src/gromacs/domdec/gpuhaloexchange.h

   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 /*! \libinternal \file
  36  * \brief Declaration of GPU halo exchange.
  37  *
  38  * \author Alan Gray <alang@nvidia.com>
  39  * \inlibraryapi
  40  * \ingroup module_domdec
  41  */
  42 #ifndef GMX_DOMDEC_GPUHALOEXCHANGE_H
  43 #define GMX_DOMDEC_GPUHALOEXCHANGE_H
  44
  45 #include "gromacs/gpu_utils/devicebuffer_datatype.h"
  46 #include "gromacs/math/vectypes.h"
  47 #include "gromacs/utility/basedefinitions.h"
  48 #include "gromacs/utility/classhelpers.h"
  49 #include "gromacs/utility/gmxmpi.h"
  50
  51 struct gmx_domdec_t;
  52 struct gmx_wallcycle;
  53 class DeviceContext;
  54 class DeviceStream;
  55 class GpuEventSynchronizer;
  56
  57 namespace gmx
  58 {
  59
  60 /*! \libinternal
  61  * \brief Manages GPU Halo Exchange object */
  62 class GpuHaloExchange
  63 {
  64
  65 public:
  66     /*! \brief Creates GPU Halo Exchange object.
  67      *
  68      * Coordinate Halo exchange will be performed in \c
  69      * StreamNonLocal, and the \c communicateHaloCoordinates
  70      * method must be called before any subsequent operations that
  71      * access non-local parts of the coordinate buffer (such as
  72      * the non-local non-bonded kernels). It also must be called
  73      * after the local coordinates buffer operations (where the
  74      * coordinates are copied to the device and hence the \c
  75      * coordinatesReadyOnDeviceEvent is recorded). Force Halo exchange
  76      * will be performed in \c streamNonLocal (also potentally
  77      * with buffer clearing in \c streamLocal)and the \c
  78      * communicateHaloForces method must be called after the
  79      * non-local buffer operations, after the local force buffer
  80      * has been copied to the GPU (if CPU forces are present), and
  81      * before the local buffer operations. The force halo exchange
  82      * does not yet support virial steps.
  83      *
  84      * \param [inout] dd                       domdec structure
  85      * \param [in]    mpi_comm_mysim           communicator used for simulation
  86      * \param [in]    deviceContext            GPU device context
  87      * \param [in]    streamLocal              local NB CUDA stream.
  88      * \param [in]    streamNonLocal           non-local NB CUDA stream.
  89      * \param [in]    pulse                    the communication pulse for this instance
  90      * \param [in]    wcycle                   The wallclock counter
  91      */
  92     GpuHaloExchange(gmx_domdec_t*        dd,
  93                     MPI_Comm             mpi_comm_mysim,
  94                     const DeviceContext& deviceContext,
  95                     const DeviceStream&  streamLocal,
  96                     const DeviceStream&  streamNonLocal,
  97                     int                  pulse,
  98                     gmx_wallcycle*       wcycle);
  99     ~GpuHaloExchange();
 100
 101     /*! \brief
 102      *
 103      * Initialization for GPU halo exchange of coordinates buffer
 104      * \param [in] d_coordinateBuffer   pointer to coordinates buffer in GPU memory
 105      * \param [in] d_forcesBuffer   pointer to coordinates buffer in GPU memory
 106      */
 107     void reinitHalo(DeviceBuffer<RVec> d_coordinateBuffer, DeviceBuffer<RVec> d_forcesBuffer);
 108
 109
 110     /*! \brief GPU halo exchange of coordinates buffer.
 111      *
 112      * Must be called after local setCoordinates (which records an
 113      * event when the coordinate data has been copied to the
 114      * device).
 115      * \param [in] box  Coordinate box (from which shifts will be constructed)
 116      * \param [in] coordinatesReadyOnDeviceEvent event recorded when coordinates have been copied to device
 117      */
 118     void communicateHaloCoordinates(const matrix box, GpuEventSynchronizer* coordinatesReadyOnDeviceEvent);
 119
 120     /*! \brief GPU halo exchange of force buffer.
 121      * \param[in] accumulateForces  True if forces should accumulate, otherwise they are set
 122      */
 123     void communicateHaloForces(bool accumulateForces);
 124
 125     /*! \brief Get the event synchronizer for the forces ready on device.
 126      *  \returns  The event to synchronize the stream that consumes forces on device.
 127      */
 128     GpuEventSynchronizer* getForcesReadyOnDeviceEvent();
 129
 130 private:
 131     class Impl;
 132     gmx::PrivateImplPointer<Impl> impl_;
 133 };
 134
 135 } // namespace gmx
 136
 137 #endif