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/listed_forces/listed_forces.h"
  55 #include "gromacs/math/vec.h"
  56 #include "gromacs/math/vecdump.h"
  57 #include "gromacs/mdlib/forcerec_threading.h"
  58 #include "gromacs/mdlib/qmmm.h"
  59 #include "gromacs/mdlib/rf_util.h"
  60 #include "gromacs/mdlib/wall.h"
  61 #include "gromacs/mdtypes/commrec.h"
  62 #include "gromacs/mdtypes/enerdata.h"
  63 #include "gromacs/mdtypes/forceoutput.h"
  64 #include "gromacs/mdtypes/forcerec.h"
  65 #include "gromacs/mdtypes/inputrec.h"
  66 #include "gromacs/mdtypes/md_enums.h"
  67 #include "gromacs/mdtypes/mdatom.h"
  68 #include "gromacs/mdtypes/simulation_workload.h"
  69 #include "gromacs/pbcutil/ishift.h"
  70 #include "gromacs/pbcutil/mshift.h"
  71 #include "gromacs/pbcutil/pbc.h"
  72 #include "gromacs/timing/wallcycle.h"
  73 #include "gromacs/utility/exceptions.h"
  74 #include "gromacs/utility/fatalerror.h"
  75 #include "gromacs/utility/smalloc.h"
  76
  77 static void clearEwaldThreadOutput(ewald_corr_thread_t* ewc_t)
  78 {
  79     ewc_t->Vcorr_q        = 0;
  80     ewc_t->Vcorr_lj       = 0;
  81     ewc_t->dvdl[efptCOUL] = 0;
  82     ewc_t->dvdl[efptVDW]  = 0;
  83     clear_mat(ewc_t->vir_q);
  84     clear_mat(ewc_t->vir_lj);
  85 }
  86
  87 static void reduceEwaldThreadOuput(int nthreads, ewald_corr_thread_t* ewc_t)
  88 {
  89     ewald_corr_thread_t& dest = ewc_t[0];
  90
  91     for (int t = 1; t < nthreads; t++)
  92     {
  93         dest.Vcorr_q += ewc_t[t].Vcorr_q;
  94         dest.Vcorr_lj += ewc_t[t].Vcorr_lj;
  95         dest.dvdl[efptCOUL] += ewc_t[t].dvdl[efptCOUL];
  96         dest.dvdl[efptVDW] += ewc_t[t].dvdl[efptVDW];
  97         m_add(dest.vir_q, ewc_t[t].vir_q, dest.vir_q);
  98         m_add(dest.vir_lj, ewc_t[t].vir_lj, dest.vir_lj);
  99     }
 100 }
 101
 102 void do_force_lowlevel(t_forcerec*                         fr,
 103                        const t_inputrec*                   ir,
 104                        const t_idef*                       idef,
 105                        const t_commrec*                    cr,
 106                        const gmx_multisim_t*               ms,
 107                        t_nrnb*                             nrnb,
 108                        gmx_wallcycle_t                     wcycle,
 109                        const t_mdatoms*                    md,
 110                        gmx::ArrayRefWithPadding<gmx::RVec> coordinates,
 111                        history_t*                          hist,
 112                        gmx::ForceOutputs*                  forceOutputs,
 113                        gmx_enerdata_t*                     enerd,
 114                        t_fcdata*                           fcd,
 115                        const matrix                        box,
 116                        const real*                         lambda,
 117                        const t_graph*                      graph,
 118                        const rvec*                         mu_tot,
 119                        const gmx::StepWorkload&            stepWork,
 120                        const DDBalanceRegionHandler&       ddBalanceRegionHandler)
 121 {
 122     // TODO: Replace all uses of x by const coordinates
 123     rvec* x = as_rvec_array(coordinates.paddedArrayRef().data());
 124
 125     auto& forceWithVirial = forceOutputs->forceWithVirial();
 126
 127     /* do QMMM first if requested */
 128     if (fr->bQMMM)
 129     {
 130         enerd->term[F_EQM] = calculate_QMMM(cr, &forceOutputs->forceWithShiftForces(), fr->qr);
 131     }
 132
 133     /* Call the short range functions all in one go. */
 134
 135     if (ir->nwall)
 136     {
 137         /* foreign lambda component for walls */
 138         real dvdl_walls = do_walls(*ir, *fr, box, *md, x, &forceWithVirial, lambda[efptVDW],
 139                                    enerd->grpp.ener[egLJSR].data(), nrnb);
 140         enerd->dvdl_lin[efptVDW] += dvdl_walls;
 141     }
 142
 143     /* Shift the coordinates. Must be done before listed forces and PPPM,
 144      * but is also necessary for SHAKE and update, therefore it can NOT
 145      * go when no listed forces have to be evaluated.
