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) 2013,2014,2015,2016,2017,2018,2019, 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 <cassert>
  44 #include <cmath>
  45 #include <cstring>
  46
  47 #include "gromacs/domdec/dlbtiming.h"
  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/force_flags.h"
  60 #include "gromacs/mdlib/forcerec_threading.h"
  61 #include "gromacs/mdlib/ns.h"
  62 #include "gromacs/mdlib/qmmm.h"
  63 #include "gromacs/mdlib/rf_util.h"
  64 #include "gromacs/mdlib/wall.h"
  65 #include "gromacs/mdtypes/commrec.h"
  66 #include "gromacs/mdtypes/enerdata.h"
  67 #include "gromacs/mdtypes/forceoutput.h"
  68 #include "gromacs/mdtypes/forcerec.h"
  69 #include "gromacs/mdtypes/inputrec.h"
  70 #include "gromacs/mdtypes/md_enums.h"
  71 #include "gromacs/pbcutil/ishift.h"
  72 #include "gromacs/pbcutil/mshift.h"
  73 #include "gromacs/pbcutil/pbc.h"
  74 #include "gromacs/timing/wallcycle.h"
  75 #include "gromacs/utility/cstringutil.h"
  76 #include "gromacs/utility/exceptions.h"
  77 #include "gromacs/utility/fatalerror.h"
  78 #include "gromacs/utility/smalloc.h"
  79
  80 void ns(FILE               *fp,
  81         t_forcerec         *fr,
  82         matrix              box,
  83         const gmx_groups_t *groups,
  84         gmx_localtop_t     *top,
  85         const t_mdatoms    *md,
  86         const t_commrec    *cr,
  87         t_nrnb             *nrnb,
  88         gmx_bool            bFillGrid)
  89 {
  90     int     nsearch;
  91
  92
  93     if (!fr->ns->nblist_initialized)
  94     {
  95         init_neighbor_list(fp, fr, md->homenr);
  96     }
  97
  98     nsearch = search_neighbours(fp, fr, box, top, groups, cr, nrnb, md,
  99                                 bFillGrid);
 100     if (debug)
 101     {
 102         fprintf(debug, "nsearch = %d\n", nsearch);
 103     }
 104
 105     /* Check whether we have to do dynamic load balancing */
 106     /*if ((nsb->nstDlb > 0) && (mod(step,nsb->nstDlb) == 0))
 107        count_nb(cr,nsb,&(top->blocks[ebCGS]),nns,fr->nlr,
 108        &(top->idef),opts->ngener);
 109      */
 110     if (fr->ns->dump_nl > 0)
 111     {
 112         dump_nblist(fp, cr, fr, fr->ns->dump_nl);
 113     }
 114 }
 115
 116 static void clearEwaldThreadOutput(ewald_corr_thread_t *ewc_t)
 117 {
 118     ewc_t->Vcorr_q        = 0;
 119     ewc_t->Vcorr_lj       = 0;
 120     ewc_t->dvdl[efptCOUL] = 0;
 121     ewc_t->dvdl[efptVDW]  = 0;
 122     clear_mat(ewc_t->vir_q);
 123     clear_mat(ewc_t->vir_lj);
 124 }
 125
 126 static void reduceEwaldThreadOuput(int nthreads, ewald_corr_thread_t *ewc_t)
 127 {
 128     ewald_corr_thread_t &dest = ewc_t[0];
 129
 130     for (int t = 1; t < nthreads; t++)
 131     {
 132         dest.Vcorr_q        += ewc_t[t].Vcorr_q;
 133         dest.Vcorr_lj       += ewc_t[t].Vcorr_lj;
 134         dest.dvdl[efptCOUL] += ewc_t[t].dvdl[efptCOUL];
 135         dest.dvdl[efptVDW]  += ewc_t[t].dvdl[efptVDW];
 136         m_add(dest.vir_q,  ewc_t[t].vir_q,  dest.vir_q);
 137         m_add(dest.vir_lj, ewc_t[t].vir_lj, dest.vir_lj);
 138     }
 139 }
 140
 141 void do_force_lowlevel(t_forcerec                   *fr,
 142                        const t_inputrec             *ir,
 143                        const t_idef                 *idef,
 144                        const t_commrec              *cr,
 145                        const gmx_multisim_t         *ms,
 146                        t_nrnb                       *nrnb,
 147                        gmx_wallcycle_t               wcycle,
 148                        const t_mdatoms              *md,
