src/mdlib/force.c

   1 /* -*- mode: c; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4; c-file-style: "stroustrup"; -*-
   2  *
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  *          GROningen MAchine for Chemical Simulations
   9  *
  10  *                        VERSION 3.2.0
  11  * Written by David van der Spoel, Erik Lindahl, Berk Hess, and others.
  12  * Copyright (c) 1991-2000, University of Groningen, The Netherlands.
  13  * Copyright (c) 2001-2004, The GROMACS development team,
  14  * check out http://www.gromacs.org for more information.
  15
  16  * This program is free software; you can redistribute it and/or
  17  * modify it under the terms of the GNU General Public License
  18  * as published by the Free Software Foundation; either version 2
  19  * of the License, or (at your option) any later version.
  20  *
  21  * If you want to redistribute modifications, please consider that
  22  * scientific software is very special. Version control is crucial -
  23  * bugs must be traceable. We will be happy to consider code for
  24  * inclusion in the official distribution, but derived work must not
  25  * be called official GROMACS. Details are found in the README & COPYING
  26  * files - if they are missing, get the official version at www.gromacs.org.
  27  *
  28  * To help us fund GROMACS development, we humbly ask that you cite
  29  * the papers on the package - you can find them in the top README file.
  30  *
  31  * For more info, check our website at http://www.gromacs.org
  32  *
  33  * And Hey:
  34  * GROwing Monsters And Cloning Shrimps
  35  */
  36 #ifdef HAVE_CONFIG_H
  37 #include <config.h>
  38 #endif
  39
  40 #include <math.h>
  41 #include <string.h>
  42 #include <assert.h>
  43 #include "sysstuff.h"
  44 #include "typedefs.h"
  45 #include "macros.h"
  46 #include "smalloc.h"
  47 #include "macros.h"
  48 #include "physics.h"
  49 #include "force.h"
  50 #include "nonbonded.h"
  51 #include "names.h"
  52 #include "network.h"
  53 #include "pbc.h"
  54 #include "ns.h"
  55 #include "nrnb.h"
  56 #include "bondf.h"
  57 #include "mshift.h"
  58 #include "txtdump.h"
  59 #include "coulomb.h"
  60 #include "pme.h"
  61 #include "mdrun.h"
  62 #include "domdec.h"
  63 #include "partdec.h"
  64 #include "qmmm.h"
  65 #include "mpelogging.h"
  66 #include "gmx_omp_nthreads.h"
  67
  68 void ns(FILE *fp,
  69         t_forcerec *fr,
  70         rvec       x[],
  71         matrix     box,
  72         gmx_groups_t *groups,
  73         t_grpopts  *opts,
  74         gmx_localtop_t *top,
  75         t_mdatoms  *md,
  76         t_commrec  *cr,
  77         t_nrnb     *nrnb,
  78         real       *lambda,
  79         real       *dvdlambda,
  80         gmx_grppairener_t *grppener,
  81         gmx_bool       bFillGrid,
  82         gmx_bool       bDoLongRangeNS)
  83 {
  84   char   *ptr;
  85   int    nsearch;
  86
  87   GMX_MPE_LOG(ev_ns_start);
  88   if (!fr->ns.nblist_initialized)
  89   {
  90       init_neighbor_list(fp, fr, md->homenr);
  91   }
  92
  93   if (fr->bTwinRange)
  94     fr->nlr=0;
  95
  96     nsearch = search_neighbours(fp,fr,x,box,top,groups,cr,nrnb,md,
  97                                 lambda,dvdlambda,grppener,
  98                                 bFillGrid,bDoLongRangeNS,TRUE);
  99   if (debug)
 100     fprintf(debug,"nsearch = %d\n",nsearch);
 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     dump_nblist(fp,cr,fr,fr->ns.dump_nl);
 109
 110   GMX_MPE_LOG(ev_ns_finish);
 111 }
 112
 113 static void reduce_thread_forces(int n,rvec *f,
 114                                  tensor vir,
 115                                  real *Vcorr,
 116                                  int efpt_ind,real *dvdl,
 117                                  int nthreads,f_thread_t *f_t)
 118 {
 119     int t,i;
 120
 121     /* This reduction can run over any number of threads */
 122 #pragma omp parallel for num_threads(gmx_omp_nthreads_get(emntBonded)) private(t) schedule(static)
 123     for(i=0; i<n; i++)
 124     {
 125         for(t=1; t<nthreads; t++)
 126         {
 127             rvec_inc(f[i],f_t[t].f[i]);
 128         }
 129     }
 130     for(t=1; t<nthreads; t++)
 131     {
 132         *Vcorr += f_t[t].Vcorr;
 133         *dvdl  += f_t[t].dvdl[efpt_ind];
 134         m_add(vir,f_t[t].vir,vir);
 135     }
 136 }
 137
 138 void do_force_lowlevel(FILE       *fplog,   gmx_large_int_t step,
 139                        t_forcerec *fr,      t_inputrec *ir,
 140                        t_idef     *idef,    t_commrec  *cr,
 141                        t_nrnb     *nrnb,    gmx_wallcycle_t wcycle,
 142                        t_mdatoms  *md,
