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 by the GROMACS development team.
   7  * Copyright (c) 2018,2019,2020, by the GROMACS development team, led by
   8  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   9  * and including many others, as listed in the AUTHORS file in the
  10  * top-level source directory and at http://www.gromacs.org.
  11  *
  12  * GROMACS is free software; you can redistribute it and/or
  13  * modify it under the terms of the GNU Lesser General Public License
  14  * as published by the Free Software Foundation; either version 2.1
  15  * of the License, or (at your option) any later version.
  16  *
  17  * GROMACS is distributed in the hope that it will be useful,
  18  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  19  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  20  * Lesser General Public License for more details.
  21  *
  22  * You should have received a copy of the GNU Lesser General Public
  23  * License along with GROMACS; if not, see
  24  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  25  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  26  *
  27  * If you want to redistribute modifications to GROMACS, please
  28  * consider that scientific software is very special. Version
  29  * control is crucial - bugs must be traceable. We will be happy to
  30  * consider code for inclusion in the official distribution, but
  31  * derived work must not be called official GROMACS. Details are found
  32  * in the README & COPYING files - if they are missing, get the
  33  * official version at http://www.gromacs.org.
  34  *
  35  * To help us fund GROMACS development, we humbly ask that you cite
  36  * the research papers on the package. Check out http://www.gromacs.org.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "force.h"
  41
  42 #include <cassert>
  43 #include <cmath>
  44 #include <cstring>
  45
  46 #include "gromacs/domdec/dlbtiming.h"
  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/math/vec.h"
  55 #include "gromacs/math/vecdump.h"
  56 #include "gromacs/mdlib/forcerec_threading.h"
  57 #include "gromacs/mdtypes/commrec.h"
  58 #include "gromacs/mdtypes/enerdata.h"
  59 #include "gromacs/mdtypes/forceoutput.h"
  60 #include "gromacs/mdtypes/forcerec.h"
  61 #include "gromacs/mdtypes/inputrec.h"
  62 #include "gromacs/mdtypes/interaction_const.h"
  63 #include "gromacs/mdtypes/md_enums.h"
  64 #include "gromacs/mdtypes/mdatom.h"
  65 #include "gromacs/mdtypes/simulation_workload.h"
  66 #include "gromacs/pbcutil/ishift.h"
  67 #include "gromacs/pbcutil/pbc.h"
  68 #include "gromacs/timing/wallcycle.h"
  69 #include "gromacs/utility/exceptions.h"
  70 #include "gromacs/utility/fatalerror.h"
  71 #include "gromacs/utility/smalloc.h"
  72
  73 using gmx::ArrayRef;
  74 using gmx::RVec;
  75
  76 static void clearEwaldThreadOutput(ewald_corr_thread_t* ewc_t)
  77 {
  78     ewc_t->Vcorr_q        = 0;
  79     ewc_t->Vcorr_lj       = 0;
  80     ewc_t->dvdl[efptCOUL] = 0;
  81     ewc_t->dvdl[efptVDW]  = 0;
  82     clear_mat(ewc_t->vir_q);
  83     clear_mat(ewc_t->vir_lj);
  84 }
  85
  86 static void reduceEwaldThreadOuput(int nthreads, ewald_corr_thread_t* ewc_t)
  87 {
  88     ewald_corr_thread_t& dest = ewc_t[0];
  89
  90     for (int t = 1; t < nthreads; t++)
  91     {
  92         dest.Vcorr_q += ewc_t[t].Vcorr_q;
  93         dest.Vcorr_lj += ewc_t[t].Vcorr_lj;
  94         dest.dvdl[efptCOUL] += ewc_t[t].dvdl[efptCOUL];
  95         dest.dvdl[efptVDW] += ewc_t[t].dvdl[efptVDW];
  96         m_add(dest.vir_q, ewc_t[t].vir_q, dest.vir_q);
  97         m_add(dest.vir_lj, ewc_t[t].vir_lj, dest.vir_lj);
  98     }
  99 }
 100
 101 void calculateLongRangeNonbondeds(t_forcerec*                   fr,
 102                                   const t_inputrec*             ir,
 103                                   const t_commrec*              cr,
 104                                   t_nrnb*                       nrnb,
 105                                   gmx_wallcycle_t               wcycle,
 106                                   const t_mdatoms*              md,
 107                                   gmx::ArrayRef<const RVec>     coordinates,
 108                                   gmx::ForceWithVirial*         forceWithVirial,
 109                                   gmx_enerdata_t*               enerd,
 110                                   const matrix                  box,
 111                                   const real*                   lambda,
 112                                   const rvec*                   mu_tot,
 113                                   const gmx::StepWorkload&      stepWork,
 114                                   const DDBalanceRegionHandler& ddBalanceRegionHandler)
 115 {
 116     const bool computePmeOnCpu = (EEL_PME(fr->ic->eeltype) || EVDW_PME(fr->ic->vdwtype))
 117                                  && thisRankHasDuty(cr, DUTY_PME)
 118                                  && (pme_run_mode(fr->pmedata) == PmeRunMode::CPU);
 119
 120     const bool haveEwaldSurfaceTerm = haveEwaldSurfaceContribution(*ir);
 121
 122     /* Do long-range electrostatics and/or LJ-PME
 123      * and compute PME surface terms when necessary.
