src/gromacs/nbnxm/sycl/nbnxm_sycl_data_mgmt.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2020,2021, 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  *  Stubs of functions that must be defined by nbnxm sycl implementation.
  39  *
  40  *  \ingroup module_nbnxm
  41  */
  42 #include "gmxpre.h"
  43
  44 #include "gromacs/gpu_utils/device_stream_manager.h"
  45 #include "gromacs/gpu_utils/pmalloc.h"
  46 #include "gromacs/hardware/device_information.h"
  47 #include "gromacs/mdtypes/interaction_const.h"
  48 #include "gromacs/nbnxm/atomdata.h"
  49 #include "gromacs/nbnxm/gpu_data_mgmt.h"
  50 #include "gromacs/nbnxm/nbnxm_gpu.h"
  51 #include "gromacs/nbnxm/nbnxm_gpu_data_mgmt.h"
  52 #include "gromacs/pbcutil/ishift.h"
  53 #include "gromacs/utility/exceptions.h"
  54
  55 #include "nbnxm_sycl_types.h"
  56
  57 namespace Nbnxm
  58 {
  59
  60 /*! \brief Initialize \p atomdata first time; it only gets filled at pair-search. */
  61 static void initAtomdataFirst(NBAtomData*          atomdata,
  62                               int                  numTypes,
  63                               const DeviceContext& deviceContext,
  64                               const DeviceStream&  localStream)
  65 {
  66     atomdata->numTypes = numTypes;
  67     allocateDeviceBuffer(&atomdata->shiftVec, SHIFTS, deviceContext);
  68     atomdata->shiftVecUploaded = false;
  69
  70     allocateDeviceBuffer(&atomdata->fShift, SHIFTS, deviceContext);
  71     allocateDeviceBuffer(&atomdata->eLJ, 1, deviceContext);
  72     allocateDeviceBuffer(&atomdata->eElec, 1, deviceContext);
  73
  74     clearDeviceBufferAsync(&atomdata->fShift, 0, SHIFTS, localStream);
  75     clearDeviceBufferAsync(&atomdata->eElec, 0, 1, localStream);
  76     clearDeviceBufferAsync(&atomdata->eLJ, 0, 1, localStream);
  77
  78     /* initialize to nullptr pointers to data that is not allocated here and will
  79        need reallocation in later */
  80     atomdata->xq = nullptr;
  81     atomdata->f  = nullptr;
  82
  83     /* size -1 indicates that the respective array hasn't been initialized yet */
  84     atomdata->numAtoms      = -1;
  85     atomdata->numAtomsAlloc = -1;
  86 }
  87
  88 /*! \brief Initialize the nonbonded parameter data structure. */
  89 static void initNbparam(NBParamGpu*                     nbp,
  90                         const interaction_const_t&      ic,
  91                         const PairlistParams&           listParams,
  92                         const nbnxn_atomdata_t::Params& nbatParams,
  93                         const DeviceContext&            deviceContext)
  94 {
  95     const int numTypes = nbatParams.numTypes;
  96
  97     set_cutoff_parameters(nbp, &ic, listParams);
  98
  99     nbp->vdwType  = nbnxmGpuPickVdwKernelType(&ic, nbatParams.ljCombinationRule);
 100     nbp->elecType = nbnxmGpuPickElectrostaticsKernelType(&ic, deviceContext.deviceInfo());
 101
 102     /* generate table for PME */
 103     nbp->coulomb_tab = nullptr;
 104     if (nbp->elecType == ElecType::EwaldTab || nbp->elecType == ElecType::EwaldTabTwin)
 105     {
 106         GMX_RELEASE_ASSERT(ic.coulombEwaldTables, "Need valid Coulomb Ewald correction tables");