 143                        t_grpopts  *opts,
 144                        rvec       x[],      history_t  *hist,
 145                        rvec       f[],
 146                        rvec       f_longrange[],
 147                        gmx_enerdata_t *enerd,
 148                        t_fcdata   *fcd,
 149                        gmx_mtop_t     *mtop,
 150                        gmx_localtop_t *top,
 151                        gmx_genborn_t *born,
 152                        t_atomtypes *atype,
 153                        gmx_bool       bBornRadii,
 154                        matrix     box,
 155                        t_lambda   *fepvals,
 156                        real       *lambda,
 157                        t_graph    *graph,
 158                        t_blocka   *excl,
 159                        rvec       mu_tot[],
 160                        int        flags,
 161                        float      *cycles_pme)
 162 {
 163     int     i,j,status;
 164     int     donb_flags;
 165     gmx_bool    bDoEpot,bSepDVDL,bSB;
 166     int     pme_flags;
 167     matrix  boxs;
 168     rvec    box_size;
 169     real    Vsr,Vlr,Vcorr=0;
 170     t_pbc   pbc;
 171     real    dvdgb;
 172     char    buf[22];
 173     gmx_enerdata_t ed_lam;
 174     double  clam_i,vlam_i;
 175     real    dvdl_dum[efptNR], dvdl, dvdl_nb[efptNR], lam_i[efptNR];
 176     real    dvdlsum;
 177
 178 #ifdef GMX_MPI
 179     double  t0=0.0,t1,t2,t3; /* time measurement for coarse load balancing */
 180 #endif
 181
 182 #define PRINT_SEPDVDL(s,v,dvdlambda) if (bSepDVDL) fprintf(fplog,sepdvdlformat,s,v,dvdlambda);
 183
 184     GMX_MPE_LOG(ev_force_start);
 185     set_pbc(&pbc,fr->ePBC,box);
 186
 187     /* reset free energy components */
 188     for (i=0;i<efptNR;i++)
 189     {
 190         dvdl_nb[i]  = 0;
 191         dvdl_dum[i] = 0;
 192     }
 193
 194     /* Reset box */
 195     for(i=0; (i<DIM); i++)
 196     {
 197         box_size[i]=box[i][i];
 198     }
 199
 200     bSepDVDL=(fr->bSepDVDL && do_per_step(step,ir->nstlog));
 201     debug_gmx();
 202
 203     /* do QMMM first if requested */
 204     if(fr->bQMMM)
 205     {
 206         enerd->term[F_EQM] = calculate_QMMM(cr,x,f,fr,md);
 207     }
 208
 209     if (bSepDVDL)
 210     {
 211         fprintf(fplog,"Step %s: non-bonded V and dVdl for node %d:\n",
 212                 gmx_step_str(step,buf),cr->nodeid);
 213     }
 214
 215     /* Call the short range functions all in one go. */
 216     GMX_MPE_LOG(ev_do_fnbf_start);
 217
 218 #ifdef GMX_MPI
 219     /*#define TAKETIME ((cr->npmenodes) && (fr->timesteps < 12))*/
 220 #define TAKETIME FALSE
 221     if (TAKETIME)
 222     {
 223         MPI_Barrier(cr->mpi_comm_mygroup);
 224         t0=MPI_Wtime();
 225     }
 226 #endif
 227
 228     if (ir->nwall)
 229     {
 230         /* foreign lambda component for walls */
 231         dvdl = do_walls(ir,fr,box,md,x,f,lambda[efptVDW],
 232                         enerd->grpp.ener[egLJSR],nrnb);
 233         PRINT_SEPDVDL("Walls",0.0,dvdl);
 234         enerd->dvdl_lin[efptVDW] += dvdl;
 235     }
 236
 237         /* If doing GB, reset dvda and calculate the Born radii */
 238         if (ir->implicit_solvent)
 239         {
 240         wallcycle_sub_start(wcycle, ewcsNONBONDED);
 241
 242                 for(i=0;i<born->nr;i++)
 243                 {
 244                         fr->dvda[i]=0;
 245                 }
 246
 247                 if(bBornRadii)
 248                 {
 249                         calc_gb_rad(cr,fr,ir,top,atype,x,&(fr->gblist),born,md,nrnb);
 250                 }
 251
 252         wallcycle_sub_stop(wcycle, ewcsNONBONDED);
 253         }
 254
 255     where();
 256     if (flags & GMX_FORCE_NONBONDED)
 257     {
 258         donb_flags = 0;
 259         /* Add short-range interactions */
 260         donb_flags |= GMX_NONBONDED_DO_SR;
 261
 262         if (flags & GMX_FORCE_FORCES)
 263         {
 264             donb_flags |= GMX_NONBONDED_DO_FORCE;
 265         }
 266         if (flags & GMX_FORCE_ENERGY)
 267         {
 268             donb_flags |= GMX_NONBONDED_DO_POTENTIAL;
 269         }
 270         if (flags & GMX_FORCE_DO_LR)
 271         {
 272             donb_flags |= GMX_NONBONDED_DO_LR;
 273         }
 274
 275         wallcycle_sub_start(wcycle, ewcsNONBONDED);
 276         do_nonbonded(cr,fr,x,f,f_longrange,md,excl,
 277                     &enerd->grpp,box_size,nrnb,
 278                     lambda,dvdl_nb,-1,-1,donb_flags);
 279         wallcycle_sub_stop(wcycle, ewcsNONBONDED);
 280     }
 281
 282     /* If we do foreign lambda and we have soft-core interactions
 283      * we have to recalculate the (non-linear) energies contributions.
