src/gromacs/listed_forces/gpubonded_impl.h

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2018,2019, 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 /*! \internal \file
  36  * \brief Declares GPU implementation class for CUDA bonded
  37  * interactions.
  38  *
  39  * This header file is needed to include from both the device-side
  40  * kernels file, and the host-side management code.
  41  *
  42  * \author Berk Hess <hess@kth.se>
  43  * \author Szilárd Páll <pall.szilard@gmail.com>
  44  * \author Mark Abraham <mark.j.abraham@gmail.com>
  45  *
  46  * \ingroup module_listed_forces
  47  */
  48 #ifndef GMX_LISTED_FORCES_GPUBONDED_IMPL_H
  49 #define GMX_LISTED_FORCES_GPUBONDED_IMPL_H
  50
  51 #include "gromacs/gpu_utils/gpu_vec.cuh"
  52 #include "gromacs/gpu_utils/gputraits.cuh"
  53 #include "gromacs/gpu_utils/hostallocator.h"
  54 #include "gromacs/listed_forces/gpubonded.h"
  55 #include "gromacs/pbcutil/pbc_aiuc.h"
  56
  57 struct gmx_ffparams_t;
  58 struct t_forcerec;
  59
  60 namespace gmx
  61 {
  62
  63 /*! \internal \brief Version of InteractionList that supports pinning */
  64 struct HostInteractionList
  65 {
  66     /*! \brief Returns the total number of elements in iatoms */
  67     int size() const
  68     {
  69         return iatoms.size();
  70     }
  71
  72     //! List of interactions, see \c HostInteractionLists
  73     HostVector<int> iatoms = {{}, gmx::HostAllocationPolicy(gmx::PinningPolicy::PinnedIfSupported)};
  74 };
  75
  76 /* \brief Bonded parameters and GPU pointers
  77  *
  78  * This is used to accumulate all the parameters and pointers so they can be passed
  79  * to the GPU as a single structure.
  80  *
  81  */
  82 struct BondedCudaKernelParameters
  83 {
  84     //! Periodic boundary data
  85     PbcAiuc             pbcAiuc;
  86     //! Scale factor
  87     float               scaleFactor;
  88     //! The bonded types on GPU
  89     int                 fTypesOnGpu[numFTypesOnGpu];
  90     //! The number of interaction atom (iatom) elements for every function type
  91     int                 numFTypeIAtoms[numFTypesOnGpu];
  92     //! The number of bonds for every function type
  93     int                 numFTypeBonds[numFTypesOnGpu];
  94     //! The start index in the range of each interaction type
  95     int                 fTypeRangeStart[numFTypesOnGpu];
  96     //! The end index in the range of each interaction type
  97     int                 fTypeRangeEnd[numFTypesOnGpu];
  98
  99     //! Force parameters (on GPU)
 100     t_iparams          *d_forceParams;
 101     //! Coordinates before the timestep (on GPU)
 102     const float4       *d_xq;
 103     //! Forces on atoms (on GPU)
 104     fvec               *d_f;
 105     //! Force shifts on atoms (on GPU)
 106     fvec               *d_fShift;
 107     //! Total Energy (on GPU)
 108     float              *d_vTot;
 109     //! Interaction list atoms (on GPU)
 110     t_iatom            *d_iatoms[numFTypesOnGpu];
 111
 112     BondedCudaKernelParameters()
 113     {
 114         matrix boxDummy = { {0, 0, 0}, {0, 0, 0}, {0, 0, 0} };
 115
 116         setPbcAiuc(0, boxDummy, &pbcAiuc);
 117
 118         scaleFactor   = 1.0;
 119         d_forceParams = nullptr;
 120         d_xq          = nullptr;
 121         d_f           = nullptr;
 122         d_fShift      = nullptr;
 123         d_vTot        = nullptr;
 124     }
 125 };
 126
 127 /*! \internal \brief Implements GPU bondeds */
 128 class GpuBonded::Impl
 129 {
 130     public:
 131         //! Constructor
 132         Impl(const gmx_ffparams_t &ffparams,
 133              void                 *streamPtr);
 134         /*! \brief Destructor, non-default needed for freeing
 135          * device-side buffers */
 136         ~Impl();
 137         /*! \brief Update lists of interactions from idef suitable for the GPU,
 138          * using the data structures prepared for PP work.
 139          *
 140          * Intended to be called after each neighbour search
 141          * stage. Copies the bonded interactions assigned to the GPU
 142          * to device data structures, and updates device buffers that
 143          * may have been updated after search. */
 144         void updateInteractionListsAndDeviceBuffers(ArrayRef<const int>  nbnxnAtomOrder,
 145                                                     const t_idef        &idef,
 146                                                     void                *xqDevice,
 147                                                     void                *forceDevice,
 148                                                     void                *fshiftDevice);
 149
 150         /*! \brief Launches bonded kernel on a GPU */
 151         template <bool calcVir, bool calcEner>
 152         void
 153         launchKernel(const t_forcerec *fr,
 154                      const matrix      box);
 155         /*! \brief Returns whether there are bonded interactions
 156          * assigned to the GPU */
 157         bool haveInteractions() const;
 158         /*! \brief Launches the transfer of computed bonded energies. */
 159         void launchEnergyTransfer();
 160         /*! \brief Waits on the energy transfer, and accumulates bonded energies to \c enerd. */
 161         void accumulateEnergyTerms(gmx_enerdata_t *enerd);
 162         /*! \brief Clears the device side energy buffer */
 163         void clearEnergies();
 164     private:
 165         /*! \brief The interaction lists
 166          *
 167          * \todo This is potentially several pinned allocations, which
 168          * could contribute to exhausting such pages. */
 169         std::array<HostInteractionList, F_NRE> iLists_;
 170
 171         //! Tells whether there are any interaction in iLists.
 172         bool                  haveInteractions_;
 173         //! Interaction lists on the device.
 174         t_ilist               d_iLists_[F_NRE];
 175         //! Bonded parameters for device-side use.
 176         t_iparams            *d_forceParams_ = nullptr;
 177         //! Position-charge vector on the device.
 178         const float4         *d_xq_ = nullptr;
 179         //! Force vector on the device.
 180         fvec                 *d_f_ = nullptr;
 181         //! Shift force vector on the device.
 182         fvec                 *d_fShift_ = nullptr;
 183         //! \brief Host-side virial buffer
 184         HostVector <float>    vTot_ = {{}, gmx::HostAllocationPolicy(gmx::PinningPolicy::PinnedIfSupported)};
 185         //! \brief Device-side total virial
 186         float                *d_vTot_   = nullptr;
 187
 188         //! \brief Bonded GPU stream, not owned by this module
 189         CommandStream         stream_;
 190
 191         //! Parameters and pointers, passed to the CUDA kernel
 192         BondedCudaKernelParameters kernelParams_;
 193 };
 194
 195 }   // namespace gmx
 196
 197 #endif