src/gromacs/nbnxm/nbnxm.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2019, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35
  36 /*! \internal \file
  37  * \brief
  38  * Implements the Nbnxm class
  39  *
  40  * \author Berk Hess <hess@kth.se>
  41  * \ingroup module_nbnxm
  42  */
  43
  44 #include "gmxpre.h"
  45
  46 #include "nbnxm.h"
  47
  48 #include "gromacs/domdec/domdec_struct.h"
  49 #include "gromacs/timing/wallcycle.h"
  50
  51 #include "atomdata.h"
  52 #include "pairlistsets.h"
  53 #include "pairsearch.h"
  54
  55 /*! \cond INTERNAL */
  56
  57 void nbnxn_put_on_grid(nonbonded_verlet_t             *nb_verlet,
  58                        const matrix                    box,
  59                        int                             gridIndex,
  60                        const rvec                      lowerCorner,
  61                        const rvec                      upperCorner,
  62                        const gmx::UpdateGroupsCog     *updateGroupsCog,
  63                        gmx::Range<int>                 atomRange,
  64                        real                            atomDensity,
  65                        gmx::ArrayRef<const int>        atomInfo,
  66                        gmx::ArrayRef<const gmx::RVec>  x,
  67                        int                             numAtomsMoved,
  68                        const int                      *move)
  69 {
  70     nb_verlet->pairSearch_->putOnGrid(box, gridIndex, lowerCorner, upperCorner,
  71                                       updateGroupsCog, atomRange, atomDensity,
  72                                       atomInfo, x, numAtomsMoved, move,
  73                                       nb_verlet->nbat.get());
  74 }
  75
  76 /* Calls nbnxn_put_on_grid for all non-local domains */
  77 void nbnxn_put_on_grid_nonlocal(nonbonded_verlet_t              *nbv,
  78                                 const struct gmx_domdec_zones_t *zones,
  79                                 gmx::ArrayRef<const int>         atomInfo,
  80                                 gmx::ArrayRef<const gmx::RVec>   x)
  81 {
  82     for (int zone = 1; zone < zones->n; zone++)
  83     {
  84         rvec c0, c1;
  85         for (int d = 0; d < DIM; d++)
  86         {
  87             c0[d] = zones->size[zone].bb_x0[d];
  88             c1[d] = zones->size[zone].bb_x1[d];
  89         }
  90
  91         nbnxn_put_on_grid(nbv, nullptr,
  92                           zone, c0, c1,
  93                           nullptr,
  94                           { zones->cg_range[zone], zones->cg_range[zone+1] },
  95                           -1,
  96                           atomInfo,
  97                           x,
  98                           0, nullptr);
  99     }
 100 }
 101
 102 bool nonbonded_verlet_t::isDynamicPruningStepCpu(int64_t step) const
 103 {
 104     return pairlistSets_->isDynamicPruningStepCpu(step);
 105 }
 106
 107 bool nonbonded_verlet_t::isDynamicPruningStepGpu(int64_t step) const
 108 {
 109     return pairlistSets_->isDynamicPruningStepGpu(step);
 110 }
 111
 112 gmx::ArrayRef<const int> nonbonded_verlet_t::getLocalAtomOrder() const
 113 {
 114     /* Return the atom order for the home cell (index 0) */
 115     const Nbnxm::Grid &grid       = pairSearch_->gridSet().grids()[0];
 116
 117     const int          numIndices = grid.atomIndexEnd() - grid.firstAtomInColumn(0);
 118
 119     return gmx::constArrayRefFromArray(pairSearch_->gridSet().atomIndices().data(), numIndices);
 120 }
 121
 122 void nonbonded_verlet_t::setLocalAtomOrder()
 123 {
 124     pairSearch_->setLocalAtomOrder();
 125 }
 126
 127 void nonbonded_verlet_t::setAtomProperties(const t_mdatoms          &mdatoms,
 128                                            gmx::ArrayRef<const int>  atomInfo)
 129 {
 130     nbnxn_atomdata_set(nbat.get(), pairSearch_->gridSet(), &mdatoms, atomInfo.data());
 131 }
 132
 133 void nonbonded_verlet_t::convertCoordinates(const gmx::AtomLocality         locality,
 134                                             const bool                      fillLocal,
 135                                             gmx::ArrayRef<const gmx::RVec>  coordinates)
 136 {
 137     wallcycle_start(wcycle_, ewcNB_XF_BUF_OPS);
 138     wallcycle_sub_start(wcycle_, ewcsNB_X_BUF_OPS);
 139
 140     nbnxn_atomdata_copy_x_to_nbat_x(pairSearch_->gridSet(), locality, fillLocal,
 141                                     as_rvec_array(coordinates.data()),
 142                                     nbat.get());
 143
 144     wallcycle_sub_stop(wcycle_, ewcsNB_X_BUF_OPS);
 145     wallcycle_stop(wcycle_, ewcNB_XF_BUF_OPS);
 146 }
 147
 148 void nonbonded_verlet_t::convertCoordinatesGpu(const gmx::AtomLocality          locality,
 149                                                const bool                       fillLocal,
 150                                                DeviceBuffer<float>              d_x,
 151                                                GpuEventSynchronizer            *xReadyOnDevice)