 284      */
 285     if (fepvals->n_lambda > 0 && (flags & GMX_FORCE_DHDL) && fepvals->sc_alpha != 0)
 286     {
 287         wallcycle_sub_start(wcycle, ewcsNONBONDED);
 288         init_enerdata(mtop->groups.grps[egcENER].nr,fepvals->n_lambda,&ed_lam);
 289
 290         for(i=0; i<enerd->n_lambda; i++)
 291         {
 292             for (j=0;j<efptNR;j++)
 293             {
 294                 lam_i[j] = (i==0 ? lambda[j] : fepvals->all_lambda[j][i-1]);
 295             }
 296             reset_enerdata(&ir->opts,fr,TRUE,&ed_lam,FALSE);
 297             do_nonbonded(cr,fr,x,f,f_longrange,md,excl,
 298                          &(ed_lam.grpp), box_size,nrnb,
 299                          lam_i,dvdl_dum,-1,-1,
 300                          GMX_NONBONDED_DO_FOREIGNLAMBDA | GMX_NONBONDED_DO_SR);
 301             sum_epot(&ir->opts,&ed_lam);
 302             enerd->enerpart_lambda[i] += ed_lam.term[F_EPOT];
 303         }
 304         destroy_enerdata(&ed_lam);
 305         wallcycle_sub_stop(wcycle, ewcsNONBONDED);
 306     }
 307     where();
 308
 309         /* If we are doing GB, calculate bonded forces and apply corrections
 310          * to the solvation forces */
 311     /* MRS: Eventually, many need to include free energy contribution here! */
 312         if (ir->implicit_solvent)
 313     {
 314                 calc_gb_forces(cr,md,born,top,atype,x,f,fr,idef,
 315                        ir->gb_algorithm,ir->sa_algorithm,nrnb,bBornRadii,&pbc,graph,enerd);
 316         wallcycle_sub_stop(wcycle, ewcsBONDED);
 317     }
 318
 319 #ifdef GMX_MPI
 320     if (TAKETIME)
 321     {
 322         t1=MPI_Wtime();
 323         fr->t_fnbf += t1-t0;
 324     }
 325 #endif
 326
 327     if (fepvals->sc_alpha!=0)
 328     {
 329         enerd->dvdl_nonlin[efptVDW] += dvdl_nb[efptVDW];
 330     }
 331     else
 332     {
 333         enerd->dvdl_lin[efptVDW] += dvdl_nb[efptVDW];
 334     }
 335
 336     if (fepvals->sc_alpha!=0)
 337
 338         /* even though coulomb part is linear, we already added it, beacuse we
 339            need to go through the vdw calculation anyway */
 340     {
 341         enerd->dvdl_nonlin[efptCOUL] += dvdl_nb[efptCOUL];
 342     }
 343     else
 344     {
 345         enerd->dvdl_lin[efptCOUL] += dvdl_nb[efptCOUL];
 346     }
 347
 348     Vsr = 0;
 349     if (bSepDVDL)
 350     {
 351         for(i=0; i<enerd->grpp.nener; i++)
 352         {
 353             Vsr +=
 354                 (fr->bBHAM ?
 355                  enerd->grpp.ener[egBHAMSR][i] :
 356                  enerd->grpp.ener[egLJSR][i])
 357                 + enerd->grpp.ener[egCOULSR][i] + enerd->grpp.ener[egGB][i];
 358         }
 359         dvdlsum = dvdl_nb[efptVDW] + dvdl_nb[efptCOUL];
 360         PRINT_SEPDVDL("VdW and Coulomb SR particle-p.",Vsr,dvdlsum);
 361     }
 362     debug_gmx();
 363
 364     GMX_MPE_LOG(ev_do_fnbf_finish);
 365
 366     if (debug)
 367     {
 368         pr_rvecs(debug,0,"fshift after SR",fr->fshift,SHIFTS);
 369     }
 370
 371     /* Shift the coordinates. Must be done before bonded forces and PPPM,
 372      * but is also necessary for SHAKE and update, therefore it can NOT
 373      * go when no bonded forces have to be evaluated.
