src/gromacs/nbnxm/nbnxm.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2019,2020, 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/nbnxm/atomdata.h"
  50 #include "gromacs/timing/wallcycle.h"
  51
  52 #include "nbnxm_gpu.h"
  53 #include "pairlistsets.h"
  54 #include "pairsearch.h"
  55
  56 /*! \cond INTERNAL */
  57
  58 void nbnxn_put_on_grid(nonbonded_verlet_t*            nb_verlet,
  59                        const matrix                   box,
  60                        int                            gridIndex,
  61                        const rvec                     lowerCorner,
  62                        const rvec                     upperCorner,
  63                        const gmx::UpdateGroupsCog*    updateGroupsCog,
  64                        gmx::Range<int>                atomRange,
  65                        real                           atomDensity,
  66                        gmx::ArrayRef<const int>       atomInfo,
  67                        gmx::ArrayRef<const gmx::RVec> x,
  68                        int                            numAtomsMoved,
  69                        const int*                     move)
  70 {
  71     nb_verlet->pairSearch_->putOnGrid(box, gridIndex, lowerCorner, upperCorner, updateGroupsCog,
  72                                       atomRange, atomDensity, 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, zone, c0, c1, nullptr,
  92                           { zones->cg_range[zone], zones->cg_range[zone + 1] }, -1, atomInfo, x, 0,
  93                           nullptr);
  94     }
  95 }
  96
  97 bool nonbonded_verlet_t::isDynamicPruningStepCpu(int64_t step) const
  98 {
  99     return pairlistSets_->isDynamicPruningStepCpu(step);
 100 }
 101
 102 bool nonbonded_verlet_t::isDynamicPruningStepGpu(int64_t step) const
 103 {
 104     return pairlistSets_->isDynamicPruningStepGpu(step);
 105 }
 106
 107 gmx::ArrayRef<const int> nonbonded_verlet_t::getLocalAtomOrder() const
 108 {
 109     /* Return the atom order for the home cell (index 0) */
 110     const Nbnxm::Grid& grid = pairSearch_->gridSet().grids()[0];
 111
 112     const int numIndices = grid.atomIndexEnd() - grid.firstAtomInColumn(0);
 113
 114     return gmx::constArrayRefFromArray(pairSearch_->gridSet().atomIndices().data(), numIndices);
 115 }
 116
 117 void nonbonded_verlet_t::setLocalAtomOrder()
 118 {
 119     pairSearch_->setLocalAtomOrder();
 120 }
 121
 122 void nonbonded_verlet_t::setAtomProperties(gmx::ArrayRef<const int>  atomTypes,
 123                                            gmx::ArrayRef<const real> atomCharges,
 124                                            gmx::ArrayRef<const int>  atomInfo)
 125 {
 126     nbnxn_atomdata_set(nbat.get(), pairSearch_->gridSet(), atomTypes, atomCharges, atomInfo);
 127 }
 128
 129 void nonbonded_verlet_t::convertCoordinates(const gmx::AtomLocality        locality,
 130                                             const bool                     fillLocal,
 131                                             gmx::ArrayRef<const gmx::RVec> coordinates)
 132 {
 133     wallcycle_start(wcycle_, ewcNB_XF_BUF_OPS);
 134     wallcycle_sub_start(wcycle_, ewcsNB_X_BUF_OPS);
 135
 136     nbnxn_atomdata_copy_x_to_nbat_x(pairSearch_->gridSet(), locality, fillLocal,
 137                                     as_rvec_array(coordinates.data()), nbat.get());
 138
 139     wallcycle_sub_stop(wcycle_, ewcsNB_X_BUF_OPS);
 140     wallcycle_stop(wcycle_, ewcNB_XF_BUF_OPS);
 141 }
 142
 143 void nonbonded_verlet_t::convertCoordinatesGpu(const gmx::AtomLocality locality,
 144                                                const bool              fillLocal,
 145                                                DeviceBuffer<gmx::RVec> d_x,
 146                                                GpuEventSynchronizer*   xReadyOnDevice)
 147 {
 148     wallcycle_start(wcycle_, ewcNB_XF_BUF_OPS);
 149     wallcycle_sub_start(wcycle_, ewcsNB_X_BUF_OPS);
 150
 151     nbnxn_atomdata_x_to_nbat_x_gpu(pairSearch_->gridSet(), locality, fillLocal, gpu_nbv, d_x, xReadyOnDevice);
 152
 153     wallcycle_sub_stop(wcycle_, ewcsNB_X_BUF_OPS);
 154     wallcycle_stop(wcycle_, ewcNB_XF_BUF_OPS);
 155 }
