2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2018,2019, 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.
38 * Implements PmeGpuProgramImpl, which stores permanent PME GPU context-derived data,
39 * such as (compiled) kernel handles.
41 * \author Aleksei Iupinov <a.yupinov@gmail.com>
42 * \ingroup module_ewald
46 #include "pme_gpu_program_impl.h"
48 #include "pme_gpu_constants.h"
49 #include "pme_gpu_internal.h" // for GridOrdering enum
50 #include "pme_gpu_types_host.h"
52 // PME interpolation order
53 constexpr int c_pmeOrder = 4;
54 // These hardcoded spread/gather parameters refer to not-implemented PME GPU 2D decomposition in X/Y
55 constexpr bool c_wrapX = true;
56 constexpr bool c_wrapY = true;
58 //! PME CUDA kernels forward declarations. Kernels are documented in their respective files.
59 template<const int order, const bool computeSplines, const bool spreadCharges, const bool wrapX, const bool wrapY, const bool writeGlobal, const bool orderThreads>
60 void pme_spline_and_spread_kernel(const PmeGpuCudaKernelParams kernelParams);
62 // Add extern declarations to inform that there will be a definition
63 // provided in another translation unit.
64 extern template void pme_spline_and_spread_kernel<c_pmeOrder, true, true, c_wrapX, c_wrapY, true, true>(
65 const PmeGpuCudaKernelParams);
66 extern template void pme_spline_and_spread_kernel<c_pmeOrder, true, true, c_wrapX, c_wrapY, true, false>(
67 const PmeGpuCudaKernelParams);
68 extern template void pme_spline_and_spread_kernel<c_pmeOrder, true, false, c_wrapX, c_wrapY, true, true>(
69 const PmeGpuCudaKernelParams);
70 extern template void pme_spline_and_spread_kernel<c_pmeOrder, true, false, c_wrapX, c_wrapY, true, false>(
71 const PmeGpuCudaKernelParams);
72 extern template void pme_spline_and_spread_kernel<c_pmeOrder, false, true, c_wrapX, c_wrapY, true, true>(
73 const PmeGpuCudaKernelParams);
74 extern template void pme_spline_and_spread_kernel<c_pmeOrder, false, true, c_wrapX, c_wrapY, true, false>(
75 const PmeGpuCudaKernelParams);
76 extern template void pme_spline_and_spread_kernel<c_pmeOrder, true, true, c_wrapX, c_wrapY, false, true>(
77 const PmeGpuCudaKernelParams);
78 extern template void pme_spline_and_spread_kernel<c_pmeOrder, true, true, c_wrapX, c_wrapY, false, false>(
79 const PmeGpuCudaKernelParams);
81 template<GridOrdering gridOrdering, bool computeEnergyAndVirial>
82 void pme_solve_kernel(const PmeGpuCudaKernelParams kernelParams);
84 // Add extern declarations to inform that there will be a definition
85 // provided in another translation unit.
86 extern template void pme_solve_kernel<GridOrdering::XYZ, false>(const PmeGpuCudaKernelParams);
87 extern template void pme_solve_kernel<GridOrdering::XYZ, true>(const PmeGpuCudaKernelParams);
88 extern template void pme_solve_kernel<GridOrdering::YZX, false>(const PmeGpuCudaKernelParams);
89 extern template void pme_solve_kernel<GridOrdering::YZX, true>(const PmeGpuCudaKernelParams);
91 template<const int order, const bool overwriteForces, const bool wrapX, const bool wrapY, const bool readGlobal, const bool orderThreads>
92 void pme_gather_kernel(const PmeGpuCudaKernelParams kernelParams);
94 // Add extern declarations to inform that there will be a definition
95 // provided in another translation unit.
