2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2019,2020,2021, 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.
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 Implements PME-PP communication using CUDA
40 * \author Alan Gray <alang@nvidia.com>
42 * \ingroup module_ewald
46 #include "pme_force_sender_gpu_impl.h"
50 #include "gromacs/gpu_utils/cudautils.cuh"
51 #include "gromacs/gpu_utils/gpueventsynchronizer.cuh"
52 #include "gromacs/utility/gmxmpi.h"
57 /*! \brief Create PME-PP GPU communication object */
58 PmeForceSenderGpu::Impl::Impl(GpuEventSynchronizer* pmeForcesReady,
60 gmx::ArrayRef<PpRanks> ppRanks) :
61 pmeForcesReady_(pmeForcesReady),
67 PmeForceSenderGpu::Impl::~Impl() = default;
69 /*! \brief sends force buffer address to PP ranks */
70 void PmeForceSenderGpu::Impl::sendForceBufferAddressToPpRanks(DeviceBuffer<Float3> d_f)
72 // Need to send address to PP rank only for thread-MPI as PP rank pulls
73 // data using cudamemcpy
81 for (const auto& receiver : ppRanks_)
84 ind_end = ind_start + receiver.numAtoms;
86 // Data will be transferred directly from GPU.
87 Float3* sendBuf = &d_f[ind_start];
89 MPI_Send(&sendBuf, sizeof(Float3*), MPI_BYTE, receiver.rankId, 0, comm_);
92 GMX_UNUSED_VALUE(d_f);
96 /*! \brief Send PME synchronizer directly using CUDA memory copy */
97 void PmeForceSenderGpu::Impl::sendFSynchronizerToPpCudaDirect(int ppRank)
99 GMX_ASSERT(GMX_THREAD_MPI,
100 "sendFSynchronizerToPpCudaDirect is expected to be called only for Thread-MPI");
102 // Data will be pulled directly from PP task
104 // TODO Using MPI_Isend would be more efficient, particularly when
105 // sending to multiple PP ranks
106 MPI_Send(&pmeForcesReady_, sizeof(GpuEventSynchronizer*), MPI_BYTE, ppRank, 0, comm_);
108 GMX_UNUSED_VALUE(ppRank);
112 /*! \brief Send PME data directly using CUDA-aware MPI */
113 void PmeForceSenderGpu::Impl::sendFToPpCudaMpi(DeviceBuffer<RVec> sendbuf,
117 MPI_Request* request)
119 GMX_ASSERT(GMX_LIB_MPI, "sendFToPpCudaMpi is expected to be called only for Lib-MPI");
122 // if using GPU direct comm with CUDA-aware MPI, make sure forces are ready on device
123 // before sending it to PP ranks
124 pmeForcesReady_->waitForEvent();
126 MPI_Isend(sendbuf[offset], numBytes, MPI_BYTE, ppRank, 0, comm_, request);
129 GMX_UNUSED_VALUE(sendbuf);
130 GMX_UNUSED_VALUE(offset);
131 GMX_UNUSED_VALUE(numBytes);
132 GMX_UNUSED_VALUE(ppRank);
133 GMX_UNUSED_VALUE(request);
137 PmeForceSenderGpu::PmeForceSenderGpu(GpuEventSynchronizer* pmeForcesReady,
139 gmx::ArrayRef<PpRanks> ppRanks) :
140 impl_(new Impl(pmeForcesReady, comm, ppRanks))
144 PmeForceSenderGpu::~PmeForceSenderGpu() = default;
146 void PmeForceSenderGpu::sendForceBufferAddressToPpRanks(DeviceBuffer<RVec> d_f)
148 impl_->sendForceBufferAddressToPpRanks(d_f);
151 void PmeForceSenderGpu::sendFSynchronizerToPpCudaDirect(int ppRank)
153 impl_->sendFSynchronizerToPpCudaDirect(ppRank);
156 void PmeForceSenderGpu::sendFToPpCudaMpi(DeviceBuffer<RVec> sendbuf,
160 MPI_Request* request)
162 impl_->sendFToPpCudaMpi(sendbuf, offset, numBytes, ppRank, request);