 156
 157 gmx::ArrayRef<const int> nonbonded_verlet_t::getGridIndices() const
 158 {
 159     return pairSearch_->gridSet().cells();
 160 }
 161
 162 void nonbonded_verlet_t::atomdata_add_nbat_f_to_f(const gmx::AtomLocality  locality,
 163                                                   gmx::ArrayRef<gmx::RVec> force)
 164 {
 165
 166     /* Skip the reduction if there was no short-range GPU work to do
 167      * (either NB or both NB and bonded work). */
 168     if (!pairlistIsSimple() && !Nbnxm::haveGpuShortRangeWork(gpu_nbv, locality))
 169     {
 170         return;
 171     }
 172
 173     wallcycle_start(wcycle_, ewcNB_XF_BUF_OPS);
 174     wallcycle_sub_start(wcycle_, ewcsNB_F_BUF_OPS);
 175
 176     reduceForces(nbat.get(), locality, pairSearch_->gridSet(), as_rvec_array(force.data()));
 177
 178     wallcycle_sub_stop(wcycle_, ewcsNB_F_BUF_OPS);
 179     wallcycle_stop(wcycle_, ewcNB_XF_BUF_OPS);
 180 }
 181
 182 void nonbonded_verlet_t::atomdata_add_nbat_f_to_f_gpu(const gmx::AtomLocality locality,
 183                                                       DeviceBuffer<gmx::RVec> totalForcesDevice,
 184                                                       void*                   forcesPmeDevice,
 185                                                       gmx::ArrayRef<GpuEventSynchronizer* const> dependencyList,
 186                                                       bool useGpuFPmeReduction,
 187                                                       bool accumulateForce)
 188 {
 189
 190     GMX_ASSERT((useGpuFPmeReduction == (forcesPmeDevice != nullptr)),
 191                "GPU PME force reduction is only valid when a non-null GPU PME force pointer is "
 192                "available");
 193
 194     /* Skip the reduction if there was no short-range GPU work to do
 195      * (either NB or both NB and bonded work). */
 196     if (!pairlistIsSimple() && !Nbnxm::haveGpuShortRangeWork(gpu_nbv, locality))
 197     {
 198         return;
 199     }
 200
 201     wallcycle_start(wcycle_, ewcNB_XF_BUF_OPS);
 202     wallcycle_sub_start(wcycle_, ewcsNB_F_BUF_OPS);
 203
 204     reduceForcesGpu(locality, totalForcesDevice, pairSearch_->gridSet(), forcesPmeDevice,
 205                     dependencyList, gpu_nbv, useGpuFPmeReduction, accumulateForce);
 206
 207     wallcycle_sub_stop(wcycle_, ewcsNB_F_BUF_OPS);
 208     wallcycle_stop(wcycle_, ewcNB_XF_BUF_OPS);
 209 }
 210
 211 void nonbonded_verlet_t::atomdata_init_add_nbat_f_to_f_gpu(GpuEventSynchronizer* const localReductionDone)
 212 {
 213
 214     wallcycle_start(wcycle_, ewcNB_XF_BUF_OPS);
 215     wallcycle_sub_start(wcycle_, ewcsNB_F_BUF_OPS);
 216
 217     const Nbnxm::GridSet& gridSet = pairSearch_->gridSet();
 218
 219     Nbnxm::nbnxn_gpu_init_add_nbat_f_to_f(gridSet.cells().data(), gpu_nbv,
 220                                           gridSet.numRealAtomsTotal(), localReductionDone);
 221
 222     wallcycle_sub_stop(wcycle_, ewcsNB_F_BUF_OPS);
 223     wallcycle_stop(wcycle_, ewcNB_XF_BUF_OPS);
 224 }
 225
 226 real nonbonded_verlet_t::pairlistInnerRadius() const
 227 {
 228     return pairlistSets_->params().rlistInner;
 229 }
 230
 231 real nonbonded_verlet_t::pairlistOuterRadius() const
 232 {
 233     return pairlistSets_->params().rlistOuter;
 234 }
 235
 236 void nonbonded_verlet_t::changePairlistRadii(real rlistOuter, real rlistInner)
 237 {
 238     pairlistSets_->changePairlistRadii(rlistOuter, rlistInner);
 239 }
 240
 241 void nonbonded_verlet_t::setupGpuShortRangeWork(const gmx::GpuBonded*          gpuBonded,
 242                                                 const gmx::InteractionLocality iLocality)
 243 {
 244     if (useGpu() && !emulateGpu())
 245     {
 246         Nbnxm::setupGpuShortRangeWork(gpu_nbv, gpuBonded, iLocality);
 247     }
 248 }
 249
 250 void nonbonded_verlet_t::atomdata_init_copy_x_to_nbat_x_gpu()
 251 {
 252     Nbnxm::nbnxn_gpu_init_x_to_nbat_x(pairSearch_->gridSet(), gpu_nbv);
 253 }
 254
 255 void nonbonded_verlet_t::insertNonlocalGpuDependency(const gmx::InteractionLocality interactionLocality)
 256 {
 257     Nbnxm::nbnxnInsertNonlocalGpuDependency(gpu_nbv, interactionLocality);
 258 }
 259
 260 /*! \endcond */