 146      *
 147      * The shifting and PBC code is deliberately not timed, since with
 148      * the Verlet scheme it only takes non-zero time with triclinic
 149      * boxes, and even then the time is around a factor of 100 less
 150      * than the next smallest counter.
 151      */
 152
 153
 154     /* Here sometimes we would not need to shift with NBFonly,
 155      * but we do so anyhow for consistency of the returned coordinates.
 156      */
 157     if (graph)
 158     {
 159         shift_self(graph, box, x);
 160         if (TRICLINIC(box))
 161         {
 162             inc_nrnb(nrnb, eNR_SHIFTX, 2 * graph->nnodes);
 163         }
 164         else
 165         {
 166             inc_nrnb(nrnb, eNR_SHIFTX, graph->nnodes);
 167         }
 168     }
 169
 170     {
 171         t_pbc pbc;
 172
 173         /* Check whether we need to take into account PBC in listed interactions. */
 174         const auto needPbcForListedForces =
 175                 fr->bMolPBC && stepWork.computeListedForces && haveCpuListedForces(*fr, *idef, *fcd);
 176         if (needPbcForListedForces)
 177         {
 178             /* Since all atoms are in the rectangular or triclinic unit-cell,
 179              * only single box vector shifts (2 in x) are required.
 180              */
 181             set_pbc_dd(&pbc, fr->pbcType, DOMAINDECOMP(cr) ? cr->dd->numCells : nullptr, TRUE, box);
 182         }
 183
 184         do_force_listed(wcycle, box, ir->fepvals, cr, ms, idef, x, hist, forceOutputs, fr, &pbc,
 185                         graph, enerd, nrnb, lambda, md, fcd,
 186                         DOMAINDECOMP(cr) ? cr->dd->globalAtomIndices.data() : nullptr, stepWork);
 187     }
 188
 189     const bool computePmeOnCpu = (EEL_PME(fr->ic->eeltype) || EVDW_PME(fr->ic->vdwtype))
 190                                  && thisRankHasDuty(cr, DUTY_PME)
 191                                  && (pme_run_mode(fr->pmedata) == PmeRunMode::CPU);
 192
 193     const bool haveEwaldSurfaceTerm = haveEwaldSurfaceContribution(*ir);
 194
 195     /* Do long-range electrostatics and/or LJ-PME
 196      * and compute PME surface terms when necessary.
 197      */
 198     if (computePmeOnCpu || fr->ic->eeltype == eelEWALD || haveEwaldSurfaceTerm)
 199     {
 200         int  status = 0;
 201         real Vlr_q = 0, Vlr_lj = 0;
 202
 203         /* We reduce all virial, dV/dlambda and energy contributions, except
 204          * for the reciprocal energies (Vlr_q, Vlr_lj) into the same struct.
