2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2019, 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 Definitions of interfaces for GPU state data propagator object.
39 * \author Artem Zhmurov <zhmurov@gmail.com>
41 * \ingroup module_mdtypes
47 #if GMX_GPU != GMX_GPU_NONE
49 #if GMX_GPU == GMX_GPU_CUDA
50 #include "gromacs/gpu_utils/cudautils.cuh"
52 #include "gromacs/gpu_utils/devicebuffer.h"
53 #if GMX_GPU == GMX_GPU_OPENCL
54 #include "gromacs/gpu_utils/oclutils.h"
56 #include "gromacs/math/vectypes.h"
57 #include "gromacs/mdtypes/state_propagator_data_gpu.h"
58 #include "gromacs/utility/classhelpers.h"
60 #include "state_propagator_data_gpu_impl.h"
65 StatePropagatorDataGpu::Impl::Impl(gmx_unused const void *commandStream,
66 gmx_unused const void *gpuContext,
67 GpuApiCallBehavior transferKind,
69 transferKind_(transferKind),
70 paddingSize_(paddingSize)
73 GMX_RELEASE_ASSERT(getenv("GMX_USE_GPU_BUFFER_OPS") == nullptr, "GPU buffer ops are not supported in this build.");
75 // Set the stream-context pair for the OpenCL builds,
76 // use the nullptr stream for CUDA builds
77 #if GMX_GPU == GMX_GPU_OPENCL
78 if (commandStream != nullptr)
80 commandStream_ = *static_cast<const CommandStream*>(commandStream);
82 if (gpuContext != nullptr)
84 gpuContext_ = *static_cast<const Context*>(gpuContext);
90 StatePropagatorDataGpu::Impl::~Impl()
94 void StatePropagatorDataGpu::Impl::reinit(int numAtomsLocal, int numAtomsAll)
96 #if GMX_GPU == GMX_GPU_OPENCL
97 GMX_ASSERT(gpuContext_ != nullptr, "GPU context should be set in OpenCL builds.");
99 numAtomsLocal_ = numAtomsLocal;
100 numAtomsAll_ = numAtomsAll;
103 if (paddingSize_ > 0)
105 numAtomsPadded = ((numAtomsAll_ + paddingSize_ - 1 ) / paddingSize_ )*paddingSize_;
109 numAtomsPadded = numAtomsAll_;
112 reallocateDeviceBuffer(&d_x_, DIM*numAtomsPadded, &d_xSize_, &d_xCapacity_, gpuContext_);
114 const size_t paddingAllocationSize = numAtomsPadded - numAtomsAll_;
115 if (paddingAllocationSize > 0)
117 clearDeviceBufferAsync(&d_x_, DIM*numAtomsAll_, DIM*paddingAllocationSize, commandStream_);
120 reallocateDeviceBuffer(&d_v_, DIM*numAtomsAll_, &d_vSize_, &d_vCapacity_, gpuContext_);
121 reallocateDeviceBuffer(&d_f_, DIM*numAtomsAll_, &d_fSize_, &d_fCapacity_, gpuContext_);
125 std::tuple<int, int> StatePropagatorDataGpu::Impl::getAtomRangesFromAtomLocality(AtomLocality atomLocality)
127 int atomsStartAt = 0;
128 int numAtomsToCopy = 0;
129 switch (atomLocality)
131 case AtomLocality::All:
133 numAtomsToCopy = numAtomsAll_;
135 case AtomLocality::Local:
137 numAtomsToCopy = numAtomsLocal_;
139 case AtomLocality::NonLocal:
140 atomsStartAt = numAtomsLocal_;
141 numAtomsToCopy = numAtomsAll_ - numAtomsLocal_;
144 GMX_RELEASE_ASSERT(false, "Wrong range of atoms requested in GPU state data manager. Should be All, Local or NonLocal.");
146 GMX_ASSERT(atomsStartAt >= 0, "The first elemtnt to copy has negative index. Probably, the GPU propagator state was not initialized.");
147 GMX_ASSERT(numAtomsToCopy >= 0, "Number of atoms to copy is negative. Probably, the GPU propagator state was not initialized.");
