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 Declaration of interfaces for GPU state data propagator object.
39 * This object stores and manages positions, velocities and forces for
40 * all particles in the system on the GPU.
42 * \todo Add cycle counters.
43 * \todo Add synchronization points.
45 * \author Artem Zhmurov <zhmurov@gmail.com>
48 * \ingroup module_mdtypes
50 #ifndef GMX_MDTYPES_STATE_PROPAGATOR_DATA_GPU_H
51 #define GMX_MDTYPES_STATE_PROPAGATOR_DATA_GPU_H
53 #include "gromacs/gpu_utils/devicebuffer_datatype.h"
54 #include "gromacs/gpu_utils/gpu_utils.h"
55 #include "gromacs/math/vectypes.h"
56 #include "gromacs/utility/arrayref.h"
57 #include "gromacs/utility/classhelpers.h"
62 class StatePropagatorDataGpu
66 /*! \brief Atom locality indicator: local, non-local, all.
68 * \todo This should be managed by a separate object, since the localities
69 * are used here and in buffer ops.
71 enum class AtomLocality : int
73 Local = 0, //!< Local atoms
74 NonLocal = 1, //!< Non-local atoms
75 All = 2, //!< Both local and non-local atoms
76 Count = 3 //!< The number of atom locality types
79 /*! \brief Constructor
81 * The buffers are reallocated only at the reinit call, the padding is
82 * used there for the coordinates buffer. It is needed for PME and added at
83 * the end of the buffer. It is assumed that if the rank has PME duties on the
84 * GPU, all coordinates are copied to the GPU and hence, for this rank, the
85 * coordinates buffer is not split into local and non-local ranges. For other
86 * ranks, the padding size is zero. This works because only one rank ever does
87 * PME work on the GPU, and if that rank also does PP work that is the only
88 * rank. So all coordinates are always transferred.
90 * \note \p commandStream and \p gpuContext are allowed to be nullptr if
91 * StatePropagatorDataGpu is not used in the OpenCL run (e.g. if PME
92 * does not run on the GPU).
94 * \todo Make \p CommandStream visible in the CPU parts of the code so we
95 * will not have to pass a void*.
96 * \todo Make \p Context visible in CPU parts of the code so we will not
97 * have to pass a void*.
99 * \param[in] commandStream GPU stream, nullptr allowed.
100 * \param[in] gpuContext GPU context, nullptr allowed.
101 * \param[in] transferKind H2D/D2H transfer call behavior (synchronous or not).
102 * \param[in] paddingSize Padding size for coordinates buffer.
104 StatePropagatorDataGpu(const void *commandStream,
105 const void *gpuContext,
106 GpuApiCallBehavior transferKind,
109 ~StatePropagatorDataGpu();
111 /*! \brief Set the ranges for local and non-local atoms and reallocates buffers.
113 * The coordinates buffer is reallocated with the padding added at the end. The
114 * size of padding is set by the constructor.
116 * \param[in] numAtomsLocal Number of atoms in local domain.
117 * \param[in] numAtomsAll Total number of atoms to handle.
119 void reinit(int numAtomsLocal, int numAtomsAll);
121 /*! \brief Returns the range of atoms to be copied based on the copy type (all, local or non-local).
123 * \todo There are at least three versions of the function with this functionality in the code:
124 * this one and two more in NBNXM. These should be unified in a shape of a general function
127 * \param[in] atomLocality If all, local or non-local ranges are needed.
129 * \returns Tuple, containing the index of the first atom in the range and the total number of atoms in the range.
131 std::tuple<int, int> getAtomRangesFromAtomLocality(AtomLocality atomLocality);
134 /*! \brief Get the positions buffer on the GPU.
136 * \returns GPU positions buffer.
138 DeviceBuffer<float> getCoordinates();
140 /*! \brief Copy positions to the GPU memory.
142 * \param[in] h_x Positions in the host memory.
143 * \param[in] atomLocality Locality of the particles to copy.
145 void copyCoordinatesToGpu(gmx::ArrayRef<const gmx::RVec> h_x,
146 AtomLocality atomLocality);
148 /*! \brief Copy positions from the GPU memory.
150 * \param[in] h_x Positions buffer in the host memory.
151 * \param[in] atomLocality Locality of the particles to copy.
153 void copyCoordinatesFromGpu(gmx::ArrayRef<gmx::RVec> h_x,
154 AtomLocality atomLocality);
157 /*! \brief Get the velocities buffer on the GPU.
159 * \returns GPU velocities buffer.
161 DeviceBuffer<float> getVelocities();
163 /*! \brief Copy velocities to the GPU memory.
165 * \param[in] h_v Velocities in the host memory.
166 * \param[in] atomLocality Locality of the particles to copy.
168 void copyVelocitiesToGpu(gmx::ArrayRef<const gmx::RVec> h_v,
169 AtomLocality atomLocality);
171 /*! \brief Copy velocities from the GPU memory.
173 * \param[in] h_v Velocities buffer in the host memory.
174 * \param[in] atomLocality Locality of the particles to copy.
176 void copyVelocitiesFromGpu(gmx::ArrayRef<gmx::RVec> h_v,
177 AtomLocality atomLocality);
180 /*! \brief Get the force buffer on the GPU.
182 * \returns GPU force buffer.
184 DeviceBuffer<float> getForces();
186 /*! \brief Copy forces to the GPU memory.
188 * \param[in] h_f Forces in the host memory.
189 * \param[in] atomLocality Locality of the particles to copy.
191 void copyForcesToGpu(gmx::ArrayRef<const gmx::RVec> h_f,
192 AtomLocality atomLocality);
194 /*! \brief Copy forces from the GPU memory.
196 * \param[in] h_f Forces buffer in the host memory.
197 * \param[in] atomLocality Locality of the particles to copy.
199 void copyForcesFromGpu(gmx::ArrayRef<gmx::RVec> h_f,
200 AtomLocality atomLocality);
201 /*! \brief Synchronize the underlying GPU stream
203 void synchronizeStream();
205 /*! \brief Getter for the number of local atoms.
207 * \returns The number of local atoms.
211 /*! \brief Getter for the total number of atoms.
213 * \returns The total number of atoms.
219 gmx::PrivateImplPointer<Impl> impl_;
225 #endif // GMX_MDTYPES_STATE_PROPAGATOR_DATA_GPU_H