 124      */
 125     if ((computePmeOnCpu || fr->ic->eeltype == eelEWALD || haveEwaldSurfaceTerm)
 126         && stepWork.computeNonbondedForces)
 127     {
 128         int  status = 0;
 129         real Vlr_q = 0, Vlr_lj = 0;
 130
 131         /* We reduce all virial, dV/dlambda and energy contributions, except
 132          * for the reciprocal energies (Vlr_q, Vlr_lj) into the same struct.
 133          */
 134         ewald_corr_thread_t& ewaldOutput = fr->ewc_t[0];
 135         clearEwaldThreadOutput(&ewaldOutput);
 136
 137         if (EEL_PME_EWALD(fr->ic->eeltype) || EVDW_PME(fr->ic->vdwtype))
 138         {
 139             /* Calculate the Ewald surface force and energy contributions, when necessary */
 140             if (haveEwaldSurfaceTerm)
 141             {
 142                 wallcycle_sub_start(wcycle, ewcsEWALD_CORRECTION);
 143
 144                 int nthreads = fr->nthread_ewc;
 145 #pragma omp parallel for num_threads(nthreads) schedule(static)
 146                 for (int t = 0; t < nthreads; t++)
 147                 {
 148                     try
 149                     {
 150                         ewald_corr_thread_t& ewc_t = fr->ewc_t[t];
 151                         if (t > 0)
 152                         {
 153                             clearEwaldThreadOutput(&ewc_t);
 154                         }
 155
 156                         /* Threading is only supported with the Verlet cut-off
 157                          * scheme and then only single particle forces (no
 158                          * exclusion forces) are calculated, so we can store
 159                          * the forces in the normal, single forceWithVirial->force_ array.
 160                          */
 161                         const rvec* x = as_rvec_array(coordinates.data());
 162                         ewald_LRcorrection(md->homenr, cr, nthreads, t, *fr, *ir, md->chargeA,
 163                                            md->chargeB, (md->nChargePerturbed != 0), x, box, mu_tot,
 164                                            as_rvec_array(forceWithVirial->force_.data()),
 165                                            &ewc_t.Vcorr_q, lambda[efptCOUL], &ewc_t.dvdl[efptCOUL]);
 166                     }
 167                     GMX_CATCH_ALL_AND_EXIT_WITH_FATAL_ERROR
 168                 }
 169                 if (nthreads > 1)
 170                 {
 171                     reduceEwaldThreadOuput(nthreads, fr->ewc_t);
 172                 }
 173                 wallcycle_sub_stop(wcycle, ewcsEWALD_CORRECTION);
 174             }
 175
 176             if (EEL_PME_EWALD(fr->ic->eeltype) && fr->n_tpi == 0)
 177             {
 178                 /* This is not in a subcounter because it takes a
 179                    negligible and constant-sized amount of time */
 180                 ewaldOutput.Vcorr_q += ewald_charge_correction(
 181                         cr, fr, lambda[efptCOUL], box, &ewaldOutput.dvdl[efptCOUL], ewaldOutput.vir_q);
 182             }
 183
 184             if (computePmeOnCpu)
 185             {
 186                 /* Do reciprocal PME for Coulomb and/or LJ. */
 187                 assert(fr->n_tpi >= 0);
 188                 if (fr->n_tpi == 0 || stepWork.stateChanged)
 189                 {
 190                     /* With domain decomposition we close the CPU side load
 191                      * balancing region here, because PME does global
 192                      * communication that acts as a global barrier.