 374      */
 375
 376     /* Here sometimes we would not need to shift with NBFonly,
 377      * but we do so anyhow for consistency of the returned coordinates.
 378      */
 379     if (graph)
 380     {
 381         shift_self(graph,box,x);
 382         if (TRICLINIC(box))
 383         {
 384             inc_nrnb(nrnb,eNR_SHIFTX,2*graph->nnodes);
 385         }
 386         else
 387         {
 388             inc_nrnb(nrnb,eNR_SHIFTX,graph->nnodes);
 389         }
 390     }
 391     /* Check whether we need to do bondeds or correct for exclusions */
 392     if (fr->bMolPBC &&
 393         ((flags & GMX_FORCE_BONDED)
 394          || EEL_RF(fr->eeltype) || EEL_FULL(fr->eeltype)))
 395     {
 396         /* Since all atoms are in the rectangular or triclinic unit-cell,
 397          * only single box vector shifts (2 in x) are required.
 398          */
 399         set_pbc_dd(&pbc,fr->ePBC,cr->dd,TRUE,box);
 400     }
 401     debug_gmx();
 402
 403     if (flags & GMX_FORCE_BONDED)
 404     {
 405         GMX_MPE_LOG(ev_calc_bonds_start);
 406
 407         wallcycle_sub_start(wcycle, ewcsBONDED);
 408         calc_bonds(fplog,cr->ms,
 409                    idef,x,hist,f,fr,&pbc,graph,enerd,nrnb,lambda,md,fcd,
 410                    DOMAINDECOMP(cr) ? cr->dd->gatindex : NULL, atype, born,
 411                    flags,
 412                    fr->bSepDVDL && do_per_step(step,ir->nstlog),step);
 413
 414         /* Check if we have to determine energy differences
 415          * at foreign lambda's.
 416          */
 417         if (fepvals->n_lambda > 0 && (flags & GMX_FORCE_DHDL) &&
 418             idef->ilsort != ilsortNO_FE)
 419         {
 420             if (idef->ilsort != ilsortFE_SORTED)
 421             {
 422                 gmx_incons("The bonded interactions are not sorted for free energy");
 423             }
 424             init_enerdata(mtop->groups.grps[egcENER].nr,fepvals->n_lambda,&ed_lam);
 425
 426             for(i=0; i<enerd->n_lambda; i++)
 427             {
 428                 reset_enerdata(&ir->opts,fr,TRUE,&ed_lam,FALSE);
 429                 for (j=0;j<efptNR;j++)
 430                 {
 431                     lam_i[j] = (i==0 ? lambda[j] : fepvals->all_lambda[j][i-1]);
 432                 }
 433                 calc_bonds_lambda(fplog,idef,x,fr,&pbc,graph,&ed_lam,nrnb,lam_i,md,
 434                                   fcd,DOMAINDECOMP(cr) ? cr->dd->gatindex : NULL);
 435                 sum_epot(&ir->opts,&ed_lam);
 436                 enerd->enerpart_lambda[i] += ed_lam.term[F_EPOT];
 437             }
 438             destroy_enerdata(&ed_lam);
 439         }
 440         debug_gmx();
 441         GMX_MPE_LOG(ev_calc_bonds_finish);
 442         wallcycle_sub_stop(wcycle, ewcsBONDED);
 443     }
 444
 445     where();
 446
 447     *cycles_pme = 0;
 448     if (EEL_FULL(fr->eeltype))
 449     {
 450         bSB = (ir->nwall == 2);
 451         if (bSB)
 452         {
 453             copy_mat(box,boxs);
 454             svmul(ir->wall_ewald_zfac,boxs[ZZ],boxs[ZZ]);
 455             box_size[ZZ] *= ir->wall_ewald_zfac;
 456         }
 457
 458         clear_mat(fr->vir_el_recip);
 459
 460         if (fr->bEwald)
 461         {
 462             Vcorr = 0;
 463             dvdl  = 0;
 464
 465             /* With the Verlet scheme exclusion forces are calculated
 466              * in the non-bonded kernel.
 467              */
 468             /* The TPI molecule does not have exclusions with the rest
 469              * of the system and no intra-molecular PME grid contributions
 470              * will be calculated in gmx_pme_calc_energy.