148 return std::make_tuple(atomsStartAt, numAtomsToCopy);
151 void StatePropagatorDataGpu::Impl::copyToDevice(DeviceBuffer<float> d_data,
152 const gmx::ArrayRef<const gmx::RVec> h_data,
154 AtomLocality atomLocality)
157 #if GMX_GPU == GMX_GPU_OPENCL
158 GMX_ASSERT(gpuContext_ != nullptr, "GPU context should be set in OpenCL builds.");
161 GMX_UNUSED_VALUE(dataSize);
163 GMX_ASSERT(dataSize >= 0, "Trying to copy to device buffer before it was allocated.");
165 int atomsStartAt, numAtomsToCopy;
166 std::tie(atomsStartAt, numAtomsToCopy) = getAtomRangesFromAtomLocality(atomLocality);
168 int elementsStartAt = atomsStartAt*DIM;
169 int numElementsToCopy = numAtomsToCopy*DIM;
171 if (numAtomsToCopy != 0)
173 GMX_ASSERT(elementsStartAt + numElementsToCopy <= dataSize, "The device allocation is smaller than requested copy range.");
174 GMX_ASSERT(atomsStartAt + numAtomsToCopy <= h_data.ssize(), "The host buffer is smaller than the requested copy range.");
176 // TODO: Use the proper stream
177 copyToDeviceBuffer(&d_data, reinterpret_cast<const float *>(&h_data.data()[atomsStartAt]),
178 elementsStartAt, numElementsToCopy,
179 commandStream_, transferKind_, nullptr);
183 void StatePropagatorDataGpu::Impl::copyFromDevice(gmx::ArrayRef<gmx::RVec> h_data,
184 DeviceBuffer<float> d_data,
186 AtomLocality atomLocality)
189 #if GMX_GPU == GMX_GPU_OPENCL
190 GMX_ASSERT(gpuContext_ != nullptr, "GPU context should be set in OpenCL builds.");
193 GMX_UNUSED_VALUE(dataSize);
195 GMX_ASSERT(dataSize >= 0, "Trying to copy from device buffer before it was allocated.");
197 int atomsStartAt, numAtomsToCopy;
198 std::tie(atomsStartAt, numAtomsToCopy) = getAtomRangesFromAtomLocality(atomLocality);
200 int elementsStartAt = atomsStartAt*DIM;
201 int numElementsToCopy = numAtomsToCopy*DIM;
203 if (numAtomsToCopy != 0)
205 GMX_ASSERT(elementsStartAt + numElementsToCopy <= dataSize, "The device allocation is smaller than requested copy range.");
206 GMX_ASSERT(atomsStartAt + numAtomsToCopy <= h_data.ssize(), "The host buffer is smaller than the requested copy range.");
208 // TODO: Use the proper stream
209 copyFromDeviceBuffer(reinterpret_cast<float*>(&h_data.data()[atomsStartAt]), &d_data,
210 elementsStartAt, numElementsToCopy,
211 commandStream_, transferKind_, nullptr);
216 DeviceBuffer<float> StatePropagatorDataGpu::Impl::getCoordinates()
221 void StatePropagatorDataGpu::Impl::copyCoordinatesToGpu(const gmx::ArrayRef<const gmx::RVec> h_x,
222 AtomLocality atomLocality)
224 copyToDevice(d_x_, h_x, d_xSize_, atomLocality);
227 void StatePropagatorDataGpu::Impl::copyCoordinatesFromGpu(gmx::ArrayRef<gmx::RVec> h_x,
228 AtomLocality atomLocality)
230 copyFromDevice(h_x, d_x_, d_xSize_, atomLocality);
234 DeviceBuffer<float> StatePropagatorDataGpu::Impl::getVelocities()
239 void StatePropagatorDataGpu::Impl::copyVelocitiesToGpu(const gmx::ArrayRef<const gmx::RVec> h_v,
240 AtomLocality atomLocality)
242 copyToDevice(d_v_, h_v, d_vSize_, atomLocality);
245 void StatePropagatorDataGpu::Impl::copyVelocitiesFromGpu(gmx::ArrayRef<gmx::RVec> h_v,
246 AtomLocality atomLocality)
