src/gromacs/ewald/pme_pp.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) 2018,2019,2020,2021, 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 /*! \internal \file
  39  *
  40  * \brief This file contains function definitions necessary for
  41  * managing the offload of long-ranged PME work to separate MPI rank,
  42  * for computing energies and forces (Coulomb and LJ).
  43  *
  44  * \author Berk Hess <hess@kth.se>
  45  * \ingroup module_ewald
  46  */
  47
  48 #include "gmxpre.h"
  49
  50 #include "pme_pp.h"
  51
  52 #include "config.h"
  53
  54 #include <cstdio>
  55 #include <cstring>
  56
  57 #include "gromacs/domdec/domdec.h"
  58 #include "gromacs/domdec/domdec_struct.h"
  59 #include "gromacs/ewald/pme.h"
  60 #include "gromacs/ewald/pme_pp_comm_gpu.h"
  61 #include "gromacs/gmxlib/network.h"
  62 #include "gromacs/math/vec.h"
  63 #include "gromacs/mdlib/gmx_omp_nthreads.h"
  64 #include "gromacs/mdtypes/commrec.h"
  65 #include "gromacs/mdtypes/forceoutput.h"
  66 #include "gromacs/mdtypes/forcerec.h"
  67 #include "gromacs/mdtypes/interaction_const.h"
  68 #include "gromacs/mdtypes/md_enums.h"
  69 #include "gromacs/mdtypes/state_propagator_data_gpu.h"
  70 #include "gromacs/nbnxm/nbnxm.h"
  71 #include "gromacs/timing/wallcycle.h"
  72 #include "gromacs/utility/fatalerror.h"
  73 #include "gromacs/utility/gmxmpi.h"
  74 #include "gromacs/utility/smalloc.h"
  75
  76 #include "pme_pp_communication.h"
  77
  78 /*! \brief Block to wait for communication to PME ranks to complete
  79  *
  80  * This should be faster with a real non-blocking MPI implementation
  81  */
  82 static constexpr bool c_useDelayedWait = false;
  83
  84 /*! \brief Wait for the pending data send requests to PME ranks to complete */
  85 static void gmx_pme_send_coeffs_coords_wait(gmx_domdec_t* dd)
  86 {
  87     if (dd->nreq_pme)
  88     {
  89 #if GMX_MPI
  90         MPI_Waitall(dd->nreq_pme, dd->req_pme, MPI_STATUSES_IGNORE);
  91 #endif
  92         dd->nreq_pme = 0;
  93     }
  94 }
  95
  96 /*! \brief Send data to PME ranks */
  97 static void gmx_pme_send_coeffs_coords(t_forcerec*                    fr,
  98                                        const t_commrec*               cr,
  99                                        unsigned int                   flags,
 100                                        gmx::ArrayRef<const real>      chargeA,
 101                                        gmx::ArrayRef<const real>      chargeB,
 102                                        gmx::ArrayRef<const real>      c6A,
 103                                        gmx::ArrayRef<const real>      c6B,
 104                                        gmx::ArrayRef<const real>      sigmaA,
 105                                        gmx::ArrayRef<const real>      sigmaB,
 106                                        const matrix                   box,
 107                                        gmx::ArrayRef<const gmx::RVec> x,
 108                                        real                           lambda_q,
 109                                        real                           lambda_lj,
 110                                        int                            maxshift_x,
 111                                        int                            maxshift_y,
 112                                        int64_t                        step,
 113                                        bool                           useGpuPmePpComms,
 114                                        bool                           reinitGpuPmePpComms,
 115                                        bool                           sendCoordinatesFromGpu,
 116                                        GpuEventSynchronizer*          coordinatesReadyOnDeviceEvent)
 117 {
 118     gmx_domdec_t*         dd;
 119     gmx_pme_comm_n_box_t* cnb;
 120     int                   n;
 121
 122     dd = cr->dd;
 123     n  = dd_numHomeAtoms(*dd);
 124
 125     if (debug)
 126     {
 127         fprintf(debug,
 128                 "PP rank %d sending to PME rank %d: %d%s%s%s%s\n",
 129                 cr->sim_nodeid,
 130                 dd->pme_nodeid,
 131                 n,
 132                 (flags & PP_PME_CHARGE) ? " charges" : "",
 133                 (flags & PP_PME_SQRTC6) ? " sqrtC6" : "",
