2 * This file is part of the GROMACS molecular simulation package.
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.
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.
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.
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.
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.
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.
37 * \brief Declares CUDA implementation of GPU Halo Exchange.
39 * This header file is needed to include from both the device-side
40 * kernels file, and the host-side management code.
42 * \author Alan Gray <alang@nvidia.com>
44 * \ingroup module_domdec
46 #ifndef GMX_DOMDEC_GPUHALOEXCHANGE_IMPL_H
47 #define GMX_DOMDEC_GPUHALOEXCHANGE_IMPL_H
49 #include "gromacs/domdec/gpuhaloexchange.h"
50 #include "gromacs/gpu_utils/device_context.h"
51 #include "gromacs/gpu_utils/gpueventsynchronizer.cuh"
52 #include "gromacs/gpu_utils/hostallocator.h"
53 #include "gromacs/utility/gmxmpi.h"
60 /*! \brief switch for whether coordinates or force halo is being applied */
61 enum class HaloQuantity
67 /*! \internal \brief Class with interfaces and data for GPU Halo Exchange */
68 class GpuHaloExchange::Impl
72 /*! \brief Creates GPU Halo Exchange object.
74 * \param [inout] dd domdec structure
75 * \param [in] dimIndex the dimension index for this instance
76 * \param [in] mpi_comm_mysim communicator used for simulation
77 * \param [in] deviceContext GPU device context
78 * \param [in] localStream local NB CUDA stream
79 * \param [in] nonLocalStream non-local NB CUDA stream
80 * \param [in] pulse the communication pulse for this instance
81 * \param [in] wcycle The wallclock counter
83 Impl(gmx_domdec_t* dd,
85 MPI_Comm mpi_comm_mysim,
86 const DeviceContext& deviceContext,
87 const DeviceStream& localStream,
88 const DeviceStream& nonLocalStream,
90 gmx_wallcycle* wcycle);
94 * (Re-) Initialization for GPU halo exchange
95 * \param [in] d_coordinatesBuffer pointer to coordinates buffer in GPU memory
96 * \param [in] d_forcesBuffer pointer to forces buffer in GPU memory
98 void reinitHalo(float3* d_coordinatesBuffer, float3* d_forcesBuffer);
102 * GPU halo exchange of coordinates buffer
103 * \param [in] box Coordinate box (from which shifts will be constructed)
104 * \param [in] coordinatesReadyOnDeviceEvent event recorded when coordinates have been copied to device
106 void communicateHaloCoordinates(const matrix box, GpuEventSynchronizer* coordinatesReadyOnDeviceEvent);
108 /*! \brief GPU halo exchange of force buffer
109 * \param[in] accumulateForces True if forces should accumulate, otherwise they are set
111 void communicateHaloForces(bool accumulateForces);
113 /*! \brief Get the event synchronizer for the forces ready on device.
114 * \returns The event to synchronize the stream that consumes forces on device.
116 GpuEventSynchronizer* getForcesReadyOnDeviceEvent();
119 /*! \brief Data transfer wrapper for GPU halo exchange
120 * \param [inout] d_ptr pointer to coordinates or force buffer in GPU memory
121 * \param [in] haloQuantity switch on whether X or F halo exchange is being performed
122 * \param [in] coordinatesReadyOnDeviceEvent event recorded when coordinates have been copied to device
124 void communicateHaloData(float3* d_ptr,
125 HaloQuantity haloQuantity,
126 GpuEventSynchronizer* coordinatesReadyOnDeviceEvent);
128 /*! \brief Data transfer for GPU halo exchange using CUDA memcopies
129 * \param [inout] sendPtr address to send data from
130 * \param [in] sendSize number of atoms to be sent
131 * \param [in] sendRank rank to send data to
132 * \param [inout] remotePtr remote address to recv data
133 * \param [in] recvRank rank to recv data from
135 void communicateHaloDataWithCudaDirect(void* sendPtr, int sendSize, int sendRank, void* remotePtr, int recvRank);
137 //! Domain decomposition object
138 gmx_domdec_t* dd_ = nullptr;
139 //! map of indices to be sent from this rank
140 gmx::HostVector<int> h_indexMap_;
141 //! device copy of index map
142 int* d_indexMap_ = nullptr;
143 //! number of elements in index map array
144 int indexMapSize_ = -1;
145 //! number of elements allocated in index map array
146 int indexMapSizeAlloc_ = -1;
147 //! device buffer for sending packed data
148 float3* d_sendBuf_ = nullptr;
149 //! number of atoms in sendbuf array
150 int sendBufSize_ = -1;
151 //! number of atoms allocated in sendbuf array
152 int sendBufSizeAlloc_ = -1;
153 //! device buffer for receiving packed data
154 float3* d_recvBuf_ = nullptr;
155 //! maximum size of packed buffer
156 int maxPackedBufferSize_ = 0;
157 //! number of atoms in recvbuf array
158 int recvBufSize_ = -1;
159 //! number of atoms allocated in recvbuf array
160 int recvBufSizeAlloc_ = -1;
161 //! rank to send data to for X
163 //! rank to recv data from for X
165 //! rank to send data to for F
167 //! rank to recv data from for F
169 //! send copy size from this rank for X
171 //! recv copy size to this rank for X
173 //! send copy size from this rank for F
175 //! recv copy size to this rank for F
177 //! number of home atoms - offset of local halo region
178 int numHomeAtoms_ = 0;
179 //! remote GPU coordinates buffer pointer for pushing data
180 void* remoteXPtr_ = nullptr;
181 //! remote GPU force buffer pointer for pushing data
182 void* remoteFPtr_ = nullptr;
183 //! Periodic Boundary Conditions for this rank
184 bool usePBC_ = false;
185 //! force shift buffer on device
186 float3* d_fShift_ = nullptr;
187 //! Event triggered when halo transfer has been launched with direct CUD memory copy
188 GpuEventSynchronizer* haloDataTransferLaunched_ = nullptr;
189 //! MPI communicator used for simulation
190 MPI_Comm mpi_comm_mysim_;
191 //! GPU context object
192 const DeviceContext& deviceContext_;
193 //! CUDA stream for local non-bonded calculations
194 const DeviceStream& localStream_;
195 //! CUDA stream for non-local non-bonded calculations
196 const DeviceStream& nonLocalStream_;
197 //! full coordinates buffer in GPU memory
198 float3* d_x_ = nullptr;
199 //! full forces buffer in GPU memory
200 float3* d_f_ = nullptr;
201 //! An event recorded once the exchanged forces are ready on the GPU
202 GpuEventSynchronizer fReadyOnDevice_;
203 //! The dimension index corresponding to this halo exchange instance
205 //! The pulse corresponding to this halo exchange instance
207 //! The wallclock counter
208 gmx_wallcycle* wcycle_ = nullptr;
209 //! The atom offset for receive (x) or send (f) for dimension index and pulse corresponding to this halo exchange instance