 107         init_ewald_coulomb_force_table(*ic.coulombEwaldTables, nbp, deviceContext);
 108     }
 109
 110     /* set up LJ parameter lookup table */
 111     if (!useLjCombRule(nbp->vdwType))
 112     {
 113         static_assert(sizeof(decltype(nbp->nbfp)) == 2 * sizeof(decltype(*nbatParams.nbfp.data())),
 114                       "Mismatch in the size of host / device data types");
 115         initParamLookupTable(&nbp->nbfp,
 116                              &nbp->nbfp_texobj,
 117                              reinterpret_cast<const Float2*>(nbatParams.nbfp.data()),
 118                              numTypes * numTypes,
 119                              deviceContext);
 120     }
 121
 122     /* set up LJ-PME parameter lookup table */
 123     if (ic.vdwtype == VanDerWaalsType::Pme)
 124     {
 125         static_assert(sizeof(decltype(nbp->nbfp_comb))
 126                               == 2 * sizeof(decltype(*nbatParams.nbfp_comb.data())),
 127                       "Mismatch in the size of host / device data types");
 128         initParamLookupTable(&nbp->nbfp_comb,
 129                              &nbp->nbfp_comb_texobj,
 130                              reinterpret_cast<const Float2*>(nbatParams.nbfp_comb.data()),
 131                              numTypes,
 132                              deviceContext);
 133     }
 134 }
 135
 136 NbnxmGpu* gpu_init(const gmx::DeviceStreamManager& deviceStreamManager,
 137                    const interaction_const_t*      ic,
 138                    const PairlistParams&           listParams,
 139                    const nbnxn_atomdata_t*         nbat,
 140                    const bool                      bLocalAndNonlocal)
 141 {
 142     auto* nb                              = new NbnxmGpu();
 143     nb->deviceContext_                    = &deviceStreamManager.context();
 144     nb->atdat                             = new NBAtomData;
 145     nb->nbparam                           = new NBParamGpu;
 146     nb->plist[InteractionLocality::Local] = new Nbnxm::gpu_plist;
 147     if (bLocalAndNonlocal)
 148     {
 149         nb->plist[InteractionLocality::NonLocal] = new Nbnxm::gpu_plist;
 150     }
 151
 152     nb->bUseTwoStreams = bLocalAndNonlocal;
 153
 154     nb->timers  = nullptr;
 155     nb->timings = nullptr;
 156
 157     /* init nbst */
 158     pmalloc(reinterpret_cast<void**>(&nb->nbst.eLJ), sizeof(*nb->nbst.eLJ));
 159     pmalloc(reinterpret_cast<void**>(&nb->nbst.eElec), sizeof(*nb->nbst.eElec));
 160     pmalloc(reinterpret_cast<void**>(&nb->nbst.fShift), SHIFTS * sizeof(*nb->nbst.fShift));
 161
 162     init_plist(nb->plist[InteractionLocality::Local]);
 163
 164     /* local/non-local GPU streams */
 165     GMX_RELEASE_ASSERT(deviceStreamManager.streamIsValid(gmx::DeviceStreamType::NonBondedLocal),
 166                        "Local non-bonded stream should be initialized to use GPU for non-bonded.");
 167     const DeviceStream& localStream = deviceStreamManager.stream(gmx::DeviceStreamType::NonBondedLocal);
 168     nb->deviceStreams[InteractionLocality::Local] = &localStream;
 169     // In general, it's not strictly necessary to use 2 streams for SYCL, since they are
 170     // out-of-order. But for the time being, it will be less disruptive to keep them.
