src/gromacs/ewald/pme-gpu-program-impl.h

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2018, 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.
   8  *
   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.
  13  *
  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.
  18  *
  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.
  23  *
  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.
  31  *
  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.
  34  */
  35 /*! \internal \file
  36  * \brief
  37  * Declares PmeGpuProgramImpl, which stores PME GPU (compiled) kernel handles.
  38  *
  39  * \author Aleksei Iupinov <a.yupinov@gmail.com>
  40  * \ingroup module_ewald
  41  */
  42 #ifndef GMX_EWALD_PME_PME_GPU_PROGRAM_IMPL_H
  43 #define GMX_EWALD_PME_PME_GPU_PROGRAM_IMPL_H
  44
  45 #include "config.h"
  46
  47 #include "gromacs/utility/classhelpers.h"
  48
  49 #if GMX_GPU == GMX_GPU_CUDA
  50 #include "gromacs/gpu_utils/gputraits.cuh"
  51 #elif GMX_GPU == GMX_GPU_OPENCL
  52 #include "gromacs/gpu_utils/gputraits_ocl.h"
  53 #elif GMX_GPU == GMX_GPU_NONE
  54 // TODO place in gputraits_stub.h
  55 using Context = void *;
  56 #endif
  57
  58 struct gmx_device_info_t;
  59
  60 /*! \internal
  61  * \brief
  62  * PME GPU persistent host program/kernel data, which should be initialized once for the whole execution.
  63  *
  64  * Primary purpose of this is to not recompile GPU kernels for each OpenCL unit test,
  65  * while the relevant GPU context (e.g. cl_context) instance persists.
  66  * In CUDA, this just assigns the kernel function pointers.
  67  * This also implicitly relies on the fact that reasonable share of the kernels are always used.
  68  * If there were more template parameters, even smaller share of all possible kernels would be used.
  69  *
  70  * \todo In future if we would need to react to either user input or
  71  * auto-tuning to compile different kernels, then we might wish to
  72  * revisit the number of kernels we pre-compile, and/or the management
  73  * of their lifetime.
  74  *
  75  * This also doesn't manage cuFFT/clFFT kernels, which depend on the PME grid dimensions.
  76  *
  77  * TODO: pass cl_context to the constructor and not create it inside.
  78  * See also Redmine #2522.
  79  */
  80 struct PmeGpuProgramImpl
  81 {
  82     /*! \brief
  83      * This is a handle to the GPU context, which is just a dummy in CUDA,
  84      * but is created/destroyed by this class in OpenCL.
  85      * TODO: Later we want to be able to own the context at a higher level and not here,
  86      * but this class would still need the non-owning context handle to build the kernels.
  87      */
  88     Context context;
  89
  90     //! Conveniently all the PME kernels use the same single argument type
  91 #if GMX_GPU == GMX_GPU_CUDA
  92     using PmeKernelHandle = void(*)(const struct PmeGpuCudaKernelParams);
  93 #elif GMX_GPU == GMX_GPU_OPENCL
  94     using PmeKernelHandle = cl_kernel;
  95 #else
  96     using PmeKernelHandle = void *;
  97 #endif
  98
  99     /*! \brief
 100      * Maximum synchronous GPU thread group execution width.
 101      * "Warp" is a CUDA term which we end up reusing in OpenCL kernels as well.
 102      * For CUDA, this is a static value that comes from gromacs/gpu_utils/cuda_arch_utils.cuh;
 103      * for OpenCL, we have to query it dynamically.
 104      */
 105     size_t warpSize;
 106
 107     //@{
 108     /**
 109      * Spread/spline kernels are compiled only for order of 4.
 110      * Spreading kernels also have hardcoded X/Y indices wrapping parameters,
 111      * as a placeholder for implementing 1/2D decomposition.
 112      */
 113     size_t          spreadWorkGroupSize;
 114
 115     PmeKernelHandle splineKernel;
 116     PmeKernelHandle spreadKernel;
 117     PmeKernelHandle splineAndSpreadKernel;
 118     //@}
 119
 120     //@{
 121     /** Same for gather: hardcoded X/Y unwrap parameters, order of 4, plus
 122      * it can either reduce with previous forces in the host buffer, or ignore them.
 123      */
 124     size_t          gatherWorkGroupSize;
 125
 126     PmeKernelHandle gatherReduceWithInputKernel;
 127     PmeKernelHandle gatherKernel;
 128     //@}
 129
 130     //@{
 131     /** Solve kernel doesn't care about the interpolation order, but can optionally
 132      * compute energy and virial, and supports XYZ and YZX grid orderings.
 133      */
 134     size_t          solveMaxWorkGroupSize;
 135
 136     PmeKernelHandle solveYZXKernel;
 137     PmeKernelHandle solveXYZKernel;
 138     PmeKernelHandle solveYZXEnergyKernel;
 139     PmeKernelHandle solveXYZEnergyKernel;
 140     //@}
 141
 142     PmeGpuProgramImpl() = delete;
 143     //! Constructor for the given device
 144     explicit PmeGpuProgramImpl(const gmx_device_info_t *deviceInfo);
 145     ~PmeGpuProgramImpl();
 146     GMX_DISALLOW_COPY_AND_ASSIGN(PmeGpuProgramImpl);
 147
 148     private:
 149         // Compiles kernels, if supported. Called by the constructor.
 150         void compileKernels(const gmx_device_info_t *deviceInfo);
 151 };
 152
 153 #endif