 149                        rvec                          x[],
 150                        history_t                    *hist,
 151                        rvec                         *forceForUseWithShiftForces,
 152                        gmx::ForceWithVirial         *forceWithVirial,
 153                        gmx_enerdata_t               *enerd,
 154                        t_fcdata                     *fcd,
 155                        matrix                        box,
 156                        t_lambda                     *fepvals,
 157                        real                         *lambda,
 158                        const t_graph                *graph,
 159                        const t_blocka               *excl,
 160                        rvec                          mu_tot[],
 161                        int                           flags,
 162                        float                        *cycles_pme,
 163                        const DDBalanceRegionHandler &ddBalanceRegionHandler)
 164 {
 165     int         i, j;
 166     int         donb_flags;
 167     int         pme_flags;
 168     t_pbc       pbc;
 169     real        dvdl_dum[efptNR], dvdl_nb[efptNR];
 170
 171 #if GMX_MPI
 172     double  t0 = 0.0, t1, t2, t3; /* time measurement for coarse load balancing */
 173 #endif
 174
 175     set_pbc(&pbc, fr->ePBC, box);
 176
 177     /* reset free energy components */
 178     for (i = 0; i < efptNR; i++)
 179     {
 180         dvdl_nb[i]  = 0;
 181         dvdl_dum[i] = 0;
 182     }
 183
 184     /* do QMMM first if requested */
 185     if (fr->bQMMM)
 186     {
 187         enerd->term[F_EQM] = calculate_QMMM(cr, forceForUseWithShiftForces, fr);
 188     }
 189
 190     /* Call the short range functions all in one go. */
 191
 192 #if GMX_MPI
 193     /*#define TAKETIME ((cr->npmenodes) && (fr->timesteps < 12))*/
 194 #define TAKETIME FALSE
 195     if (TAKETIME)
 196     {
 197         MPI_Barrier(cr->mpi_comm_mygroup);
 198         t0 = MPI_Wtime();
 199     }
 200 #endif
 201
 202     if (ir->nwall)
 203     {
 204         /* foreign lambda component for walls */
 205         real dvdl_walls = do_walls(*ir, *fr, box, *md, x,
 206                                    forceWithVirial, lambda[efptVDW],
 207                                    enerd->grpp.ener[egLJSR], nrnb);
 208         enerd->dvdl_lin[efptVDW] += dvdl_walls;
 209     }
 210
 211     /* We only do non-bonded calculation with group scheme here, the verlet
 212      * calls are done from do_force_cutsVERLET(). */
 213     if (fr->cutoff_scheme == ecutsGROUP && (flags & GMX_FORCE_NONBONDED))
 214     {
 215         donb_flags = 0;
 216         /* Add short-range interactions */
 217         donb_flags |= GMX_NONBONDED_DO_SR;
 218
 219         /* Currently all group scheme kernels always calculate (shift-)forces */
 220         if (flags & GMX_FORCE_FORCES)
 221         {
 222             donb_flags |= GMX_NONBONDED_DO_FORCE;
 223         }
 224         if (flags & GMX_FORCE_VIRIAL)
 225         {
 226             donb_flags |= GMX_NONBONDED_DO_SHIFTFORCE;
 227         }
 228         if (flags & GMX_FORCE_ENERGY)
 229         {
 230             donb_flags |= GMX_NONBONDED_DO_POTENTIAL;
 231         }
 232
 233         wallcycle_sub_start(wcycle, ewcsNONBONDED);
 234         do_nonbonded(fr, x, forceForUseWithShiftForces, md, excl,
 235                      &enerd->grpp, nrnb,
 236                      lambda, dvdl_nb, -1, -1, donb_flags);
 237
 238         /* If we do foreign lambda and we have soft-core interactions
 239          * we have to recalculate the (non-linear) energies contributions.