 134                 (flags & PP_PME_SIGMA) ? " sigma" : "",
 135                 (flags & PP_PME_COORD) ? " coordinates" : "");
 136     }
 137
 138     if (useGpuPmePpComms)
 139     {
 140         flags |= PP_PME_GPUCOMMS;
 141     }
 142
 143     if (c_useDelayedWait)
 144     {
 145         /* We can not use cnb until pending communication has finished */
 146         gmx_pme_send_coeffs_coords_wait(dd);
 147     }
 148
 149     if (dd->pme_receive_vir_ener)
 150     {
 151         /* Peer PP node: communicate all data */
 152         if (dd->cnb == nullptr)
 153         {
 154             snew(dd->cnb, 1);
 155         }
 156         cnb = dd->cnb;
 157
 158         cnb->flags      = flags;
 159         cnb->natoms     = n;
 160         cnb->maxshift_x = maxshift_x;
 161         cnb->maxshift_y = maxshift_y;
 162         cnb->lambda_q   = lambda_q;
 163         cnb->lambda_lj  = lambda_lj;
 164         cnb->step       = step;
 165         if (flags & PP_PME_COORD)
 166         {
 167             copy_mat(box, cnb->box);
 168         }
 169 #if GMX_MPI
 170         MPI_Isend(cnb,
 171                   sizeof(*cnb),
 172                   MPI_BYTE,
 173                   dd->pme_nodeid,
 174                   eCommType_CNB,
 175                   cr->mpi_comm_mysim,
 176                   &dd->req_pme[dd->nreq_pme++]);
 177 #endif
 178     }
 179     else if (flags & (PP_PME_CHARGE | PP_PME_SQRTC6 | PP_PME_SIGMA))
 180     {
 181 #if GMX_MPI
 182         /* Communicate only the number of atoms */
 183         MPI_Isend(&n,
 184                   sizeof(n),
 185                   MPI_BYTE,
 186                   dd->pme_nodeid,
 187                   eCommType_CNB,
 188                   cr->mpi_comm_mysim,
 189                   &dd->req_pme[dd->nreq_pme++]);
 190 #endif
 191     }
 192
 193 #if GMX_MPI
 194     if (n > 0)
 195     {
 196         if (flags & PP_PME_CHARGE)
 197         {
 198             MPI_Isend(chargeA.data(),
 199                       n * sizeof(real),
 200                       MPI_BYTE,
 201                       dd->pme_nodeid,
 202                       eCommType_ChargeA,
 203                       cr->mpi_comm_mysim,
 204                       &dd->req_pme[dd->nreq_pme++]);
 205         }
 206         if (flags & PP_PME_CHARGEB)
 207         {
 208             MPI_Isend(chargeB.data(),
 209                       n * sizeof(real),
 210                       MPI_BYTE,
 211                       dd->pme_nodeid,
 212                       eCommType_ChargeB,
 213                       cr->mpi_comm_mysim,
 214                       &dd->req_pme[dd->nreq_pme++]);
 215         }
 216         if (flags & PP_PME_SQRTC6)
 217         {
 218             MPI_Isend(c6A.data(),
 219                       n * sizeof(real),
 220                       MPI_BYTE,
 221                       dd->pme_nodeid,
 222                       eCommType_SQRTC6A,
 223                       cr->mpi_comm_mysim,
 224                       &dd->req_pme[dd->nreq_pme++]);
 225         }
 226         if (flags & PP_PME_SQRTC6B)
 227         {
 228             MPI_Isend(c6B.data(),
 229                       n * sizeof(real),
 230                       MPI_BYTE,
 231                       dd->pme_nodeid,
 232                       eCommType_SQRTC6B,
 233                       cr->mpi_comm_mysim,
 234                       &dd->req_pme[dd->nreq_pme++]);
 235         }
 236         if (flags & PP_PME_SIGMA)
 237         {
 238             MPI_Isend(sigmaA.data(),
 239                       n * sizeof(real),
 240                       MPI_BYTE,
 241                       dd->pme_nodeid,
 242                       eCommType_SigmaA,
 243                       cr->mpi_comm_mysim,
 244                       &dd->req_pme[dd->nreq_pme++]);
 245         }
 246         if (flags & PP_PME_SIGMAB)
 247         {
 248             MPI_Isend(sigmaB.data(),
 249                       n * sizeof(real),
 250                       MPI_BYTE,
 251                       dd->pme_nodeid,
 252                       eCommType_SigmaB,
 253                       cr->mpi_comm_mysim,
 254                       &dd->req_pme[dd->nreq_pme++]);
 255         }
 256         if (flags & PP_PME_COORD)
 257         {
 258             if (reinitGpuPmePpComms)
 259             {
 260                 fr->pmePpCommGpu->reinit(n);
