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.
36 /*! \libinternal \file
37 * \brief Defines the host-side PME GPU data structures.
38 * \todo Some renaming/refactoring, which does not impair the performance:
39 * -- bringing the function names up to guidelines
40 * -- PmeGpuSettings -> PmeGpuTasks
41 * -- refining GPU notation application (#2053)
42 * -- renaming coefficients to charges (?)
44 * \author Aleksei Iupinov <a.yupinov@gmail.com>
45 * \ingroup module_ewald
48 #ifndef GMX_EWALD_PME_GPU_TYPES_HOST_H
49 #define GMX_EWALD_PME_GPU_TYPES_HOST_H
56 #include "gromacs/ewald/pme.h"
57 #include "gromacs/ewald/pme_gpu_program.h"
58 #include "gromacs/gpu_utils/clfftinitializer.h"
59 #include "gromacs/gpu_utils/hostallocator.h"
60 #include "gromacs/math/vectypes.h"
62 #include "pme_gpu_settings.h"
63 #include "pme_gpu_staging.h"
66 struct PmeGpuSpecific;
68 /*! \brief A dummy typedef for the GPU host data placeholder on non-GPU builds */
69 typedef int PmeGpuSpecific;
73 struct PmeGpuCudaKernelParams;
74 /*! \brief A typedef for including the GPU kernel arguments data by pointer */
75 typedef PmeGpuCudaKernelParams PmeGpuKernelParams;
76 #elif GMX_GPU_OPENCL || GMX_GPU_SYCL
77 struct PmeGpuKernelParamsBase;
78 /*! \brief A typedef for including the GPU kernel arguments data by pointer */
79 typedef PmeGpuKernelParamsBase PmeGpuKernelParams;
81 /*! \brief A dummy typedef for the GPU kernel arguments data placeholder on non-GPU builds */
82 typedef int PmeGpuKernelParams;
85 struct DeviceInformation;
88 * The PME GPU structure for all the data copied directly from the CPU PME structure.
89 * The copying is done when the CPU PME structure is already (re-)initialized
90 * (pme_gpu_reinit is called at the end of gmx_pme_init).
91 * All the variables here are named almost the same way as in gmx_pme_t.
92 * The types are different: pointers are replaced by vectors.
93 * TODO: use the shared data with the PME CPU.
94 * Included in the main PME GPU structure by value.
98 /*! \brief Grid count */
100 /*! \brief Grid dimensions - nkx, nky, nkz */
102 /*! \brief PME interpolation order */
104 /*! \brief Ewald splitting coefficient for Coulomb */
106 /*! \brief Electrostatics parameter */
108 /*! \brief Gridline indices - nnx, nny, nnz */
110 /*! \brief Fractional shifts - fshx, fshy, fshz */
111 std::vector<real> fsh;
112 /*! \brief Precomputed B-spline values */
113 std::vector<real> bsp_mod[DIM];
114 /*! \brief The PME codepath being taken */
116 /*! \brief Whether PME execution is happening on a PME-only rank (from gmx_pme_t.bPPnode). */
118 /*! \brief The box scaler based on inputrec - created in pme_init and managed by CPU structure */
119 class EwaldBoxZScaler* boxScaler;
120 /*! \brief The previous computation box to know if we even need to update the current box params.
121 * \todo Manage this on higher level.
122 * \todo Alternatively, when this structure is used by CPU PME code, make use of this field there as well.
128 * The main PME GPU host structure, included in the PME CPU structure by pointer.
132 /*! \brief The information copied once per reinit from the CPU structure. */
133 std::shared_ptr<PmeShared> common; // TODO: make the CPU structure use the same type
135 //! A handle to the program created by buildPmeGpuProgram()
136 const PmeGpuProgram* programHandle_;
138 //! Handle that ensures the clFFT library has been initialized once per process.
139 std::unique_ptr<gmx::ClfftInitializer> initializedClfftLibrary_;
141 /*! \brief The settings. */
142 PmeGpuSettings settings;
144 /*! \brief The host-side buffers.
145 * The device-side buffers are buried in kernelParams, but that will have to change.
147 PmeGpuStaging staging;
149 /*! \brief Number of local atoms, padded to be divisible by c_pmeAtomDataAlignment.
151 * Used only as a basic size for almost all the atom data allocations
152 * (spline parameter data is also aligned by PME_SPREADGATHER_PARTICLES_PER_WARP).
153 * kernelParams.atoms.nAtoms is the actual atom count to be used for most data copying.
155 * TODO: memory allocation/padding properties should be handled by
156 * something like a container
160 /*! \brief Kernel scheduling grid width limit in X - derived from deviceinfo compute capability in CUDA.
161 * Declared as very large int to make it useful in computations with type promotion, to avoid overflows.
162 * OpenCL seems to not have readily available global work size limit, so we just assign a large arbitrary constant to this instead.
163 * TODO: this should be in PmeGpuProgram(Impl)
165 std::intmax_t maxGridWidthX;
167 /*! \brief A single structure encompassing all the PME data used on GPU.
168 * Its value is the only argument to all the PME GPU kernels.
169 * \todo Test whether this should be copied to the constant GPU memory once for each computation
170 * (or even less often with no box updates) instead of being an argument.
172 std::shared_ptr<PmeGpuKernelParams> kernelParams;
174 /*! \brief The pointer to GPU-framework specific host-side data, such as CUDA streams and events. */
175 std::shared_ptr<PmeGpuSpecific> archSpecific; /* FIXME: make it an unique_ptr */