 240          */
 241         if (fepvals->n_lambda > 0 && (flags & GMX_FORCE_DHDL) && fepvals->sc_alpha != 0)
 242         {
 243             for (i = 0; i < enerd->n_lambda; i++)
 244             {
 245                 real lam_i[efptNR];
 246
 247                 for (j = 0; j < efptNR; j++)
 248                 {
 249                     lam_i[j] = (i == 0 ? lambda[j] : fepvals->all_lambda[j][i-1]);
 250                 }
 251                 reset_foreign_enerdata(enerd);
 252                 do_nonbonded(fr, x, forceForUseWithShiftForces, md, excl,
 253                              &(enerd->foreign_grpp), nrnb,
 254                              lam_i, dvdl_dum, -1, -1,
 255                              (donb_flags & ~GMX_NONBONDED_DO_FORCE) | GMX_NONBONDED_DO_FOREIGNLAMBDA);
 256                 sum_epot(&(enerd->foreign_grpp), enerd->foreign_term);
 257                 enerd->enerpart_lambda[i] += enerd->foreign_term[F_EPOT];
 258             }
 259         }
 260         wallcycle_sub_stop(wcycle, ewcsNONBONDED);
 261     }
 262
 263 #if GMX_MPI
 264     if (TAKETIME)
 265     {
 266         t1          = MPI_Wtime();
 267         fr->t_fnbf += t1-t0;
 268     }
 269 #endif
 270
 271     if (fepvals->sc_alpha != 0)
 272     {
 273         enerd->dvdl_nonlin[efptVDW] += dvdl_nb[efptVDW];
 274     }
 275     else
 276     {
 277         enerd->dvdl_lin[efptVDW] += dvdl_nb[efptVDW];
 278     }
 279
 280     if (fepvals->sc_alpha != 0)
 281
 282     /* even though coulomb part is linear, we already added it, beacuse we
 283        need to go through the vdw calculation anyway */
 284     {
 285         enerd->dvdl_nonlin[efptCOUL] += dvdl_nb[efptCOUL];
 286     }
 287     else
 288     {
 289         enerd->dvdl_lin[efptCOUL] += dvdl_nb[efptCOUL];
 290     }
 291
 292     if (debug)
 293     {
 294         pr_rvecs(debug, 0, "fshift after SR", fr->fshift, SHIFTS);
 295     }
 296
 297     /* Shift the coordinates. Must be done before listed forces and PPPM,
 298      * but is also necessary for SHAKE and update, therefore it can NOT
 299      * go when no listed forces have to be evaluated.
 300      *
 301      * The shifting and PBC code is deliberately not timed, since with
 302      * the Verlet scheme it only takes non-zero time with triclinic
 303      * boxes, and even then the time is around a factor of 100 less
 304      * than the next smallest counter.
 305      */
 306
 307
 308     /* Here sometimes we would not need to shift with NBFonly,
 309      * but we do so anyhow for consistency of the returned coordinates.
 310      */
 311     if (graph)
 312     {
 313         shift_self(graph, box, x);
 314         if (TRICLINIC(box))
 315         {
 316             inc_nrnb(nrnb, eNR_SHIFTX, 2*graph->nnodes);
 317         }
 318         else
 319         {
 320             inc_nrnb(nrnb, eNR_SHIFTX, graph->nnodes);
 321         }
 322     }
 323     /* Check whether we need to do listed interactions or correct for exclusions */
 324     if (fr->bMolPBC &&
 325         ((flags & GMX_FORCE_LISTED)
 326          || EEL_RF(fr->ic->eeltype) || EEL_FULL(fr->ic->eeltype) || EVDW_PME(fr->ic->vdwtype)))
 327     {
 328         /* TODO There are no electrostatics methods that require this
 329            transformation, when using the Verlet scheme, so update the
 330            above conditional. */
 331         /* Since all atoms are in the rectangular or triclinic unit-cell,
 332          * only single box vector shifts (2 in x) are required.
 333          */
 334         set_pbc_dd(&pbc, fr->ePBC, DOMAINDECOMP(cr) ? cr->dd->nc : nullptr,
 335                    TRUE, box);
 336     }
 337
 338     do_force_listed(wcycle, box, ir->fepvals, cr, ms,
 339                     idef, x, hist,
 340                     forceForUseWithShiftForces, forceWithVirial,
 341                     fr, &pbc, graph, enerd, nrnb, lambda, md, fcd,
 342                     DOMAINDECOMP(cr) ? cr->dd->globalAtomIndices.data() : nullptr,
 343                     flags);
 344
 345
 346     *cycles_pme = 0;
 347
 348     /* Do long-range electrostatics and/or LJ-PME, including related short-range
 349      * corrections.
 350      */
 351     if (EEL_FULL(fr->ic->eeltype) || EVDW_PME(fr->ic->vdwtype))
 352     {
 353         int  status            = 0;
 354         real Vlr_q             = 0, Vlr_lj = 0;
 355
 356         /* We reduce all virial, dV/dlambda and energy contributions, except
 357          * for the reciprocal energies (Vlr_q, Vlr_lj) into the same struct.