 261             }
 262
 263             if (useGpuPmePpComms && (fr != nullptr))
 264             {
 265                 if (sendCoordinatesFromGpu)
 266                 {
 267                     fr->pmePpCommGpu->sendCoordinatesToPmeFromGpu(
 268                             fr->stateGpu->getCoordinates(), n, coordinatesReadyOnDeviceEvent);
 269                 }
 270                 else
 271                 {
 272                     fr->pmePpCommGpu->sendCoordinatesToPmeFromCpu(
 273                             const_cast<gmx::RVec*>(x.data()), n, coordinatesReadyOnDeviceEvent);
 274                 }
 275             }
 276             else
 277             {
 278                 MPI_Isend(x.data(),
 279                           n * sizeof(rvec),
 280                           MPI_BYTE,
 281                           dd->pme_nodeid,
 282                           eCommType_COORD,
 283                           cr->mpi_comm_mysim,
 284                           &dd->req_pme[dd->nreq_pme++]);
 285             }
 286         }
 287     }
 288 #else
 289     GMX_UNUSED_VALUE(fr);
 290     GMX_UNUSED_VALUE(chargeA);
 291     GMX_UNUSED_VALUE(chargeB);
 292     GMX_UNUSED_VALUE(c6A);
 293     GMX_UNUSED_VALUE(c6B);
 294     GMX_UNUSED_VALUE(sigmaA);
 295     GMX_UNUSED_VALUE(sigmaB);
 296     GMX_UNUSED_VALUE(reinitGpuPmePpComms);
 297     GMX_UNUSED_VALUE(sendCoordinatesFromGpu);
 298     GMX_UNUSED_VALUE(coordinatesReadyOnDeviceEvent);
 299 #endif
 300     if (!c_useDelayedWait)
 301     {
 302         /* Wait for the data to arrive */
 303         /* We can skip this wait as we are sure x and q will not be modified
 304          * before the next call to gmx_pme_send_x_q or gmx_pme_receive_f.
 305          */
 306         gmx_pme_send_coeffs_coords_wait(dd);
 307     }
 308 }
 309
 310 void gmx_pme_send_parameters(const t_commrec*           cr,
 311                              const interaction_const_t& interactionConst,
 312                              bool                       bFreeEnergy_q,
 313                              bool                       bFreeEnergy_lj,
 314                              gmx::ArrayRef<const real>  chargeA,
 315                              gmx::ArrayRef<const real>  chargeB,
 316                              gmx::ArrayRef<const real>  sqrt_c6A,
 317                              gmx::ArrayRef<const real>  sqrt_c6B,
 318                              gmx::ArrayRef<const real>  sigmaA,
 319                              gmx::ArrayRef<const real>  sigmaB,
 320                              int                        maxshift_x,
 321                              int                        maxshift_y)
 322 {
 323     unsigned int flags = 0;
 324
 325     if (EEL_PME(interactionConst.eeltype))
 326     {
 327         flags |= PP_PME_CHARGE;
 328     }
 329     if (EVDW_PME(interactionConst.vdwtype))
 330     {
 331         flags |= (PP_PME_SQRTC6 | PP_PME_SIGMA);
 332     }
 333     if (bFreeEnergy_q || bFreeEnergy_lj)
 334     {
 335         /* Assumes that the B state flags are in the bits just above
 336          * the ones for the A state. */
 337         flags |= (flags << 1);
 338     }
 339
 340     gmx_pme_send_coeffs_coords(nullptr,
 341                                cr,
 342                                flags,
 343                                chargeA,
 344                                chargeB,
 345                                sqrt_c6A,
 346                                sqrt_c6B,
 347                                sigmaA,
 348                                sigmaB,
 349                                nullptr,
 350                                gmx::ArrayRef<gmx::RVec>(),
 351                                0,
 352                                0,
 353                                maxshift_x,
 354                                maxshift_y,
 355                                -1,
 356                                false,
 357                                false,
 358                                false,
 359                                nullptr);
 360 }
 361
 362 void gmx_pme_send_coordinates(t_forcerec*                    fr,
 363                               const t_commrec*               cr,
 364                               const matrix                   box,
