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