 358          */
 359         ewald_corr_thread_t &ewaldOutput = fr->ewc_t[0];
 360         clearEwaldThreadOutput(&ewaldOutput);
 361
 362         if (EEL_PME_EWALD(fr->ic->eeltype) || EVDW_PME(fr->ic->vdwtype))
 363         {
 364             /* With the Verlet scheme exclusion forces are calculated
 365              * in the non-bonded kernel.
 366              */
 367             /* The TPI molecule does not have exclusions with the rest
 368              * of the system and no intra-molecular PME grid
 369              * contributions will be calculated in
 370              * gmx_pme_calc_energy.
 371              */
 372             if ((ir->cutoff_scheme == ecutsGROUP && fr->n_tpi == 0) ||
 373                 ir->ewald_geometry != eewg3D ||
 374                 ir->epsilon_surface != 0)
 375             {
 376                 int nthreads, t;
 377
 378                 wallcycle_sub_start(wcycle, ewcsEWALD_CORRECTION);
 379
 380                 if (fr->n_tpi > 0)
 381                 {
 382                     gmx_fatal(FARGS, "TPI with PME currently only works in a 3D geometry with tin-foil boundary conditions");
 383                 }
 384
 385                 nthreads = fr->nthread_ewc;
 386 #pragma omp parallel for num_threads(nthreads) schedule(static)
 387                 for (t = 0; t < nthreads; t++)
 388                 {
 389                     try
 390                     {
 391                         ewald_corr_thread_t &ewc_t = fr->ewc_t[t];
 392                         if (t > 0)
 393                         {
 394                             clearEwaldThreadOutput(&ewc_t);
 395                         }
 396
 397                         /* Threading is only supported with the Verlet cut-off
 398                          * scheme and then only single particle forces (no
 399                          * exclusion forces) are calculated, so we can store
 400                          * the forces in the normal, single forceWithVirial->force_ array.
 401                          */
 402                         ewald_LRcorrection(md->homenr, cr, nthreads, t, fr, ir,
 403                                            md->chargeA, md->chargeB,
 404                                            md->sqrt_c6A, md->sqrt_c6B,
 405                                            md->sigmaA, md->sigmaB,
 406                                            md->sigma3A, md->sigma3B,
 407                                            (md->nChargePerturbed != 0) || (md->nTypePerturbed != 0),
 408                                            ir->cutoff_scheme != ecutsVERLET,
 409                                            excl, x, box, mu_tot,
 410                                            ir->ewald_geometry,
 411                                            ir->epsilon_surface,
 412                                            as_rvec_array(forceWithVirial->force_.data()),
 413                                            ewc_t.vir_q, ewc_t.vir_lj,
 414                                            &ewc_t.Vcorr_q, &ewc_t.Vcorr_lj,
 415                                            lambda[efptCOUL], lambda[efptVDW],
 416                                            &ewc_t.dvdl[efptCOUL], &ewc_t.dvdl[efptVDW]);
 417                     }
 418                     GMX_CATCH_ALL_AND_EXIT_WITH_FATAL_ERROR;
 419                 }
 420                 if (nthreads > 1)
 421                 {
 422                     reduceEwaldThreadOuput(nthreads, fr->ewc_t);
 423                 }
 424                 wallcycle_sub_stop(wcycle, ewcsEWALD_CORRECTION);
 425             }
 426
 427             if (EEL_PME_EWALD(fr->ic->eeltype) && fr->n_tpi == 0)
 428             {
 429                 /* This is not in a subcounter because it takes a
 430                    negligible and constant-sized amount of time */
 431                 ewaldOutput.Vcorr_q +=
 432                     ewald_charge_correction(cr, fr, lambda[efptCOUL], box,
 433                                             &ewaldOutput.dvdl[efptCOUL],
 434                                             ewaldOutput.vir_q);
 435             }
 436
 437             if ((EEL_PME(fr->ic->eeltype) || EVDW_PME(fr->ic->vdwtype)) &&
 438                 thisRankHasDuty(cr, DUTY_PME) && (pme_run_mode(fr->pmedata) == PmeRunMode::CPU))
 439             {
 440                 /* Do reciprocal PME for Coulomb and/or LJ. */
 441                 assert(fr->n_tpi >= 0);
 442                 if (fr->n_tpi == 0 || (flags & GMX_FORCE_STATECHANGED))
 443                 {
 444                     pme_flags = GMX_PME_SPREAD | GMX_PME_SOLVE;
 445
 446                     if (flags & GMX_FORCE_FORCES)
 447                     {
 448                         pme_flags |= GMX_PME_CALC_F;
 449                     }
 450                     if (flags & GMX_FORCE_VIRIAL)
 451                     {
 452                         pme_flags |= GMX_PME_CALC_ENER_VIR;
 453                     }
 454                     if (fr->n_tpi > 0)
 455                     {
 456                         /* We don't calculate f, but we do want the potential */
 457                         pme_flags |= GMX_PME_CALC_POT;
 458                     }
 459
 460                     /* With domain decomposition we close the CPU side load
 461                      * balancing region here, because PME does global
 462                      * communication that acts as a global barrier.
