2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2018,2019,2020, 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<int order, bool computeSplines, bool spreadCharges, bool wrapX, bool wrapY, bool writeGlobal, ThreadsPerAtom threadsPerAtom>
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.
66 pme_spline_and_spread_kernel<c_pmeOrder, true, true, c_wrapX, c_wrapY, true, ThreadsPerAtom::Order> (const PmeGpuCudaKernelParams);
68 pme_spline_and_spread_kernel<c_pmeOrder, true, true, c_wrapX, c_wrapY, true, ThreadsPerAtom::OrderSquared>(const PmeGpuCudaKernelParams);
70 pme_spline_and_spread_kernel<c_pmeOrder, true, false, c_wrapX, c_wrapY, true, ThreadsPerAtom::Order> (const PmeGpuCudaKernelParams);
72 pme_spline_and_spread_kernel<c_pmeOrder, true, false, c_wrapX, c_wrapY, true, ThreadsPerAtom::OrderSquared>(const PmeGpuCudaKernelParams);
74 pme_spline_and_spread_kernel<c_pmeOrder, false, true, c_wrapX, c_wrapY, true, ThreadsPerAtom::Order> (const PmeGpuCudaKernelParams);
76 pme_spline_and_spread_kernel<c_pmeOrder, false, true, c_wrapX, c_wrapY, true, ThreadsPerAtom::OrderSquared>(const PmeGpuCudaKernelParams);
78 pme_spline_and_spread_kernel<c_pmeOrder, true, true, c_wrapX, c_wrapY, false, ThreadsPerAtom::Order> (const PmeGpuCudaKernelParams);
80 pme_spline_and_spread_kernel<c_pmeOrder, true, true, c_wrapX, c_wrapY, false, ThreadsPerAtom::OrderSquared>(const PmeGpuCudaKernelParams);
83 template<GridOrdering gridOrdering, bool computeEnergyAndVirial>
84 void pme_solve_kernel(const PmeGpuCudaKernelParams kernelParams);
86 // Add extern declarations to inform that there will be a definition
87 // provided in another translation unit.
89 extern template void pme_solve_kernel<GridOrdering::XYZ, false>(const PmeGpuCudaKernelParams);
90 extern template void pme_solve_kernel<GridOrdering::XYZ, true> (const PmeGpuCudaKernelParams);
91 extern template void pme_solve_kernel<GridOrdering::YZX, false>(const PmeGpuCudaKernelParams);
92 extern template void pme_solve_kernel<GridOrdering::YZX, true> (const PmeGpuCudaKernelParams);
95 template<int order, bool wrapX, bool wrapY, bool readGlobal, ThreadsPerAtom threadsPerAtom>
96 void pme_gather_kernel(const PmeGpuCudaKernelParams kernelParams);
98 // Add extern declarations to inform that there will be a definition
99 // provided in another translation unit.
101 extern template void pme_gather_kernel<c_pmeOrder, c_wrapX, c_wrapY, true, ThreadsPerAtom::Order> (const PmeGpuCudaKernelParams);
102 extern template void pme_gather_kernel<c_pmeOrder, c_wrapX, c_wrapY, false, ThreadsPerAtom::Order> (const PmeGpuCudaKernelParams);
103 extern template void pme_gather_kernel<c_pmeOrder, c_wrapX, c_wrapY, true, ThreadsPerAtom::OrderSquared>(const PmeGpuCudaKernelParams);
104 extern template void pme_gather_kernel<c_pmeOrder, c_wrapX, c_wrapY, false, ThreadsPerAtom::OrderSquared>(const PmeGpuCudaKernelParams);
107 PmeGpuProgramImpl::PmeGpuProgramImpl(const DeviceContext& deviceContext) :
108 deviceContext_(deviceContext)
111 warpSize_ = warp_size;
112 spreadWorkGroupSize = c_spreadMaxThreadsPerBlock;
113 solveMaxWorkGroupSize = c_solveMaxThreadsPerBlock;
114 gatherWorkGroupSize = c_gatherMaxThreadsPerBlock;
116 /* Not all combinations of the splineAndSpread, spline and Spread kernels are required
117 * If only the spline (without the spread) then it does not make sense not to write the data to global memory
118 * Similarly the spread kernel (without the spline) implies that we should read the spline data from global memory
121 splineAndSpreadKernel = pme_spline_and_spread_kernel<c_pmeOrder, true, true, c_wrapX, c_wrapY, false, ThreadsPerAtom::OrderSquared>;
122 splineAndSpreadKernelThPerAtom4 = pme_spline_and_spread_kernel<c_pmeOrder, true, true, c_wrapX, c_wrapY, false, ThreadsPerAtom::Order>;
123 splineAndSpreadKernelWriteSplines = pme_spline_and_spread_kernel<c_pmeOrder, true, true, c_wrapX, c_wrapY, true, ThreadsPerAtom::OrderSquared>;
124 splineAndSpreadKernelWriteSplinesThPerAtom4 = pme_spline_and_spread_kernel<c_pmeOrder, true, true, c_wrapX, c_wrapY, true, ThreadsPerAtom::Order>;
125 splineKernel = pme_spline_and_spread_kernel<c_pmeOrder, true, false, c_wrapX, c_wrapY, true, ThreadsPerAtom::OrderSquared>;
126 splineKernelThPerAtom4 = pme_spline_and_spread_kernel<c_pmeOrder, true, false, c_wrapX, c_wrapY, true, ThreadsPerAtom::Order>;
127 spreadKernel = pme_spline_and_spread_kernel<c_pmeOrder, false, true, c_wrapX, c_wrapY, true, ThreadsPerAtom::OrderSquared>;
128 spreadKernelThPerAtom4 = pme_spline_and_spread_kernel<c_pmeOrder, false, true, c_wrapX, c_wrapY, true, ThreadsPerAtom::Order>;
129 gatherKernel = pme_gather_kernel<c_pmeOrder, c_wrapX, c_wrapY, false, ThreadsPerAtom::OrderSquared>;
130 gatherKernelThPerAtom4 = pme_gather_kernel<c_pmeOrder, c_wrapX, c_wrapY, false, ThreadsPerAtom::Order>;
131 gatherKernelReadSplines = pme_gather_kernel<c_pmeOrder, c_wrapX, c_wrapY, true, ThreadsPerAtom::OrderSquared>;
132 gatherKernelReadSplinesThPerAtom4 = pme_gather_kernel<c_pmeOrder, c_wrapX, c_wrapY, true, ThreadsPerAtom::Order>;
133 solveXYZKernel = pme_solve_kernel<GridOrdering::XYZ, false>;
134 solveXYZEnergyKernel = pme_solve_kernel<GridOrdering::XYZ, true>;
135 solveYZXKernel = pme_solve_kernel<GridOrdering::YZX, false>;
136 solveYZXEnergyKernel = pme_solve_kernel<GridOrdering::YZX, true>;
140 PmeGpuProgramImpl::~PmeGpuProgramImpl() {}