreign_enerdata(enerd);
 892 }