96 extern template void pme_gather_kernel<c_pmeOrder, true, c_wrapX, c_wrapY, true, true>(const PmeGpuCudaKernelParams);
97 extern template void pme_gather_kernel<c_pmeOrder, true, c_wrapX, c_wrapY, false, true>(const PmeGpuCudaKernelParams);
98 extern template void pme_gather_kernel<c_pmeOrder, false, c_wrapX, c_wrapY, true, true>(const PmeGpuCudaKernelParams);
99 extern template void pme_gather_kernel<c_pmeOrder, false, c_wrapX, c_wrapY, false, true>(const PmeGpuCudaKernelParams);
100 extern template void pme_gather_kernel<c_pmeOrder, true, c_wrapX, c_wrapY, true, false>(const PmeGpuCudaKernelParams);
101 extern template void pme_gather_kernel<c_pmeOrder, true, c_wrapX, c_wrapY, false, false>(const PmeGpuCudaKernelParams);
102 extern template void pme_gather_kernel<c_pmeOrder, false, c_wrapX, c_wrapY, true, false>(const PmeGpuCudaKernelParams);
103 extern template void pme_gather_kernel<c_pmeOrder, false, c_wrapX, c_wrapY, false, false>(const PmeGpuCudaKernelParams);
105 PmeGpuProgramImpl::PmeGpuProgramImpl(const gmx_device_info_t*)
108 warpSize = warp_size;
109 spreadWorkGroupSize = c_spreadMaxThreadsPerBlock;
110 solveMaxWorkGroupSize = c_solveMaxThreadsPerBlock;
111 gatherWorkGroupSize = c_gatherMaxThreadsPerBlock;
114 * Not all combinations of the splineAndSpread, spline and Spread kernels are required
115 * If only the spline (without the spread) then it does not make sense not to write the data to global memory
116 * Similarly the spread kernel (without the spline) implies that we should read the spline data from global memory
118 splineAndSpreadKernel =
119 pme_spline_and_spread_kernel<c_pmeOrder, true, true, c_wrapX, c_wrapY, false, false>;
120 splineAndSpreadKernelThPerAtom4 =
121 pme_spline_and_spread_kernel<c_pmeOrder, true, true, c_wrapX, c_wrapY, false, true>;
122 splineAndSpreadKernelWriteSplines =
123 pme_spline_and_spread_kernel<c_pmeOrder, true, true, c_wrapX, c_wrapY, true, false>;
124 splineAndSpreadKernelWriteSplinesThPerAtom4 =
125 pme_spline_and_spread_kernel<c_pmeOrder, true, true, c_wrapX, c_wrapY, true, true>;
126 splineKernel = pme_spline_and_spread_kernel<c_pmeOrder, true, false, c_wrapX, c_wrapY, true, false>;
127 splineKernelThPerAtom4 =
128 pme_spline_and_spread_kernel<c_pmeOrder, true, false, c_wrapX, c_wrapY, true, true>;
129 spreadKernel = pme_spline_and_spread_kernel<c_pmeOrder, false, true, c_wrapX, c_wrapY, true, false>;
130 spreadKernelThPerAtom4 =
131 pme_spline_and_spread_kernel<c_pmeOrder, false, true, c_wrapX, c_wrapY, true, true>;
132 gatherKernel = pme_gather_kernel<c_pmeOrder, true, c_wrapX, c_wrapY, false, false>;
133 gatherKernelThPerAtom4 = pme_gather_kernel<c_pmeOrder, true, c_wrapX, c_wrapY, false, true>;
134 gatherKernelReadSplines = pme_gather_kernel<c_pmeOrder, true, c_wrapX, c_wrapY, true, false>;
135 gatherKernelReadSplinesThPerAtom4 = pme_gather_kernel<c_pmeOrder, true, c_wrapX, c_wrapY, true, true>;
136 gatherReduceWithInputKernel = pme_gather_kernel<c_pmeOrder, false, c_wrapX, c_wrapY, false, false>;
137 gatherReduceWithInputKernelThPerAtom4 =
138 pme_gather_kernel<c_pmeOrder, false, c_wrapX, c_wrapY, false, true>;
139 gatherReduceWithInputKernelReadSplines =
140 pme_gather_kernel<c_pmeOrder, false, c_wrapX, c_wrapY, true, false>;
141 gatherReduceWithInputKernelReadSplinesThPerAtom4 =
142 pme_gather_kernel<c_pmeOrder, false, c_wrapX, c_wrapY, true, true>;
143 solveXYZKernel = pme_solve_kernel<GridOrdering::XYZ, false>;
144 solveXYZEnergyKernel = pme_solve_kernel<GridOrdering::XYZ, true>;
145 solveYZXKernel = pme_solve_kernel<GridOrdering::YZX, false>;
146 solveYZXEnergyKernel = pme_solve_kernel<GridOrdering::YZX, true>;
149 PmeGpuProgramImpl::~PmeGpuProgramImpl() {}