 152 {
 153     wallcycle_start(wcycle_, ewcNB_XF_BUF_OPS);
 154     wallcycle_sub_start(wcycle_, ewcsNB_X_BUF_OPS);
 155
 156     nbnxn_atomdata_x_to_nbat_x_gpu(pairSearch_->gridSet(), locality, fillLocal,
 157                                    gpu_nbv,
 158                                    d_x,
 159                                    xReadyOnDevice);
 160
 161     wallcycle_sub_stop(wcycle_, ewcsNB_X_BUF_OPS);
 162     wallcycle_stop(wcycle_, ewcNB_XF_BUF_OPS);
 163 }
 164
 165 gmx::ArrayRef<const int> nonbonded_verlet_t::getGridIndices() const
 166 {
 167     return pairSearch_->gridSet().cells();
 168 }
 169
 170 void
 171 nonbonded_verlet_t::atomdata_add_nbat_f_to_f(const gmx::AtomLocality  locality,
 172                                              gmx::ArrayRef<gmx::RVec> force)
 173 {
 174
 175     /* Skip the reduction if there was no short-range GPU work to do
 176      * (either NB or both NB and bonded work). */
 177     if (!pairlistIsSimple() && !haveGpuShortRangeWork(locality))
 178     {
 179         return;
 180     }
 181
 182     wallcycle_start(wcycle_, ewcNB_XF_BUF_OPS);
 183     wallcycle_sub_start(wcycle_, ewcsNB_F_BUF_OPS);
 184
 185     reduceForces(nbat.get(), locality, pairSearch_->gridSet(), as_rvec_array(force.data()));
 186
 187     wallcycle_sub_stop(wcycle_, ewcsNB_F_BUF_OPS);
 188     wallcycle_stop(wcycle_, ewcNB_XF_BUF_OPS);
 189 }
 190
 191 void
 192 nonbonded_verlet_t::atomdata_add_nbat_f_to_f_gpu(const gmx::AtomLocality                     locality,
 193                                                  DeviceBuffer<float>                         totalForcesDevice,
 194                                                  void                                       *forcesPmeDevice,
 195                                                  gmx::ArrayRef<GpuEventSynchronizer* const>  dependencyList,
 196                                                  bool                                        useGpuFPmeReduction,
 197                                                  bool                                        accumulateForce)
 198 {
 199
 200     GMX_ASSERT((useGpuFPmeReduction == (forcesPmeDevice != nullptr)),
 201                "GPU PME force reduction is only valid when a non-null GPU PME force pointer is available");
 202
 203     /* Skip the reduction if there was no short-range GPU work to do
 204      * (either NB or both NB and bonded work). */
 205     if (!pairlistIsSimple() && !haveGpuShortRangeWork(locality))
 206     {
 207         return;
 208     }
 209
 210     wallcycle_start(wcycle_, ewcNB_XF_BUF_OPS);
 211     wallcycle_sub_start(wcycle_, ewcsNB_F_BUF_OPS);
 212
 213     reduceForcesGpu(locality, totalForcesDevice, pairSearch_->gridSet(), forcesPmeDevice, dependencyList, gpu_nbv, useGpuFPmeReduction, accumulateForce);
 214
 215     wallcycle_sub_stop(wcycle_, ewcsNB_F_BUF_OPS);
 216     wallcycle_stop(wcycle_, ewcNB_XF_BUF_OPS);
 217 }
 218
 219 void
 220 nonbonded_verlet_t::atomdata_init_add_nbat_f_to_f_gpu(GpuEventSynchronizer* const localReductionDone)
 221 {
 222
 223     wallcycle_start(wcycle_, ewcNB_XF_BUF_OPS);
 224     wallcycle_sub_start(wcycle_, ewcsNB_F_BUF_OPS);
 225
 226     const Nbnxm::GridSet      &gridSet = pairSearch_->gridSet();
 227
 228     Nbnxm::nbnxn_gpu_init_add_nbat_f_to_f(gridSet.cells().data(),
 229                                           gpu_nbv,
 230                                           gridSet.numRealAtomsTotal(),
 231                                           localReductionDone);
 232
 233     wallcycle_sub_stop(wcycle_, ewcsNB_F_BUF_OPS);
 234     wallcycle_stop(wcycle_, ewcNB_XF_BUF_OPS);
 235 }
 236
 237 real nonbonded_verlet_t::pairlistInnerRadius() const
 238 {
 239     return pairlistSets_->params().rlistInner;
 240 }
 241
 242 real nonbonded_verlet_t::pairlistOuterRadius() const
 243 {
 244     return pairlistSets_->params().rlistOuter;
 245 }
 246
 247 void nonbonded_verlet_t::changePairlistRadii(real rlistOuter,
 248                                              real rlistInner)
 249 {
 250     pairlistSets_->changePairlistRadii(rlistOuter, rlistInner);
 251 }
 252
 253 void
 254 nonbonded_verlet_t::atomdata_init_copy_x_to_nbat_x_gpu()
 255 {
 256     Nbnxm::nbnxn_gpu_init_x_to_nbat_x(pairSearch_->gridSet(), gpu_nbv);
 257 }
 258
 259 void nonbonded_verlet_t::insertNonlocalGpuDependency(const gmx::InteractionLocality interactionLocality)
 260 {
 261     Nbnxm::nbnxnInsertNonlocalGpuDependency(gpu_nbv, interactionLocality);
 262 }
 263
 264 void* nonbonded_verlet_t::get_x_on_device_event()
 265 {
 266     return Nbnxm::nbnxn_get_x_on_device_event(gpu_nbv);
 267 }
 268
 269 void nonbonded_verlet_t::wait_nonlocal_x_copy_D2H_done()
 270 {
 271     Nbnxm::nbnxn_wait_nonlocal_x_copy_D2H_done(gpu_nbv);
 272 }
 273
 274 void nonbonded_verlet_t::stream_local_wait_for_nonlocal()
 275 {
 276     Nbnxm::nbnxn_stream_local_wait_for_nonlocal(gpu_nbv);
 277 }
 278
 279 /*! \endcond */