2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2018,2019,2020,2021, 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 Defines the host-side PME GPU data structure, which is dependent on the GPU types.
38 * It's included by pointer in the general PmeGpu host structure in pme_gpu_types_host.h.
40 * \author Aleksei Iupinov <a.yupinov@gmail.com>
41 * \ingroup module_ewald
44 #ifndef PMEGPUTYPESHOSTIMPL_H
45 #define PMEGPUTYPESHOSTIMPL_H
48 #include "gromacs/utility/enumerationhelpers.h"
55 # include "gromacs/gpu_utils/gpuregiontimer.cuh"
57 # include "gromacs/gpu_utils/gpuregiontimer_ocl.h"
59 # include "gromacs/gpu_utils/gpuregiontimer_sycl.h"
62 #include "gromacs/gpu_utils/gpueventsynchronizer.h"
64 #include "gromacs/fft/gpu_3dfft.h"
65 #include "gromacs/timing/gpu_timing.h" // for gtPME_EVENT_COUNT
68 //! Number of FEP states.
69 # define NUMFEPSTATES 2
78 * The main PME CUDA/OpenCL-specific host data structure, included in the PME GPU structure by the archSpecific pointer.
82 /*! \brief Constructor
84 * \param[in] deviceContext GPU device context
85 * \param[in] pmeStream GPU pme stream.
87 PmeGpuSpecific(const DeviceContext& deviceContext, const DeviceStream& pmeStream) :
88 deviceContext_(deviceContext), pmeStream_(pmeStream)
93 * A handle to the GPU context.
94 * TODO: this is currently extracted from the implementation of pmeGpu->programHandle_,
95 * but should be a constructor parameter to PmeGpu, as well as PmeGpuProgram,
96 * managed by high-level code.
98 const DeviceContext& deviceContext_;
100 /*! \brief The GPU stream where everything related to the PME happens. */
101 const DeviceStream& pmeStream_;
103 /* Synchronization events */
104 /*! \brief Triggered after the PME Force Calculations have been completed */
105 GpuEventSynchronizer pmeForcesReady;
106 /*! \brief Triggered after the grid has been copied to the host (after the spreading stage). */
107 GpuEventSynchronizer syncSpreadGridD2H;
109 /* Settings which are set at the start of the run */
110 /*! \brief A boolean which tells whether the complex and real grids for cu/clFFT are different or same. Currently true. */
111 bool performOutOfPlaceFFT = false;
112 /*! \brief A boolean which tells if the GPU timing events are enabled.
113 * False by default, can be enabled by setting the environment variable GMX_ENABLE_GPU_TIMING.
114 * Note: will not be reliable when multiple GPU tasks are running concurrently on the same
115 * device context, as CUDA events on multiple streams are untrustworthy.
117 bool useTiming = false;
119 //! Vector of FFT setups
120 std::vector<std::unique_ptr<gmx::Gpu3dFft>> fftSetup;
122 //! All the timers one might use
123 gmx::EnumerationArray<PmeStage, GpuRegionTimer> timingEvents;
125 //! Indices of timingEvents actually used
126 std::set<PmeStage> activeTimers;
128 /* GPU arrays element counts (not the arrays sizes in bytes!).
129 * They might be larger than the actual meaningful data sizes.
130 * These are paired: the actual element count + the maximum element count that can fit in the current allocated memory.
131 * These integer pairs are mostly meaningful for the reallocateDeviceBuffer calls.
132 * As such, if DeviceBuffer is refactored into a class, they can be freely changed, too.
133 * The only exceptions are realGridSize and complexGridSize which are also used for grid clearing/copying.
134 * TODO: these should live in a clean buffered container type, and be refactored in the NB/cudautils as well.
136 /*! \brief The kernelParams.atoms.coordinates float element count (actual)*/
137 int coordinatesSize = 0;
138 /*! \brief The kernelParams.atoms.coordinates float element count (reserved) */
139 int coordinatesSizeAlloc = 0;
140 /*! \brief The kernelParams.atoms.forces float element count (actual) */
142 /*! \brief The kernelParams.atoms.forces float element count (reserved) */
143 int forcesSizeAlloc = 0;
144 /*! \brief The kernelParams.atoms.gridlineIndices int element count (actual) */
145 int gridlineIndicesSize = 0;
146 /*! \brief The kernelParams.atoms.gridlineIndices int element count (reserved) */
147 int gridlineIndicesSizeAlloc = 0;
148 /*! \brief Both the kernelParams.atoms.theta and kernelParams.atoms.dtheta float element count (actual) */
149 int splineDataSize = 0;
150 /*! \brief Both the kernelParams.atoms.theta and kernelParams.atoms.dtheta float element count (reserved) */
151 int splineDataSizeAlloc = 0;
152 /*! \brief The kernelParams.atoms.coefficients float element count (actual) */
153 int coefficientsSize[NUMFEPSTATES] = { 0, 0 };
154 /*! \brief The kernelParams.atoms.coefficients float element count (reserved) */
155 int coefficientsCapacity[NUMFEPSTATES] = { 0, 0 };
156 /*! \brief The kernelParams.grid.splineValuesArray float element count (actual) */
157 int splineValuesSize[NUMFEPSTATES] = { 0, 0 };
158 /*! \brief The kernelParams.grid.splineValuesArray float element count (reserved) */
159 int splineValuesCapacity[NUMFEPSTATES] = { 0, 0 };
160 /*! \brief The kernelParams.grid.realGrid float element count (actual) */
161 int realGridSize[NUMFEPSTATES] = { 0, 0 };
162 /*! \brief The kernelParams.grid.realGrid float element count (reserved) */
163 int realGridCapacity[NUMFEPSTATES] = { 0, 0 };
164 /*! \brief The kernelParams.grid.fourierGrid float (not float2!) element count (actual) */
165 int complexGridSize[NUMFEPSTATES] = { 0, 0 };
166 /*! \brief The kernelParams.grid.fourierGrid float (not float2!) element count (reserved) */
167 int complexGridCapacity[NUMFEPSTATES] = { 0, 0 };