 365                               gmx::ArrayRef<const gmx::RVec> x,
 366                               real                           lambda_q,
 367                               real                           lambda_lj,
 368                               bool                           computeEnergyAndVirial,
 369                               int64_t                        step,
 370                               bool                           useGpuPmePpComms,
 371                               bool                           receiveCoordinateAddressFromPme,
 372                               bool                           sendCoordinatesFromGpu,
 373                               GpuEventSynchronizer*          coordinatesReadyOnDeviceEvent,
 374                               gmx_wallcycle*                 wcycle)
 375 {
 376     wallcycle_start(wcycle, WallCycleCounter::PpPmeSendX);
 377
 378     unsigned int flags = PP_PME_COORD;
 379     if (computeEnergyAndVirial)
 380     {
 381         flags |= PP_PME_ENER_VIR;
 382     }
 383     gmx_pme_send_coeffs_coords(fr,
 384                                cr,
 385                                flags,
 386                                {},
 387                                {},
 388                                {},
 389                                {},
 390                                {},
 391                                {},
 392                                box,
 393                                x,
 394                                lambda_q,
 395                                lambda_lj,
 396                                0,
 397                                0,
 398                                step,
 399                                useGpuPmePpComms,
 400                                receiveCoordinateAddressFromPme,
 401                                sendCoordinatesFromGpu,
 402                                coordinatesReadyOnDeviceEvent);
 403
 404     wallcycle_stop(wcycle, WallCycleCounter::PpPmeSendX);
 405 }
 406
 407 void gmx_pme_send_finish(const t_commrec* cr)
 408 {
 409     unsigned int flags = PP_PME_FINISH;
 410
 411     gmx_pme_send_coeffs_coords(
 412             nullptr, cr, flags, {}, {}, {}, {}, {}, {}, nullptr, gmx::ArrayRef<gmx::RVec>(), 0, 0, 0, 0, -1, false, false, false, nullptr);
 413 }
 414
 415 void gmx_pme_send_switchgrid(const t_commrec* cr, ivec grid_size, real ewaldcoeff_q, real ewaldcoeff_lj)
 416 {
 417 #if GMX_MPI
 418     gmx_pme_comm_n_box_t cnb;
 419
 420     /* Only let one PP node signal each PME node */
 421     if (cr->dd->pme_receive_vir_ener)
 422     {
 423         cnb.flags = PP_PME_SWITCHGRID;
 424         copy_ivec(grid_size, cnb.grid_size);
 425         cnb.ewaldcoeff_q  = ewaldcoeff_q;
 426         cnb.ewaldcoeff_lj = ewaldcoeff_lj;
 427
 428         /* We send this, uncommon, message blocking to simplify the code */
 429         MPI_Send(&cnb, sizeof(cnb), MPI_BYTE, cr->dd->pme_nodeid, eCommType_CNB, cr->mpi_comm_mysim);
 430     }
 431 #else
 432     GMX_UNUSED_VALUE(cr);
 433     GMX_UNUSED_VALUE(grid_size);
 434     GMX_UNUSED_VALUE(ewaldcoeff_q);
 435     GMX_UNUSED_VALUE(ewaldcoeff_lj);
 436 #endif
 437 }
 438
 439 void gmx_pme_send_resetcounters(const t_commrec gmx_unused* cr, int64_t gmx_unused step)
 440 {
 441 #if GMX_MPI
 442     gmx_pme_comm_n_box_t cnb;
 443
 444     /* Only let one PP node signal each PME node */
 445     if (cr->dd->pme_receive_vir_ener)
 446     {
 447         cnb.flags = PP_PME_RESETCOUNTERS;
 448         cnb.step  = step;
 449
 450         /* We send this, uncommon, message blocking to simplify the code */
 451         MPI_Send(&cnb, sizeof(cnb), MPI_BYTE, cr->dd->pme_nodeid, eCommType_CNB, cr->mpi_comm_mysim);
 452     }
 453 #endif
 454 }
 455
 456 /*! \brief Receive virial and energy from PME rank */
 457 static void receive_virial_energy(const t_commrec*      cr,
 458                                   gmx::ForceWithVirial* forceWithVirial,
 459                                   real*                 energy_q,
 460                                   real*                 energy_lj,
 461                                   real*                 dvdlambda_q,
 462                                   real*                 dvdlambda_lj,
 463                                   float*                pme_cycles)
 464 {
