src/gromacs/mdlib/gpuforcereduction_impl.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 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 backend-agnostic GPU Force Reduction functions
  38  *
  39  * \author Alan Gray <alang@nvidia.com>
  40  *
  41  * \ingroup module_mdlib
  42  */
  43
  44 #include "gmxpre.h"
  45
  46 #include "gpuforcereduction_impl.h"
  47
  48 #include "gromacs/gpu_utils/device_stream.h"
  49 #include "gromacs/gpu_utils/devicebuffer.h"
  50 #if GMX_GPU_CUDA
  51 #    include "gromacs/gpu_utils/gpueventsynchronizer.cuh"
  52 #elif GMX_GPU_SYCL
  53 #    include "gromacs/gpu_utils/gpueventsynchronizer_sycl.h"
  54 #endif
  55 #include "gromacs/mdlib/gpuforcereduction_impl_internal.h"
  56 #include "gromacs/utility/gmxassert.h"
  57
  58 namespace gmx
  59 {
  60
  61 GpuForceReduction::Impl::Impl(const DeviceContext& deviceContext,
  62                               const DeviceStream&  deviceStream,
  63                               gmx_wallcycle*       wcycle) :
  64     baseForce_(),
  65     deviceContext_(deviceContext),
  66     deviceStream_(deviceStream),
  67     nbnxmForceToAdd_(),
  68     rvecForceToAdd_(),
  69     wcycle_(wcycle)
  70 {
  71 }
  72
  73 void GpuForceReduction::Impl::reinit(DeviceBuffer<Float3>  baseForcePtr,
  74                                      const int             numAtoms,
  75                                      ArrayRef<const int>   cell,
  76                                      const int             atomStart,
  77                                      const bool            accumulate,
  78                                      GpuEventSynchronizer* completionMarker)
  79 {
  80     GMX_ASSERT(baseForcePtr, "Input base force for reduction has no data");
  81     baseForce_        = baseForcePtr;
  82     numAtoms_         = numAtoms;
  83     atomStart_        = atomStart;
  84     accumulate_       = accumulate;
  85     completionMarker_ = completionMarker;
  86     cellInfo_.cell    = cell.data();
  87
  88     wallcycle_start_nocount(wcycle_, WallCycleCounter::LaunchGpu);
  89     reallocateDeviceBuffer(
  90             &cellInfo_.d_cell, numAtoms_, &cellInfo_.cellSize, &cellInfo_.cellSizeAlloc, deviceContext_);
  91     copyToDeviceBuffer(&cellInfo_.d_cell,
  92                        &(cellInfo_.cell[atomStart]),
  93                        0,
  94                        numAtoms_,
  95                        deviceStream_,
  96                        GpuApiCallBehavior::Async,
  97                        nullptr);
  98     wallcycle_stop(wcycle_, WallCycleCounter::LaunchGpu);
  99
 100     dependencyList_.clear();
 101 };
 102
 103 void GpuForceReduction::Impl::registerNbnxmForce(DeviceBuffer<RVec> forcePtr)
 104 {
 105     GMX_ASSERT(forcePtr, "Input force for reduction has no data");
 106     nbnxmForceToAdd_ = forcePtr;
 107 };
 108
 109 void GpuForceReduction::Impl::registerRvecForce(DeviceBuffer<RVec> forcePtr)
 110 {
 111     GMX_ASSERT(forcePtr, "Input force for reduction has no data");
 112     rvecForceToAdd_ = forcePtr;
 113 };
 114
 115 void GpuForceReduction::Impl::addDependency(GpuEventSynchronizer* dependency)
 116 {
 117     dependencyList_.push_back(dependency);
 118 }
 119
 120 void GpuForceReduction::Impl::execute()
 121 {
 122     wallcycle_start_nocount(wcycle_, WallCycleCounter::LaunchGpu);
 123     wallcycle_sub_start(wcycle_, WallCycleSubCounter::LaunchGpuNBFBufOps);
 124
 125     if (numAtoms_ == 0)
 126     {
 127         return;
 128     }
 129
 130     GMX_ASSERT(nbnxmForceToAdd_, "Nbnxm force for reduction has no data");
 131
 132     // Enqueue wait on all dependencies passed
 133     for (auto* synchronizer : dependencyList_)
 134     {
 135         synchronizer->enqueueWaitEvent(deviceStream_);
 136     }
 137
 138     const bool addRvecForce = static_cast<bool>(rvecForceToAdd_); // True iff initialized
 139
 140     launchForceReductionKernel(numAtoms_,
 141                                atomStart_,
 142                                addRvecForce,
 143                                accumulate_,
 144                                nbnxmForceToAdd_,
 145                                rvecForceToAdd_,
 146                                baseForce_,
 147                                cellInfo_.d_cell,
 148                                deviceStream_);
 149
 150     // Mark that kernel has been launched
 151     if (completionMarker_ != nullptr)
 152     {
 153         completionMarker_->markEvent(deviceStream_);
 154     }
 155
 156     wallcycle_sub_stop(wcycle_, WallCycleSubCounter::LaunchGpuNBFBufOps);
 157     wallcycle_stop(wcycle_, WallCycleCounter::LaunchGpu);
 158 }
 159
 160 GpuForceReduction::GpuForceReduction(const DeviceContext& deviceContext,
 161                                      const DeviceStream&  deviceStream,
 162                                      gmx_wallcycle*       wcycle) :
 163     impl_(new Impl(deviceContext, deviceStream, wcycle))
 164 {
 165 }
 166
 167 void GpuForceReduction::registerNbnxmForce(DeviceBuffer<RVec> forcePtr)
 168 {
 169     impl_->registerNbnxmForce(forcePtr);
 170 }
 171
 172 void GpuForceReduction::registerRvecForce(DeviceBuffer<RVec> forcePtr)
 173 {
 174     impl_->registerRvecForce(forcePtr);
 175 }
 176
 177 void GpuForceReduction::addDependency(GpuEventSynchronizer* dependency)
 178 {
 179     impl_->addDependency(dependency);
 180 }
 181
 182 void GpuForceReduction::reinit(DeviceBuffer<RVec>    baseForcePtr,
 183                                const int             numAtoms,
 184                                ArrayRef<const int>   cell,
 185                                const int             atomStart,
 186                                const bool            accumulate,
 187                                GpuEventSynchronizer* completionMarker)
 188 {
 189     impl_->reinit(baseForcePtr, numAtoms, cell, atomStart, accumulate, completionMarker);
 190 }
 191 void GpuForceReduction::execute()
 192 {
 193     impl_->execute();
 194 }
 195
 196 GpuForceReduction::~GpuForceReduction() = default;
 197
 198 } // namespace gmx