 471              */
 472             if ((ir->cutoff_scheme == ecutsGROUP && fr->n_tpi == 0) ||
 473                 ir->ewald_geometry != eewg3D ||
 474                 ir->epsilon_surface != 0)
 475             {
 476                 int nthreads,t;
 477
 478                 wallcycle_sub_start(wcycle, ewcsEWALD_CORRECTION);
 479
 480                 if (fr->n_tpi > 0)
 481                 {
 482                     gmx_fatal(FARGS,"TPI with PME currently only works in a 3D geometry with tin-foil boundary conditions");
 483                 }
 484
 485                 nthreads = gmx_omp_nthreads_get(emntBonded);
 486 #pragma omp parallel for num_threads(nthreads) schedule(static)
 487                 for(t=0; t<nthreads; t++)
 488                 {
 489                     int s,e,i;
 490                     rvec *fnv;
 491                     tensor *vir;
 492                     real *Vcorrt,*dvdlt;
 493                     if (t == 0)
 494                     {
 495                         fnv    = fr->f_novirsum;
 496                         vir    = &fr->vir_el_recip;
 497                         Vcorrt = &Vcorr;
 498                         dvdlt  = &dvdl;
 499                     }
 500                     else
 501                     {
 502                         fnv    = fr->f_t[t].f;
 503                         vir    = &fr->f_t[t].vir;
 504                         Vcorrt = &fr->f_t[t].Vcorr;
 505                         dvdlt  = &fr->f_t[t].dvdl[efptCOUL];
 506                         for(i=0; i<fr->natoms_force; i++)
 507                         {
 508                             clear_rvec(fnv[i]);
 509                         }
 510                         clear_mat(*vir);
 511                     }
 512                     *dvdlt = 0;
 513                     *Vcorrt =
 514                         ewald_LRcorrection(fplog,
 515                                            fr->excl_load[t],fr->excl_load[t+1],
 516                                            cr,t,fr,
 517                                            md->chargeA,
 518                                            md->nChargePerturbed ? md->chargeB : NULL,
 519                                            ir->cutoff_scheme != ecutsVERLET,
 520                                            excl,x,bSB ? boxs : box,mu_tot,
 521                                            ir->ewald_geometry,
 522                                            ir->epsilon_surface,
 523                                            fnv,*vir,
 524                                            lambda[efptCOUL],dvdlt);
 525                 }
 526                 if (nthreads > 1)
 527                 {
 528                     reduce_thread_forces(fr->natoms_force,fr->f_novirsum,
 529                                          fr->vir_el_recip,
 530                                          &Vcorr,efptCOUL,&dvdl,
 531                                          nthreads,fr->f_t);
 532                 }
 533
 534                 wallcycle_sub_stop(wcycle, ewcsEWALD_CORRECTION);
 535             }
 536
 537             if (fr->n_tpi == 0)
 538             {
 539                 Vcorr += ewald_charge_correction(cr,fr,lambda[efptCOUL],box,
 540                                                  &dvdl,fr->vir_el_recip);
 541             }
 542
 543             PRINT_SEPDVDL("Ewald excl./charge/dip. corr.",Vcorr,dvdl);
 544             enerd->dvdl_lin[efptCOUL] += dvdl;
 545         }
 546
 547         status = 0;
 548         Vlr  = 0;
 549         dvdl = 0;
 550         switch (fr->eeltype)
 551         {
 552         case eelPME:
 553         case eelPMESWITCH:
 554         case eelPMEUSER:
 555         case eelPMEUSERSWITCH:
 556         case eelP3M_AD:
 557             if (cr->duty & DUTY_PME)
 558             {
 559                 assert(fr->n_tpi >= 0);
 560                 if (fr->n_tpi == 0 || (flags & GMX_FORCE_STATECHANGED))
 561                 {
 562                     pme_flags = GMX_PME_SPREAD_Q | GMX_PME_SOLVE;
 563                     if (flags & GMX_FORCE_FORCES)
 564                     {
 565                         pme_flags |= GMX_PME_CALC_F;
 566                     }
 567                     if (flags & (GMX_FORCE_VIRIAL | GMX_FORCE_ENERGY))
 568                     {
 569                         pme_flags |= GMX_PME_CALC_ENER_VIR;
 570                     }
 571                     if (fr->n_tpi > 0)
 572                     {
 573                         /* We don't calculate f, but we do want the potential */
 574                         pme_flags |= GMX_PME_CALC_POT;
 575                     }
 576                     wallcycle_start(wcycle,ewcPMEMESH);
 577                     status = gmx_pme_do(fr->pmedata,
 578                                         md->start,md->homenr - fr->n_tpi,
 579                                         x,fr->f_novirsum,
 580                                         md->chargeA,md->chargeB,
 581                                         bSB ? boxs : box,cr,
 582                                         DOMAINDECOMP(cr) ? dd_pme_maxshift_x(cr->dd) : 0,
 583                                         DOMAINDECOMP(cr) ? dd_pme_maxshift_y(cr->dd) : 0,
 584                                         nrnb,wcycle,
 585                                         fr->vir_el_recip,fr->ewaldcoeff,
 586                                         &Vlr,lambda[efptCOUL],&dvdl,
 587                                         pme_flags);
 588                     *cycles_pme = wallcycle_stop(wcycle,ewcPMEMESH);
 589
 590                     /* We should try to do as little computation after
 591                      * this as possible, because parallel PME synchronizes
 592                      * the nodes, so we want all load imbalance of the rest
 593                      * of the force calculation to be before the PME call.