 463                      */
 464                     ddBalanceRegionHandler.closeAfterForceComputationCpu();
 465
 466                     wallcycle_start(wcycle, ewcPMEMESH);
 467                     status = gmx_pme_do(fr->pmedata,
 468                                         0, md->homenr - fr->n_tpi,
 469                                         x,
 470                                         as_rvec_array(forceWithVirial->force_.data()),
 471                                         md->chargeA, md->chargeB,
 472                                         md->sqrt_c6A, md->sqrt_c6B,
 473                                         md->sigmaA, md->sigmaB,
 474                                         box, cr,
 475                                         DOMAINDECOMP(cr) ? dd_pme_maxshift_x(cr->dd) : 0,
 476                                         DOMAINDECOMP(cr) ? dd_pme_maxshift_y(cr->dd) : 0,
 477                                         nrnb, wcycle,
 478                                         ewaldOutput.vir_q, ewaldOutput.vir_lj,
 479                                         &Vlr_q, &Vlr_lj,
 480                                         lambda[efptCOUL], lambda[efptVDW],
 481                                         &ewaldOutput.dvdl[efptCOUL],
 482                                         &ewaldOutput.dvdl[efptVDW],
 483                                         pme_flags);
 484                     *cycles_pme = wallcycle_stop(wcycle, ewcPMEMESH);
 485                     if (status != 0)
 486                     {
 487                         gmx_fatal(FARGS, "Error %d in reciprocal PME routine", status);
 488                     }
 489
 490                     /* We should try to do as little computation after
 491                      * this as possible, because parallel PME synchronizes
 492                      * the nodes, so we want all load imbalance of the
 493                      * rest of the force calculation to be before the PME
 494                      * call.  DD load balancing is done on the whole time
 495                      * of the force call (without PME).
 496                      */
 497                 }
 498                 if (fr->n_tpi > 0)
 499                 {
 500                     if (EVDW_PME(ir->vdwtype))
 501                     {
 502
 503                         gmx_fatal(FARGS, "Test particle insertion not implemented with LJ-PME");
 504                     }
 505                     /* Determine the PME grid energy of the test molecule
 506                      * with the PME grid potential of the other charges.
 507                      */
 508                     gmx_pme_calc_energy(fr->pmedata, fr->n_tpi,
 509                                         x + md->homenr - fr->n_tpi,
 510                                         md->chargeA + md->homenr - fr->n_tpi,
 511                                         &Vlr_q);
 512                 }
 513             }
 514         }
 515
 516         if (!EEL_PME(fr->ic->eeltype) && EEL_PME_EWALD(fr->ic->eeltype))
 517         {
 518             Vlr_q = do_ewald(ir, x, as_rvec_array(forceWithVirial->force_.data()),
 519                              md->chargeA, md->chargeB,
 520                              box, cr, md->homenr,
 521                              ewaldOutput.vir_q, fr->ic->ewaldcoeff_q,
 522                              lambda[efptCOUL], &ewaldOutput.dvdl[efptCOUL],
 523                              fr->ewald_table);
 524         }
 525
 526         /* Note that with separate PME nodes we get the real energies later */
 527         // TODO it would be simpler if we just accumulated a single
 528         // long-range virial contribution.