 205          */
 206         ewald_corr_thread_t& ewaldOutput = fr->ewc_t[0];
 207         clearEwaldThreadOutput(&ewaldOutput);
 208
 209         if (EEL_PME_EWALD(fr->ic->eeltype) || EVDW_PME(fr->ic->vdwtype))
 210         {
 211             /* Calculate the Ewald surface force and energy contributions, when necessary */
 212             if (haveEwaldSurfaceTerm)
 213             {
 214                 wallcycle_sub_start(wcycle, ewcsEWALD_CORRECTION);
 215
 216                 if (fr->n_tpi > 0)
 217                 {
 218                     gmx_fatal(FARGS,
 219                               "TPI with PME currently only works in a 3D geometry with tin-foil "
 220                               "boundary conditions");
 221                 }
 222
 223                 int nthreads = fr->nthread_ewc;
 224 #pragma omp parallel for num_threads(nthreads) schedule(static)
 225                 for (int t = 0; t < nthreads; t++)
 226                 {
 227                     try
 228                     {
 229                         ewald_corr_thread_t& ewc_t = fr->ewc_t[t];
 230                         if (t > 0)
 231                         {
 232                             clearEwaldThreadOutput(&ewc_t);
 233                         }
 234
 235                         /* Threading is only supported with the Verlet cut-off
 236                          * scheme and then only single particle forces (no
 237                          * exclusion forces) are calculated, so we can store
 238                          * the forces in the normal, single forceWithVirial->force_ array.
 239                          */
 240                         ewald_LRcorrection(md->homenr, cr, nthreads, t, *fr, *ir, md->chargeA,
 241                                            md->chargeB, (md->nChargePerturbed != 0), x, box, mu_tot,
 242                                            as_rvec_array(forceWithVirial.force_.data()),
 243                                            &ewc_t.Vcorr_q, lambda[efptCOUL], &ewc_t.dvdl[efptCOUL]);
 244                     }
 245                     GMX_CATCH_ALL_AND_EXIT_WITH_FATAL_ERROR
 246                 }
 247                 if (nthreads > 1)
 248                 {
 249                     reduceEwaldThreadOuput(nthreads, fr->ewc_t);
 250                 }
 251                 wallcycle_sub_stop(wcycle, ewcsEWALD_CORRECTION);
 252             }
 253
 254             if (EEL_PME_EWALD(fr->ic->eeltype) && fr->n_tpi == 0)
 255             {
 256                 /* This is not in a subcounter because it takes a
 257                    negligible and constant-sized amount of time */
 258                 ewaldOutput.Vcorr_q += ewald_charge_correction(
 259                         cr, fr, lambda[efptCOUL], box, &ewaldOutput.dvdl[efptCOUL], ewaldOutput.vir_q);
 260             }
 261
 262             if (computePmeOnCpu)
 263             {
 264                 /* Do reciprocal PME for Coulomb and/or LJ. */
 265                 assert(fr->n_tpi >= 0);
 266                 if (fr->n_tpi == 0 || stepWork.stateChanged)
 267                 {
 268                     int pme_flags = GMX_PME_SPREAD | GMX_PME_SOLVE;
 269
 270                     if (stepWork.computeForces)
 271                     {
 272                         pme_flags |= GMX_PME_CALC_F;
 273                     }
 274                     if (stepWork.computeVirial)
 275                     {
 276                         pme_flags |= GMX_PME_CALC_ENER_VIR;
 277                     }
 278                     if (fr->n_tpi > 0)
 279                     {
 280                         /* We don't calculate f, but we do want the potential */
 281                         pme_flags |= GMX_PME_CALC_POT;
 282                     }
 283
 284                     /* With domain decomposition we close the CPU side load
 285                      * balancing region here, because PME does global
 286                      * communication that acts as a global barrier.
 287                      */
 288                     ddBalanceRegionHandler.closeAfterForceComputationCpu();
 289
 290                     wallcycle_start(wcycle, ewcPMEMESH);
 291                     status = gmx_pme_do(
 292                             fr->pmedata,
 293                             gmx::constArrayRefFromArray(coordinates.unpaddedConstArrayRef().data(),
 294                                                         md->homenr - fr->n_tpi),
 295                             forceWithVirial.force_, md->chargeA, md->chargeB, md->sqrt_c6A,
 296                             md->sqrt_c6B, md->sigmaA, md->sigmaB, box, cr,
 297                             DOMAINDECOMP(cr) ? dd_pme_maxshift_x(cr->dd) : 0,
 298                             DOMAINDECOMP(cr) ? dd_pme_maxshift_y(cr->dd) : 0, nrnb, wcycle,
 299                             ewaldOutput.vir_q, ewaldOutput.vir_lj, &Vlr_q, &Vlr_lj,
 300                             lambda[efptCOUL], lambda[efptVDW], &ewaldOutput.dvdl[efptCOUL],
 301                             &ewaldOutput.dvdl[efptVDW], pme_flags);
 302                     wallcycle_stop(wcycle, ewcPMEMESH);
 303                     if (status != 0)
 304                     {
 305                         gmx_fatal(FARGS, "Error %d in reciprocal PME routine", status);
 306                     }
 307
 308                     /* We should try to do as little computation after
 309                      * this as possible, because parallel PME synchronizes
 310                      * the nodes, so we want all load imbalance of the
 311                      * rest of the force calculation to be before the PME
 312                      * call.  DD load balancing is done on the whole time
 313                      * of the force call (without PME).