 594                      * DD load balancing is done on the whole time of
 595                      * the force call (without PME).
 596                      */
 597                 }
 598                 if (fr->n_tpi > 0)
 599                 {
 600                     /* Determine the PME grid energy of the test molecule
 601                      * with the PME grid potential of the other charges.
 602                      */
 603                     gmx_pme_calc_energy(fr->pmedata,fr->n_tpi,
 604                                         x + md->homenr - fr->n_tpi,
 605                                         md->chargeA + md->homenr - fr->n_tpi,
 606                                         &Vlr);
 607                 }
 608                 PRINT_SEPDVDL("PME mesh",Vlr,dvdl);
 609             }
 610             break;
 611         case eelEWALD:
 612             Vlr = do_ewald(fplog,FALSE,ir,x,fr->f_novirsum,
 613                            md->chargeA,md->chargeB,
 614                            box_size,cr,md->homenr,
 615                            fr->vir_el_recip,fr->ewaldcoeff,
 616                            lambda[efptCOUL],&dvdl,fr->ewald_table);
 617             PRINT_SEPDVDL("Ewald long-range",Vlr,dvdl);
 618             break;
 619         default:
 620             gmx_fatal(FARGS,"No such electrostatics method implemented %s",
 621                       eel_names[fr->eeltype]);
 622         }
 623         if (status != 0)
 624         {
 625             gmx_fatal(FARGS,"Error %d in long range electrostatics routine %s",
 626                       status,EELTYPE(fr->eeltype));
 627                 }
 628         /* Note that with separate PME nodes we get the real energies later */
 629         enerd->dvdl_lin[efptCOUL] += dvdl;
 630         enerd->term[F_COUL_RECIP] = Vlr + Vcorr;
 631         if (debug)
 632         {
 633             fprintf(debug,"Vlr = %g, Vcorr = %g, Vlr_corr = %g\n",
 634                     Vlr,Vcorr,enerd->term[F_COUL_RECIP]);
 635             pr_rvecs(debug,0,"vir_el_recip after corr",fr->vir_el_recip,DIM);
 636             pr_rvecs(debug,0,"fshift after LR Corrections",fr->fshift,SHIFTS);
 637         }
 638     }
 639     else
 640     {
 641         if (EEL_RF(fr->eeltype))
 642         {
 643             /* With the Verlet scheme exclusion forces are calculated
 644              * in the non-bonded kernel.
 645              */
 646             if (ir->cutoff_scheme != ecutsVERLET && fr->eeltype != eelRF_NEC)
 647             {
 648                 dvdl = 0;
 649                 enerd->term[F_RF_EXCL] =
 650                     RF_excl_correction(fplog,fr,graph,md,excl,x,f,
 651                                        fr->fshift,&pbc,lambda[efptCOUL],&dvdl);
 652             }
 653
 654             enerd->dvdl_lin[efptCOUL] += dvdl;
 655             PRINT_SEPDVDL("RF exclusion correction",
 656                           enerd->term[F_RF_EXCL],dvdl);
 657         }
 658     }
 659     where();
 660     debug_gmx();
 661
 662     if (debug)
 663     {
 664         print_nrnb(debug,nrnb);
 665     }
 666     debug_gmx();
 667
 668 #ifdef GMX_MPI
 669     if (TAKETIME)
 670     {
 671         t2=MPI_Wtime();
 672         MPI_Barrier(cr->mpi_comm_mygroup);
 673         t3=MPI_Wtime();
 674         fr->t_wait += t3-t2;
 675         if (fr->timesteps == 11)
 676         {
 677             fprintf(stderr,"* PP load balancing info: node %d, step %s, rel wait time=%3.0f%% , load string value: %7.2f\n",
 678                     cr->nodeid, gmx_step_str(fr->timesteps,buf),
 679                     100*fr->t_wait/(fr->t_wait+fr->t_fnbf),
 680                     (fr->t_fnbf+fr->t_wait)/fr->t_fnbf);
 681         }
 682         fr->timesteps++;
 683     }
 684 #endif
 685
 686     if (debug)
 687     {
 688         pr_rvecs(debug,0,"fshift after bondeds",fr->fshift,SHIFTS);
 689     }
 690
 691     GMX_MPE_LOG(ev_force_finish);
 692
 693 }
 694
 695 void init_enerdata(int ngener,int n_lambda,gmx_enerdata_t *enerd)
 696 {
 697     int i,n2;
 698
 699     for(i=0; i<F_NRE; i++)
 700     {
 701         enerd->term[i] = 0;
 702     }
 703
 704
 705     for(i=0; i<efptNR; i++) {
 706         enerd->dvdl_lin[i]  = 0;
 707         enerd->dvdl_nonlin[i]  = 0;
 708     }
 709