 529         forceWithVirial->addVirialContribution(ewaldOutput.vir_q);
 530         forceWithVirial->addVirialContribution(ewaldOutput.vir_lj);
 531         enerd->dvdl_lin[efptCOUL] += ewaldOutput.dvdl[efptCOUL];
 532         enerd->dvdl_lin[efptVDW]  += ewaldOutput.dvdl[efptVDW];
 533         enerd->term[F_COUL_RECIP]  = Vlr_q + ewaldOutput.Vcorr_q;
 534         enerd->term[F_LJ_RECIP]    = Vlr_lj + ewaldOutput.Vcorr_lj;
 535
 536         if (debug)
 537         {
 538             fprintf(debug, "Vlr_q = %g, Vcorr_q = %g, Vlr_corr_q = %g\n",
 539                     Vlr_q, ewaldOutput.Vcorr_q, enerd->term[F_COUL_RECIP]);
 540             pr_rvecs(debug, 0, "vir_el_recip after corr", ewaldOutput.vir_q, DIM);
 541             pr_rvecs(debug, 0, "fshift after LR Corrections", fr->fshift, SHIFTS);
 542             fprintf(debug, "Vlr_lj: %g, Vcorr_lj = %g, Vlr_corr_lj = %g\n",
 543                     Vlr_lj, ewaldOutput.Vcorr_lj, enerd->term[F_LJ_RECIP]);
 544             pr_rvecs(debug, 0, "vir_lj_recip after corr", ewaldOutput.vir_lj, DIM);
 545         }
 546     }
 547     else
 548     {
 549         /* Is there a reaction-field exclusion correction needed?
 550          * With the Verlet scheme, exclusion forces are calculated
 551          * in the non-bonded kernel.
 552          */
 553         if (ir->cutoff_scheme != ecutsVERLET && EEL_RF(fr->ic->eeltype))
 554         {
 555             real dvdl_rf_excl      = 0;
 556             enerd->term[F_RF_EXCL] =
 557                 RF_excl_correction(fr, graph, md, excl, DOMAINDECOMP(cr),
 558                                    x, forceForUseWithShiftForces,
 559                                    fr->fshift, &pbc, lambda[efptCOUL], &dvdl_rf_excl);
 560
 561             enerd->dvdl_lin[efptCOUL] += dvdl_rf_excl;
 562         }
 563     }
 564
 565     if (debug)
 566     {
 567         print_nrnb(debug, nrnb);
 568     }
 569
 570 #if GMX_MPI
 571     if (TAKETIME)
 572     {
 573         t2 = MPI_Wtime();
 574         MPI_Barrier(cr->mpi_comm_mygroup);
 575         t3          = MPI_Wtime();
 576         fr->t_wait += t3-t2;
 577         if (fr->timesteps == 11)
 578         {
 579             char buf[22];
 580             fprintf(stderr, "* PP load balancing info: rank %d, step %s, rel wait time=%3.0f%% , load string value: %7.2f\n",
 581                     cr->nodeid, gmx_step_str(fr->timesteps, buf),
 582                     100*fr->t_wait/(fr->t_wait+fr->t_fnbf),
 583                     (fr->t_fnbf+fr->t_wait)/fr->t_fnbf);
 584         }
 585         fr->timesteps++;
 586     }
 587 #endif
 588
 589     if (debug)
 590     {
 591         pr_rvecs(debug, 0, "fshift after bondeds", fr->fshift, SHIFTS);
 592     }
 593
 594 }
 595
 596 void init_enerdata(int ngener, int n_lambda, gmx_enerdata_t *enerd)
 597 {
 598     int i, n2;
 599
 600     for (i = 0; i < F_NRE; i++)
 601     {
 602         enerd->term[i]         = 0;
 603         enerd->foreign_term[i] = 0;
 604     }
 605
 606
 607     for (i = 0; i < efptNR; i++)
 608     {
 609         enerd->dvdl_lin[i]     = 0;
 610         enerd->dvdl_nonlin[i]  = 0;
 611     }
 612
 613     n2 = ngener*ngener;
 614     if (debug)
 615     {
 616         fprintf(debug, "Creating %d sized group matrix for energies\n", n2);
 617     }
 618     enerd->grpp.nener         = n2;
 619     enerd->foreign_grpp.nener = n2;
 620     for (i = 0; (i < egNR); i++)
 621     {
 622         snew(enerd->grpp.ener[i], n2);
 623         snew(enerd->foreign_grpp.ener[i], n2);
 624     }
 625
 626     if (n_lambda)
 627     {
 628         enerd->n_lambda = 1 + n_lambda;
 629         snew(enerd->enerpart_lambda, enerd->n_lambda);
 630     }
 631     else
 632     {
 633         enerd->n_lambda = 0;
 634     }
 635 }
 636