248 copyFromDevice(h_v, d_v_, d_vSize_, atomLocality);
252 DeviceBuffer<float> StatePropagatorDataGpu::Impl::getForces()
257 void StatePropagatorDataGpu::Impl::copyForcesToGpu(const gmx::ArrayRef<const gmx::RVec> h_f,
258 AtomLocality atomLocality)
260 copyToDevice(d_f_, h_f, d_fSize_, atomLocality);
263 void StatePropagatorDataGpu::Impl::copyForcesFromGpu(gmx::ArrayRef<gmx::RVec> h_f,
264 AtomLocality atomLocality)
266 copyFromDevice(h_f, d_f_, d_fSize_, atomLocality);
269 void StatePropagatorDataGpu::Impl::synchronizeStream()
271 gpuStreamSynchronize(commandStream_);
274 int StatePropagatorDataGpu::Impl::numAtomsLocal()
276 return numAtomsLocal_;
279 int StatePropagatorDataGpu::Impl::numAtomsAll()
286 StatePropagatorDataGpu::StatePropagatorDataGpu(const void *commandStream,
287 const void *gpuContext,
288 GpuApiCallBehavior transferKind,
290 : impl_(new Impl(commandStream,
297 StatePropagatorDataGpu::StatePropagatorDataGpu(StatePropagatorDataGpu && /* other */) noexcept = default;
299 StatePropagatorDataGpu &StatePropagatorDataGpu::operator=(StatePropagatorDataGpu && /* other */) noexcept = default;
301 StatePropagatorDataGpu::~StatePropagatorDataGpu() = default;
304 void StatePropagatorDataGpu::reinit(int numAtomsLocal, int numAtomsAll)
306 return impl_->reinit(numAtomsLocal, numAtomsAll);
309 std::tuple<int, int> StatePropagatorDataGpu::getAtomRangesFromAtomLocality(AtomLocality atomLocality)
311 return impl_->getAtomRangesFromAtomLocality(atomLocality);
315 DeviceBuffer<float> StatePropagatorDataGpu::getCoordinates()
317 return impl_->getCoordinates();
320 void StatePropagatorDataGpu::copyCoordinatesToGpu(const gmx::ArrayRef<const gmx::RVec> h_x,
321 AtomLocality atomLocality)
323 return impl_->copyCoordinatesToGpu(h_x, atomLocality);
326 void StatePropagatorDataGpu::copyCoordinatesFromGpu(gmx::ArrayRef<RVec> h_x,
327 AtomLocality atomLocality)
329 return impl_->copyCoordinatesFromGpu(h_x, atomLocality);
333 DeviceBuffer<float> StatePropagatorDataGpu::getVelocities()
335 return impl_->getVelocities();
338 void StatePropagatorDataGpu::copyVelocitiesToGpu(const gmx::ArrayRef<const gmx::RVec> h_v,
339 AtomLocality atomLocality)
341 return impl_->copyVelocitiesToGpu(h_v, atomLocality);
344 void StatePropagatorDataGpu::copyVelocitiesFromGpu(gmx::ArrayRef<RVec> h_v,
345 AtomLocality atomLocality)
347 return impl_->copyVelocitiesFromGpu(h_v, atomLocality);
351 DeviceBuffer<float> StatePropagatorDataGpu::getForces()
353 return impl_->getForces();
356 void StatePropagatorDataGpu::copyForcesToGpu(const gmx::ArrayRef<const gmx::RVec> h_f,
357 AtomLocality atomLocality)
359 return impl_->copyForcesToGpu(h_f, atomLocality);
362 void StatePropagatorDataGpu::copyForcesFromGpu(gmx::ArrayRef<RVec> h_f,
363 AtomLocality atomLocality)
365 return impl_->copyForcesFromGpu(h_f, atomLocality);
368 void StatePropagatorDataGpu::synchronizeStream()
370 return impl_->synchronizeStream();
373 int StatePropagatorDataGpu::numAtomsLocal()
375 return impl_->numAtomsLocal();
378 int StatePropagatorDataGpu::numAtomsAll()
380 return impl_->numAtomsAll();
385 #endif // GMX_GPU == GMX_GPU_NONE