 171     if (nb->bUseTwoStreams)
 172     {
 173         init_plist(nb->plist[InteractionLocality::NonLocal]);
 174
 175         GMX_RELEASE_ASSERT(deviceStreamManager.streamIsValid(gmx::DeviceStreamType::NonBondedNonLocal),
 176                            "Non-local non-bonded stream should be initialized to use GPU for "
 177                            "non-bonded with domain decomposition.");
 178         nb->deviceStreams[InteractionLocality::NonLocal] =
 179                 &deviceStreamManager.stream(gmx::DeviceStreamType::NonBondedNonLocal);
 180     }
 181
 182     nb->bDoTime = false;
 183
 184     const nbnxn_atomdata_t::Params& nbatParams    = nbat->params();
 185     const DeviceContext&            deviceContext = *nb->deviceContext_;
 186
 187     initNbparam(nb->nbparam, *ic, listParams, nbatParams, deviceContext);
 188     initAtomdataFirst(nb->atdat, nbatParams.numTypes, deviceContext, localStream);
 189
 190     return nb;
 191 }
 192
 193 void gpu_upload_shiftvec(NbnxmGpu* nb, const nbnxn_atomdata_t* nbatom)
 194 {
 195     NBAtomData*         adat        = nb->atdat;
 196     const DeviceStream& localStream = *nb->deviceStreams[InteractionLocality::Local];
 197
 198     /* only if we have a dynamic box */
 199     if (nbatom->bDynamicBox || !adat->shiftVecUploaded)
 200     {
 201         GMX_ASSERT(adat->shiftVec.elementSize() == sizeof(nbatom->shift_vec[0]),
 202                    "Sizes of host- and device-side shift vectors should be the same.");
 203         copyToDeviceBuffer(&adat->shiftVec,
 204                            reinterpret_cast<const Float3*>(nbatom->shift_vec.data()),
 205                            0,
 206                            SHIFTS,
 207                            localStream,
 208                            GpuApiCallBehavior::Async,
 209                            nullptr);
 210         adat->shiftVecUploaded = true;
 211     }
 212 }
 213
 214 void gpu_free(NbnxmGpu* nb)
 215 {
 216     if (nb == nullptr)
 217     {
 218         return;
 219     }
 220
 221     NBAtomData* atdat   = nb->atdat;
 222     NBParamGpu* nbparam = nb->nbparam;
 223
 224     if ((!nbparam->coulomb_tab)
 225         && (nbparam->elecType == ElecType::EwaldTab || nbparam->elecType == ElecType::EwaldTabTwin))
 226     {
 227         destroyParamLookupTable(&nbparam->coulomb_tab, nbparam->coulomb_tab_texobj);
 228     }
 229
 230     if (!useLjCombRule(nb->nbparam->vdwType))
 231     {
 232         destroyParamLookupTable(&nbparam->nbfp, nbparam->nbfp_texobj);
 233     }
 234
 235     if (nbparam->vdwType == VdwType::EwaldGeom || nbparam->vdwType == VdwType::EwaldLB)
 236     {
 237         destroyParamLookupTable(&nbparam->nbfp_comb, nbparam->nbfp_comb_texobj);
 238     }
 239
 240     /* Free plist */
 241     auto* plist = nb->plist[InteractionLocality::Local];
 242     delete plist;
 243     if (nb->bUseTwoStreams)
 244     {
 245         auto* plist_nl = nb->plist[InteractionLocality::NonLocal];
 246         delete plist_nl;
 247     }
 248
 249     /* Free nbst */
 250     pfree(nb->nbst.eLJ);
 251     nb->nbst.eLJ = nullptr;
 252
 253     pfree(nb->nbst.eElec);
 254     nb->nbst.eElec = nullptr;
 255
 256     pfree(nb->nbst.fShift);
 257     nb->nbst.fShift = nullptr;
 258
 259     delete atdat;
 260     delete nbparam;
 261     delete nb;
 262 }
 263
 264 int gpu_min_ci_balanced(NbnxmGpu* nb)
 265 {
 266     // SYCL-TODO: Logic and magic values taken from OpenCL
 267     static constexpr unsigned int balancedFactor = 50;
 268     if (nb == nullptr)
 269     {
 270         return 0;
 271     }
 272     const cl::sycl::device device = nb->deviceContext_->deviceInfo().syclDevice;
 273     const int numComputeUnits     = device.get_info<cl::sycl::info::device::max_compute_units>();
 274     return balancedFactor * numComputeUnits;
 275 }
 276
 277 } // namespace Nbnxm