 637 void destroy_enerdata(gmx_enerdata_t *enerd)
 638 {
 639     int i;
 640
 641     for (i = 0; (i < egNR); i++)
 642     {
 643         sfree(enerd->grpp.ener[i]);
 644     }
 645
 646     for (i = 0; (i < egNR); i++)
 647     {
 648         sfree(enerd->foreign_grpp.ener[i]);
 649     }
 650
 651     if (enerd->n_lambda)
 652     {
 653         sfree(enerd->enerpart_lambda);
 654     }
 655 }
 656
 657 static real sum_v(int n, const real v[])
 658 {
 659     real t;
 660     int  i;
 661
 662     t = 0.0;
 663     for (i = 0; (i < n); i++)
 664     {
 665         t = t + v[i];
 666     }
 667
 668     return t;
 669 }
 670
 671 void sum_epot(gmx_grppairener_t *grpp, real *epot)
 672 {
 673     int i;
 674
 675     /* Accumulate energies */
 676     epot[F_COUL_SR]  = sum_v(grpp->nener, grpp->ener[egCOULSR]);
 677     epot[F_LJ]       = sum_v(grpp->nener, grpp->ener[egLJSR]);
 678     epot[F_LJ14]     = sum_v(grpp->nener, grpp->ener[egLJ14]);
 679     epot[F_COUL14]   = sum_v(grpp->nener, grpp->ener[egCOUL14]);
 680
 681 /* lattice part of LR doesnt belong to any group
 682  * and has been added earlier
 683  */
 684     epot[F_BHAM]     = sum_v(grpp->nener, grpp->ener[egBHAMSR]);
 685
 686     epot[F_EPOT] = 0;
 687     for (i = 0; (i < F_EPOT); i++)
 688     {
 689         if (i != F_DISRESVIOL && i != F_ORIRESDEV)
 690         {
 691             epot[F_EPOT] += epot[i];
 692         }
 693     }
 694 }
 695
 696 void sum_dhdl(gmx_enerdata_t *enerd, gmx::ArrayRef<const real> lambda, t_lambda *fepvals)
 697 {
 698     int    index;
 699
 700     enerd->dvdl_lin[efptVDW] += enerd->term[F_DVDL_VDW];  /* include dispersion correction */
 701     enerd->term[F_DVDL]       = 0.0;
 702     for (int i = 0; i < efptNR; i++)
 703     {
 704         if (fepvals->separate_dvdl[i])
 705         {
 706             /* could this be done more readably/compactly? */
 707             switch (i)
 708             {
 709                 case (efptMASS):
 710                     index = F_DKDL;
 711                     break;
 712                 case (efptCOUL):
 713                     index = F_DVDL_COUL;
 714                     break;
 715                 case (efptVDW):
 716                     index = F_DVDL_VDW;
 717                     break;
 718                 case (efptBONDED):
 719                     index = F_DVDL_BONDED;
 720                     break;
 721                 case (efptRESTRAINT):
 722                     index = F_DVDL_RESTRAINT;
 723                     break;
 724                 default:
 725                     index = F_DVDL;
 726                     break;
 727             }
 728             enerd->term[index] = enerd->dvdl_lin[i] + enerd->dvdl_nonlin[i];
 729             if (debug)
 730             {
 731                 fprintf(debug, "dvdl-%s[%2d]: %f: non-linear %f + linear %f\n",
 732                         efpt_names[i], i, enerd->term[index], enerd->dvdl_nonlin[i], enerd->dvdl_lin[i]);
 733             }
 734         }
 735         else
 736         {
 737             enerd->term[F_DVDL] += enerd->dvdl_lin[i] + enerd->dvdl_nonlin[i];
 738             if (debug)
 739             {
 740                 fprintf(debug, "dvd-%sl[%2d]: %f: non-linear %f + linear %f\n",
 741                         efpt_names[0], i, enerd->term[F_DVDL], enerd->dvdl_nonlin[i], enerd->dvdl_lin[i]);
 742             }
 743         }
 744     }
 745
 746     if (fepvals->separate_dvdl[efptBONDED])
 747     {
 748         enerd->term[F_DVDL_BONDED] += enerd->term[F_DVDL_CONSTR];
 749     }
 750     else
 751     {
 752         enerd->term[F_DVDL] += enerd->term[F_DVDL_CONSTR];
 753     }
 754
 755     for (int i = 0; i < fepvals->n_lambda; i++)
 756     {
 757         /* note we are iterating over fepvals here!