 193                      */
 194                     ddBalanceRegionHandler.closeAfterForceComputationCpu();
 195
 196                     wallcycle_start(wcycle, ewcPMEMESH);
 197                     status = gmx_pme_do(
 198                             fr->pmedata,
 199                             gmx::constArrayRefFromArray(coordinates.data(), md->homenr - fr->n_tpi),
 200                             forceWithVirial->force_, md->chargeA, md->chargeB, md->sqrt_c6A,
 201                             md->sqrt_c6B, md->sigmaA, md->sigmaB, box, cr,
 202                             DOMAINDECOMP(cr) ? dd_pme_maxshift_x(cr->dd) : 0,
 203                             DOMAINDECOMP(cr) ? dd_pme_maxshift_y(cr->dd) : 0, nrnb, wcycle,
 204                             ewaldOutput.vir_q, ewaldOutput.vir_lj, &Vlr_q, &Vlr_lj,
 205                             lambda[efptCOUL], lambda[efptVDW], &ewaldOutput.dvdl[efptCOUL],
 206                             &ewaldOutput.dvdl[efptVDW], stepWork);
 207                     wallcycle_stop(wcycle, ewcPMEMESH);
 208                     if (status != 0)
 209                     {
 210                         gmx_fatal(FARGS, "Error %d in reciprocal PME routine", status);
 211                     }
 212
 213                     /* We should try to do as little computation after
 214                      * this as possible, because parallel PME synchronizes
 215                      * the nodes, so we want all load imbalance of the
 216                      * rest of the force calculation to be before the PME
 217                      * call.  DD load balancing is done on the whole time
 218                      * of the force call (without PME).
 219                      */
 220                 }
 221                 if (fr->n_tpi > 0)
 222                 {
 223                     /* Determine the PME grid energy of the test molecule
 224                      * with the PME grid potential of the other charges.
 225                      */
 226                     gmx_pme_calc_energy(
 227                             fr->pmedata, coordinates.subArray(md->homenr - fr->n_tpi, fr->n_tpi),
 228                             gmx::arrayRefFromArray(md->chargeA + md->homenr - fr->n_tpi, fr->n_tpi),
 229                             &Vlr_q);
 230                 }
 231             }
 232         }
 233
 234         if (fr->ic->eeltype == eelEWALD)
 235         {
 236             const rvec* x = as_rvec_array(coordinates.data());
 237             Vlr_q = do_ewald(ir, x, as_rvec_array(forceWithVirial->force_.data()), md->chargeA,
 238                              md->chargeB, box, cr, md->homenr, ewaldOutput.vir_q, fr->ic->ewaldcoeff_q,
 239                              lambda[efptCOUL], &ewaldOutput.dvdl[efptCOUL], fr->ewald_table);
 240         }
 241
 242         /* Note that with separate PME nodes we get the real energies later */
 243         // TODO it would be simpler if we just accumulated a single
 244         // long-range virial contribution.
 245         forceWithVirial->addVirialContribution(ewaldOutput.vir_q);
 246         forceWithVirial->addVirialContribution(ewaldOutput.vir_lj);
 247         enerd->dvdl_lin[efptCOUL] += ewaldOutput.dvdl[efptCOUL];
 248         enerd->dvdl_lin[efptVDW] += ewaldOutput.dvdl[efptVDW];
 249         enerd->term[F_COUL_RECIP] = Vlr_q + ewaldOutput.Vcorr_q;
 250         enerd->term[F_LJ_RECIP]   = Vlr_lj + ewaldOutput.Vcorr_lj;
 251
 252         if (debug)
 253         {
 254             fprintf(debug, "Vlr_q = %g, Vcorr_q = %g, Vlr_corr_q = %g\n", Vlr_q,
 255                     ewaldOutput.Vcorr_q, enerd->term[F_COUL_RECIP]);
 256             pr_rvecs(debug, 0, "vir_el_recip after corr", ewaldOutput.vir_q, DIM);
 257             fprintf(debug, "Vlr_lj: %g, Vcorr_lj = %g, Vlr_corr_lj = %g\n", Vlr_lj,
 258                     ewaldOutput.Vcorr_lj, enerd->term[F_LJ_RECIP]);
 259             pr_rvecs(debug, 0, "vir_lj_recip after corr", ewaldOutput.vir_lj, DIM);
 260         }
 261     }
 262
 263     if (debug)
 264     {
 265         print_nrnb(debug, nrnb);
 266     }
 267 }