 710     n2=ngener*ngener;
 711     if (debug)
 712     {
 713         fprintf(debug,"Creating %d sized group matrix for energies\n",n2);
 714     }
 715     enerd->grpp.nener = n2;
 716     for(i=0; (i<egNR); i++)
 717     {
 718         snew(enerd->grpp.ener[i],n2);
 719     }
 720
 721     if (n_lambda)
 722     {
 723         enerd->n_lambda = 1 + n_lambda;
 724         snew(enerd->enerpart_lambda,enerd->n_lambda);
 725     }
 726     else
 727     {
 728         enerd->n_lambda = 0;
 729     }
 730 }
 731
 732 void destroy_enerdata(gmx_enerdata_t *enerd)
 733 {
 734     int i;
 735
 736     for(i=0; (i<egNR); i++)
 737     {
 738         sfree(enerd->grpp.ener[i]);
 739     }
 740
 741     if (enerd->n_lambda)
 742     {
 743         sfree(enerd->enerpart_lambda);
 744     }
 745 }
 746
 747 static real sum_v(int n,real v[])
 748 {
 749   real t;
 750   int  i;
 751
 752   t = 0.0;
 753   for(i=0; (i<n); i++)
 754     t = t + v[i];
 755
 756   return t;
 757 }
 758
 759 void sum_epot(t_grpopts *opts,gmx_enerdata_t *enerd)
 760 {
 761   gmx_grppairener_t *grpp;
 762   real *epot;
 763   int i;
 764
 765   grpp = &enerd->grpp;
 766   epot = enerd->term;
 767
 768   /* Accumulate energies */
 769   epot[F_COUL_SR]  = sum_v(grpp->nener,grpp->ener[egCOULSR]);
 770   epot[F_LJ]       = sum_v(grpp->nener,grpp->ener[egLJSR]);
 771   epot[F_LJ14]     = sum_v(grpp->nener,grpp->ener[egLJ14]);
 772   epot[F_COUL14]   = sum_v(grpp->nener,grpp->ener[egCOUL14]);
 773   epot[F_COUL_LR]  = sum_v(grpp->nener,grpp->ener[egCOULLR]);
 774   epot[F_LJ_LR]    = sum_v(grpp->nener,grpp->ener[egLJLR]);
 775   /* We have already added 1-2,1-3, and 1-4 terms to F_GBPOL */
 776   epot[F_GBPOL]   += sum_v(grpp->nener,grpp->ener[egGB]);
 777
 778 /* lattice part of LR doesnt belong to any group
 779  * and has been added earlier
 780  */
 781   epot[F_BHAM]     = sum_v(grpp->nener,grpp->ener[egBHAMSR]);
 782   epot[F_BHAM_LR]  = sum_v(grpp->nener,grpp->ener[egBHAMLR]);
 783
 784   epot[F_EPOT] = 0;
 785   for(i=0; (i<F_EPOT); i++)
 786   {
 787       if (i != F_DISRESVIOL && i != F_ORIRESDEV)
 788       {
 789           epot[F_EPOT] += epot[i];
 790       }
 791   }
 792 }
 793
 794 void sum_dhdl(gmx_enerdata_t *enerd, real *lambda, t_lambda *fepvals)
 795 {
 796     int i,j,index;
 797     double dlam;
 798
 799     enerd->dvdl_lin[efptVDW] += enerd->term[F_DVDL_VDW];  /* include dispersion correction */
 800     enerd->term[F_DVDL] = 0.0;
 801     for (i=0;i<efptNR;i++)
 802     {
 803         if (fepvals->separate_dvdl[i])
 804         {
 805             /* could this be done more readably/compactly? */
 806             switch (i) {
 807             case (efptCOUL):
 808                 index = F_DVDL_COUL;
 809                 break;
 810             case (efptVDW):
 811                 index = F_DVDL_VDW;
 812                 break;
 813             case (efptBONDED):
 814                 index = F_DVDL_BONDED;
 815                 break;
 816             case (efptRESTRAINT):
 817                 index = F_DVDL_RESTRAINT;
 818                 break;
 819             case (efptMASS):
 820                 index = F_DKDL;
 821                 break;
 822             default:
 823                 index = F_DVDL;
 824                 break;
 825             }
 826             enerd->term[index] = enerd->dvdl_lin[i] + enerd->dvdl_nonlin[i];
 827             if (debug)
 828             {
 829                 fprintf(debug,"dvdl-%s[%2d]: %f: non-linear %f + linear %f\n",
 830                         efpt_names[i],i,enerd->term[index],enerd->dvdl_nonlin[i],enerd->dvdl_lin[i]);
 831             }
 832         }
 833         else
 834         {
 835             enerd->term[F_DVDL] += enerd->dvdl_lin[i] + enerd->dvdl_nonlin[i];
 836             if (debug)
 837             {
 838                 fprintf(debug,"dvd-%sl[%2d]: %f: non-linear %f + linear %f\n",
 839                         efpt_names[0],i,enerd->term[F_DVDL],enerd->dvdl_nonlin[i],enerd->dvdl_lin[i]);
 840             }
 841         }
 842     }
 843
 844     /* Notes on the foreign lambda free energy difference evaluation:
 845      * Adding the potential and ekin terms that depend linearly on lambda
 846      * as delta lam * dvdl to the energy differences is exact.