 758            For the current lam, dlam = 0 automatically,
 759            so we don't need to add anything to the
 760            enerd->enerpart_lambda[0] */
 761
 762         /* we don't need to worry about dvdl_lin contributions to dE at
 763            current lambda, because the contributions to the current
 764            lambda are automatically zeroed */
 765
 766         double &enerpart_lambda = enerd->enerpart_lambda[i + 1];
 767
 768         for (gmx::index j = 0; j < lambda.ssize(); j++)
 769         {
 770             /* Note that this loop is over all dhdl components, not just the separated ones */
 771             const double dlam  = fepvals->all_lambda[j][i] - lambda[j];
 772
 773             enerpart_lambda   += dlam*enerd->dvdl_lin[j];
 774
 775             /* Constraints can not be evaluated at foreign lambdas, so we add
 776              * a linear extrapolation. This is an approximation, but usually
 777              * quite accurate since constraints change little between lambdas.
 778              */
 779             if ((j == efptBONDED && fepvals->separate_dvdl[efptBONDED]) ||
 780                 (j == efptFEP && !fepvals->separate_dvdl[efptBONDED]))
 781             {
 782                 enerpart_lambda += dlam*enerd->term[F_DVDL_CONSTR];
 783             }
 784
 785             if (j == efptMASS && !fepvals->separate_dvdl[j])
 786             {
 787                 enerpart_lambda += dlam*enerd->term[F_DKDL];
 788             }
 789
 790             if (debug)
 791             {
 792                 fprintf(debug, "enerdiff lam %g: (%15s), non-linear %f linear %f*%f\n",
 793                         fepvals->all_lambda[j][i], efpt_names[j],
 794                         enerpart_lambda - enerd->enerpart_lambda[0],
 795                         dlam, enerd->dvdl_lin[j]);
 796             }
 797         }
 798     }
 799
 800     /* The constrain contribution is now included in other terms, so clear it */
 801     enerd->term[F_DVDL_CONSTR] = 0;
 802 }
 803
 804
 805 void reset_foreign_enerdata(gmx_enerdata_t *enerd)
 806 {
 807     int  i, j;
 808
 809     /* First reset all foreign energy components.  Foreign energies always called on
 810        neighbor search steps */
 811     for (i = 0; (i < egNR); i++)
 812     {
 813         for (j = 0; (j < enerd->grpp.nener); j++)
 814         {
 815             enerd->foreign_grpp.ener[i][j] = 0.0;
 816         }
 817     }
 818
 819     /* potential energy components */
 820     for (i = 0; (i <= F_EPOT); i++)
 821     {
 822         enerd->foreign_term[i] = 0.0;
 823     }
 824 }
 825
 826 void reset_enerdata(gmx_enerdata_t *enerd)
 827 {
 828     int      i, j;
 829
 830     /* First reset all energy components. */
 831     for (i = 0; (i < egNR); i++)
 832     {
 833         for (j = 0; (j < enerd->grpp.nener); j++)
 834         {
 835             enerd->grpp.ener[i][j] = 0.0;
 836         }
 837     }
 838     for (i = 0; i < efptNR; i++)
 839     {
 840         enerd->dvdl_lin[i]    = 0.0;
 841         enerd->dvdl_nonlin[i] = 0.0;
 842     }
 843
 844     /* Normal potential energy components */
 845     for (i = 0; (i <= F_EPOT); i++)
 846     {
 847         enerd->term[i] = 0.0;
 848     }
 849     enerd->term[F_DVDL]            = 0.0;
 850     enerd->term[F_DVDL_COUL]       = 0.0;
 851     enerd->term[F_DVDL_VDW]        = 0.0;
 852     enerd->term[F_DVDL_BONDED]     = 0.0;
 853     enerd->term[F_DVDL_RESTRAINT]  = 0.0;
 854     enerd->term[F_DKDL]            = 0.0;
 855     if (enerd->n_lambda > 0)
 856     {
 857         for (i = 0; i < enerd->n_lambda; i++)
 858         {
 859             enerd->enerpart_lambda[i] = 0.0;
 860         }
 861     }
 862     /* reset foreign energy data - separate function since we also call it elsewhere */
 863     reset_foreign_enerdata(enerd);
 864 }