 465     gmx_pme_comm_vir_ene_t cve;
 466
 467     if (cr->dd->pme_receive_vir_ener)
 468     {
 469         if (debug)
 470         {
 471             fprintf(debug,
 472                     "PP rank %d receiving from PME rank %d: virial and energy\n",
 473                     cr->sim_nodeid,
 474                     cr->dd->pme_nodeid);
 475         }
 476 #if GMX_MPI
 477         MPI_Recv(&cve, sizeof(cve), MPI_BYTE, cr->dd->pme_nodeid, 1, cr->mpi_comm_mysim, MPI_STATUS_IGNORE);
 478 #else
 479         memset(&cve, 0, sizeof(cve));
 480 #endif
 481
 482         forceWithVirial->addVirialContribution(cve.vir_q);
 483         forceWithVirial->addVirialContribution(cve.vir_lj);
 484         *energy_q  = cve.energy_q;
 485         *energy_lj = cve.energy_lj;
 486         *dvdlambda_q += cve.dvdlambda_q;
 487         *dvdlambda_lj += cve.dvdlambda_lj;
 488         *pme_cycles = cve.cycles;
 489
 490         if (cve.stop_cond != StopCondition::None)
 491         {
 492             gmx_set_stop_condition(cve.stop_cond);
 493         }
 494     }
 495     else
 496     {
 497         *energy_q   = 0;
 498         *energy_lj  = 0;
 499         *pme_cycles = 0;
 500     }
 501 }
 502
 503 /*! \brief Recieve force data from PME ranks */
 504 static void recvFFromPme(gmx::PmePpCommGpu* pmePpCommGpu,
 505                          void*              recvptr,
 506                          int                n,
 507                          const t_commrec*   cr,
 508                          bool               useGpuPmePpComms,
 509                          bool               receivePmeForceToGpu)
 510 {
 511     if (useGpuPmePpComms)
 512     {
 513         GMX_ASSERT(pmePpCommGpu != nullptr, "Need valid pmePpCommGpu");
 514         // Receive forces from PME rank
 515         pmePpCommGpu->receiveForceFromPme(static_cast<gmx::RVec*>(recvptr), n, receivePmeForceToGpu);
 516     }
 517     else
 518     {
 519         // Receive data using MPI
 520 #if GMX_MPI
 521         MPI_Recv(recvptr, n * sizeof(rvec), MPI_BYTE, cr->dd->pme_nodeid, 0, cr->mpi_comm_mysim, MPI_STATUS_IGNORE);
 522 #else
 523         GMX_UNUSED_VALUE(cr);
 524 #endif
 525     }
 526 }
 527
 528
 529 void gmx_pme_receive_f(gmx::PmePpCommGpu*    pmePpCommGpu,
 530                        const t_commrec*      cr,
 531                        gmx::ForceWithVirial* forceWithVirial,
 532                        real*                 energy_q,
 533                        real*                 energy_lj,
 534                        real*                 dvdlambda_q,
 535                        real*                 dvdlambda_lj,
 536                        bool                  useGpuPmePpComms,
 537                        bool                  receivePmeForceToGpu,
 538                        float*                pme_cycles)
 539 {
 540     if (c_useDelayedWait)
 541     {
 542         /* Wait for the x request to finish */
 543         gmx_pme_send_coeffs_coords_wait(cr->dd);
 544     }
 545
 546     const int               natoms = dd_numHomeAtoms(*cr->dd);
 547     std::vector<gmx::RVec>& buffer = cr->dd->pmeForceReceiveBuffer;
 548     buffer.resize(natoms);
 549
 550     void* recvptr = reinterpret_cast<void*>(buffer.data());
 551     recvFFromPme(pmePpCommGpu, recvptr, natoms, cr, useGpuPmePpComms, receivePmeForceToGpu);
 552
 553     int nt = gmx_omp_nthreads_get_simple_rvec_task(ModuleMultiThread::Default, natoms);
 554
 555     gmx::ArrayRef<gmx::RVec> f = forceWithVirial->force_;
 556
 557     if (!receivePmeForceToGpu)
 558     {
 559         /* Note that we would like to avoid this conditional by putting it
 560          * into the omp pragma instead, but then we still take the full
 561          * omp parallel for overhead (at least with gcc5).
 562          */
 563         if (nt == 1)
 564         {
 565             for (int i = 0; i < natoms; i++)
 566             {
 567                 f[i] += buffer[i];
 568             }
 569         }
 570         else
 571         {
 572 #pragma omp parallel for num_threads(nt) schedule(static)
 573             for (int i = 0; i < natoms; i++)
 574             {
 575                 f[i] += buffer[i];
 576             }
 577         }
 578     }
 579
 580     receive_virial_energy(cr, forceWithVirial, energy_q, energy_lj, dvdlambda_q, dvdlambda_lj, pme_cycles);
 581 }