 847      * For the constraints this is not exact, but we have no other option
 848      * without literally changing the lengths and reevaluating the energies at each step.
 849      * (try to remedy this post 4.6 - MRS)
 850      * For the non-bonded LR term we assume that the soft-core (if present)
 851      * no longer affects the energy beyond the short-range cut-off,
 852      * which is a very good approximation (except for exotic settings).
 853      * (investigate how to overcome this post 4.6 - MRS)
 854      */
 855
 856     for(i=0; i<fepvals->n_lambda; i++)
 857     {                                         /* note we are iterating over fepvals here!
 858                                                  For the current lam, dlam = 0 automatically,
 859                                                  so we don't need to add anything to the
 860                                                  enerd->enerpart_lambda[0] */
 861
 862         /* we don't need to worry about dvdl contributions to the current lambda, because
 863            it's automatically zero */
 864
 865         /* first kinetic energy term */
 866         dlam = (fepvals->all_lambda[efptMASS][i] - lambda[efptMASS]);
 867
 868         enerd->enerpart_lambda[i+1] += enerd->term[F_DKDL]*dlam;
 869
 870         for (j=0;j<efptNR;j++)
 871         {
 872             if (j==efptMASS) {continue;} /* no other mass term to worry about */
 873
 874             dlam = (fepvals->all_lambda[j][i]-lambda[j]);
 875             enerd->enerpart_lambda[i+1] += dlam*enerd->dvdl_lin[j];
 876             if (debug)
 877             {
 878                 fprintf(debug,"enerdiff lam %g: (%15s), non-linear %f linear %f*%f\n",
 879                         fepvals->all_lambda[j][i],efpt_names[j],
 880                         (enerd->enerpart_lambda[i+1] - enerd->enerpart_lambda[0]),
 881                         dlam,enerd->dvdl_lin[j]);
 882             }
 883         }
 884     }
 885 }
 886
 887 void reset_enerdata(t_grpopts *opts,
 888                     t_forcerec *fr,gmx_bool bNS,
 889                     gmx_enerdata_t *enerd,
 890                     gmx_bool bMaster)
 891 {
 892     gmx_bool bKeepLR;
 893     int  i,j;
 894
 895     /* First reset all energy components, except for the long range terms
 896      * on the master at non neighbor search steps, since the long range
 897      * terms have already been summed at the last neighbor search step.
 898      */
 899     bKeepLR = (fr->bTwinRange && !bNS);
 900     for(i=0; (i<egNR); i++) {
 901         if (!(bKeepLR && bMaster && (i == egCOULLR || i == egLJLR))) {
 902             for(j=0; (j<enerd->grpp.nener); j++)
 903                 enerd->grpp.ener[i][j] = 0.0;
 904         }
 905     }
 906     for (i=0;i<efptNR;i++)
 907     {
 908         enerd->dvdl_lin[i]    = 0.0;
 909         enerd->dvdl_nonlin[i] = 0.0;
 910     }
 911
 912     /* Normal potential energy components */
 913     for(i=0; (i<=F_EPOT); i++) {
 914         enerd->term[i] = 0.0;
 915     }
 916     /* Initialize the dVdlambda term with the long range contribution */
 917     /* Initialize the dvdl term with the long range contribution */
 918     enerd->term[F_DVDL]            = 0.0;
 919     enerd->term[F_DVDL_COUL]       = 0.0;
 920     enerd->term[F_DVDL_VDW]        = 0.0;
 921     enerd->term[F_DVDL_BONDED]     = 0.0;
 922     enerd->term[F_DVDL_RESTRAINT]  = 0.0;
 923     enerd->term[F_DKDL]            = 0.0;
 924     if (enerd->n_lambda > 0)
 925     {
 926         for(i=0; i<enerd->n_lambda; i++)
 927         {
 928             enerd->enerpart_lambda[i] = 0.0;
 929         }
 930     }
 931 }