 314                      */
 315                 }
 316                 if (fr->n_tpi > 0)
 317                 {
 318                     if (EVDW_PME(ir->vdwtype))
 319                     {
 320
 321                         gmx_fatal(FARGS, "Test particle insertion not implemented with LJ-PME");
 322                     }
 323                     /* Determine the PME grid energy of the test molecule
 324                      * with the PME grid potential of the other charges.
 325                      */
 326                     gmx_pme_calc_energy(
 327                             fr->pmedata,
 328                             coordinates.unpaddedConstArrayRef().subArray(md->homenr - fr->n_tpi, fr->n_tpi),
 329                             gmx::arrayRefFromArray(md->chargeA + md->homenr - fr->n_tpi, fr->n_tpi),
 330                             &Vlr_q);
 331                 }
 332             }
 333         }
 334
 335         if (fr->ic->eeltype == eelEWALD)
 336         {
 337             Vlr_q = do_ewald(ir, x, as_rvec_array(forceWithVirial.force_.data()), md->chargeA,
 338                              md->chargeB, box, cr, md->homenr, ewaldOutput.vir_q, fr->ic->ewaldcoeff_q,
 339                              lambda[efptCOUL], &ewaldOutput.dvdl[efptCOUL], fr->ewald_table);
 340         }
 341
 342         /* Note that with separate PME nodes we get the real energies later */
 343         // TODO it would be simpler if we just accumulated a single
 344         // long-range virial contribution.
 345         forceWithVirial.addVirialContribution(ewaldOutput.vir_q);
 346         forceWithVirial.addVirialContribution(ewaldOutput.vir_lj);
 347         enerd->dvdl_lin[efptCOUL] += ewaldOutput.dvdl[efptCOUL];
 348         enerd->dvdl_lin[efptVDW] += ewaldOutput.dvdl[efptVDW];
 349         enerd->term[F_COUL_RECIP] = Vlr_q + ewaldOutput.Vcorr_q;
 350         enerd->term[F_LJ_RECIP]   = Vlr_lj + ewaldOutput.Vcorr_lj;
 351
 352         if (debug)
 353         {
 354             fprintf(debug, "Vlr_q = %g, Vcorr_q = %g, Vlr_corr_q = %g\n", Vlr_q,
 355                     ewaldOutput.Vcorr_q, enerd->term[F_COUL_RECIP]);
 356             pr_rvecs(debug, 0, "vir_el_recip after corr", ewaldOutput.vir_q, DIM);
 357             rvec* fshift = as_rvec_array(forceOutputs->forceWithShiftForces().shiftForces().data());
 358             pr_rvecs(debug, 0, "fshift after LR Corrections", fshift, SHIFTS);
 359             fprintf(debug, "Vlr_lj: %g, Vcorr_lj = %g, Vlr_corr_lj = %g\n", Vlr_lj,
 360                     ewaldOutput.Vcorr_lj, enerd->term[F_LJ_RECIP]);
 361             pr_rvecs(debug, 0, "vir_lj_recip after corr", ewaldOutput.vir_lj, DIM);
 362         }
 363     }
 364
 365     if (debug)
 366     {
 367         print_nrnb(debug, nrnb);
 368     }
 369
 370     if (debug)
 371     {
 372         rvec* fshift = as_rvec_array(forceOutputs->forceWithShiftForces().shiftForces().data());
 373         pr_rvecs(debug, 0, "fshift after bondeds", fshift, SHIFTS);
 374     }
 375 }