src/gromacs/ewald/pme_pp_comm_gpu_impl.cu

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   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.
   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  *
  37  * \brief Implements PME-PP communication using CUDA
  38  *
  39  *
  40  * \author Alan Gray <alang@nvidia.com>
  41  *
  42  * \ingroup module_ewald
  43  */
  44 #include "gmxpre.h"
  45
  46 #include "gromacs/ewald/pme_pp_communication.h"
  47 #include "pme_pp_comm_gpu_impl.h"
  48
  49 #include "config.h"
  50
  51 #include "gromacs/gpu_utils/cudautils.cuh"
  52 #include "gromacs/gpu_utils/device_context.h"
  53 #include "gromacs/gpu_utils/device_stream.h"
  54 #include "gromacs/gpu_utils/devicebuffer.h"
  55 #include "gromacs/gpu_utils/gpueventsynchronizer.cuh"
  56 #include "gromacs/gpu_utils/typecasts.cuh"
  57 #include "gromacs/utility/gmxmpi.h"
  58
  59 namespace gmx
  60 {
  61
  62 PmePpCommGpu::Impl::Impl(MPI_Comm             comm,
  63                          int                  pmeRank,
  64                          const DeviceContext& deviceContext,
  65                          const DeviceStream&  deviceStream) :
  66     deviceContext_(deviceContext),
  67     pmePpCommStream_(deviceStream),
  68     comm_(comm),
  69     pmeRank_(pmeRank),
  70     d_pmeForces_(nullptr)
  71 {
  72 }
  73
  74 PmePpCommGpu::Impl::~Impl() = default;
  75
  76 void PmePpCommGpu::Impl::reinit(int size)
  77 {
  78     // This rank will access PME rank memory directly, so needs to receive the remote PME buffer addresses.
  79 #if GMX_MPI
  80
  81     if (GMX_THREAD_MPI)
  82     {
  83         // receive device buffer address from PME rank
  84         MPI_Recv(&remotePmeXBuffer_, sizeof(float3*), MPI_BYTE, pmeRank_, 0, comm_, MPI_STATUS_IGNORE);
  85         MPI_Recv(&remotePmeFBuffer_, sizeof(float3*), MPI_BYTE, pmeRank_, 0, comm_, MPI_STATUS_IGNORE);
  86     }
  87
  88 #endif
  89
  90     // Reallocate buffer used for staging PME force on GPU
  91     reallocateDeviceBuffer(&d_pmeForces_, size, &d_pmeForcesSize_, &d_pmeForcesSizeAlloc_, deviceContext_);
  92 }
  93
  94 void PmePpCommGpu::Impl::receiveForceFromPmeCudaDirect(float3* pmeForcePtr, int recvSize, bool receivePmeForceToGpu)
  95 {
  96 #if GMX_MPI
  97     // Receive event from PME task and add to stream, to ensure pull of data doesn't
  98     // occur before PME force calc is completed
  99     GpuEventSynchronizer* pmeSync;
 100     // NOLINTNEXTLINE(bugprone-sizeof-expression)
 101     MPI_Recv(&pmeSync, sizeof(GpuEventSynchronizer*), MPI_BYTE, pmeRank_, 0, comm_, MPI_STATUS_IGNORE);
 102     pmeSync->enqueueWaitEvent(pmePpCommStream_);
 103 #endif
 104
 105     // Pull force data from remote GPU
 106     cudaError_t stat = cudaMemcpyAsync(pmeForcePtr,
 107                                        remotePmeFBuffer_,
 108                                        recvSize * DIM * sizeof(float),
 109                                        cudaMemcpyDefault,
 110                                        pmePpCommStream_.stream());
 111     CU_RET_ERR(stat, "cudaMemcpyAsync on Recv from PME CUDA direct data transfer failed");
 112
 113     if (receivePmeForceToGpu)
 114     {
 115         // Record event to be enqueued in the GPU local buffer operations, to
 116         // satisfy dependency on receiving the PME force data before
 117         // reducing it with the other force contributions.
 118         forcesReadySynchronizer_.markEvent(pmePpCommStream_);
 119     }
 120     else
 121     {
 122         // Ensure CPU waits for PME forces to be copied before reducing
 123         // them with other forces on the CPU
 124         pmePpCommStream_.synchronize();
 125     }
 126 }
 127
 128 void PmePpCommGpu::Impl::receiveForceFromPmeCudaMpi(float3* pmeForcePtr, int recvSize)
 129 {
 130 #if GMX_MPI
 131     MPI_Recv(pmeForcePtr, recvSize * DIM, MPI_FLOAT, pmeRank_, 0, comm_, MPI_STATUS_IGNORE);
 132 #else
 133     GMX_UNUSED_VALUE(pmeForcePtr);
 134     GMX_UNUSED_VALUE(recvSize);
 135 #endif
 136 }
 137
 138 void PmePpCommGpu::Impl::receiveForceFromPme(float3* recvPtr, int recvSize, bool receivePmeForceToGpu)
 139 {
 140     float3* pmeForcePtr = receivePmeForceToGpu ? asFloat3(d_pmeForces_) : recvPtr;
 141     if (GMX_THREAD_MPI)
 142     {
 143         receiveForceFromPmeCudaDirect(pmeForcePtr, recvSize, receivePmeForceToGpu);
 144     }
 145     else
 146     {
 147         receiveForceFromPmeCudaMpi(pmeForcePtr, recvSize);
 148     }
 149 }
 150
 151 void PmePpCommGpu::Impl::sendCoordinatesToPmeCudaDirect(float3*               sendPtr,
 152                                                         int                   sendSize,
 153                                                         GpuEventSynchronizer* coordinatesReadyOnDeviceEvent)
 154 {
 155     // ensure stream waits until coordinate data is available on device
 156     coordinatesReadyOnDeviceEvent->enqueueWaitEvent(pmePpCommStream_);
 157
 158     cudaError_t stat = cudaMemcpyAsync(remotePmeXBuffer_,
 159                                        sendPtr,
 160                                        sendSize * DIM * sizeof(float),
 161                                        cudaMemcpyDefault,
 162                                        pmePpCommStream_.stream());
 163     CU_RET_ERR(stat, "cudaMemcpyAsync on Send to PME CUDA direct data transfer failed");
 164
 165 #if GMX_MPI
 166     // Record and send event to allow PME task to sync to above transfer before commencing force calculations
 167     pmeCoordinatesSynchronizer_.markEvent(pmePpCommStream_);
 168     GpuEventSynchronizer* pmeSync = &pmeCoordinatesSynchronizer_;
 169     // NOLINTNEXTLINE(bugprone-sizeof-expression)
 170     MPI_Send(&pmeSync, sizeof(GpuEventSynchronizer*), MPI_BYTE, pmeRank_, 0, comm_);
 171 #endif
 172 }
 173
 174 void PmePpCommGpu::Impl::sendCoordinatesToPmeCudaMpi(float3*               sendPtr,
 175                                                      int                   sendSize,
 176                                                      GpuEventSynchronizer* coordinatesReadyOnDeviceEvent)
 177 {
 178     // ensure coordinate data is available on device before we start transfer
 179     coordinatesReadyOnDeviceEvent->waitForEvent();
 180
 181 #if GMX_MPI
 182     float3* sendptr_x = sendPtr;
 183
 184     MPI_Send(sendptr_x, sendSize * DIM, MPI_FLOAT, pmeRank_, eCommType_COORD_GPU, comm_);
 185 #else
 186     GMX_UNUSED_VALUE(sendPtr);
 187     GMX_UNUSED_VALUE(sendSize);
 188 #endif
 189 }
 190
 191 void PmePpCommGpu::Impl::sendCoordinatesToPme(float3*               sendPtr,
 192                                               int                   sendSize,
 193                                               GpuEventSynchronizer* coordinatesReadyOnDeviceEvent)
 194 {
 195     if (GMX_THREAD_MPI)
 196     {
 197         sendCoordinatesToPmeCudaDirect(sendPtr, sendSize, coordinatesReadyOnDeviceEvent);
 198     }
 199     else
 200     {
 201         sendCoordinatesToPmeCudaMpi(sendPtr, sendSize, coordinatesReadyOnDeviceEvent);
 202     }
 203 }
 204 DeviceBuffer<Float3> PmePpCommGpu::Impl::getGpuForceStagingPtr()
 205 {
 206     return d_pmeForces_;
 207 }
 208
 209 GpuEventSynchronizer* PmePpCommGpu::Impl::getForcesReadySynchronizer()
 210 {
 211     if (GMX_THREAD_MPI)
 212     {
 213         return &forcesReadySynchronizer_;
 214     }
 215     else
 216     {
 217         return nullptr;
 218     }
 219 }
 220
 221 PmePpCommGpu::PmePpCommGpu(MPI_Comm             comm,
 222                            int                  pmeRank,
 223                            const DeviceContext& deviceContext,
 224                            const DeviceStream&  deviceStream) :
 225     impl_(new Impl(comm, pmeRank, deviceContext, deviceStream))
 226 {
 227 }
 228
 229 PmePpCommGpu::~PmePpCommGpu() = default;
 230
 231 void PmePpCommGpu::reinit(int size)
 232 {
 233     impl_->reinit(size);
 234 }
 235
 236 void PmePpCommGpu::receiveForceFromPme(RVec* recvPtr, int recvSize, bool receivePmeForceToGpu)
 237 {
 238     impl_->receiveForceFromPme(asFloat3(recvPtr), recvSize, receivePmeForceToGpu);
 239 }
 240
 241 void PmePpCommGpu::sendCoordinatesToPmeFromGpu(DeviceBuffer<RVec>    sendPtr,
 242                                                int                   sendSize,
 243                                                GpuEventSynchronizer* coordinatesReadyOnDeviceEvent)
 244 {
 245     impl_->sendCoordinatesToPme(asFloat3(sendPtr), sendSize, coordinatesReadyOnDeviceEvent);
 246 }
 247
 248 void PmePpCommGpu::sendCoordinatesToPmeFromCpu(RVec*                 sendPtr,
 249                                                int                   sendSize,
 250                                                GpuEventSynchronizer* coordinatesReadyOnDeviceEvent)
 251 {
 252     impl_->sendCoordinatesToPme(asFloat3(sendPtr), sendSize, coordinatesReadyOnDeviceEvent);
 253 }
 254
 255 DeviceBuffer<Float3> PmePpCommGpu::getGpuForceStagingPtr()
 256 {
 257     return impl_->getGpuForceStagingPtr();
 258 }
 259
 260 GpuEventSynchronizer* PmePpCommGpu::getForcesReadySynchronizer()
 261 {
 262     return impl_->getForcesReadySynchronizer();
 263 }
 264
 265 } // namespace gmx