src/gromacs/ewald/pme-gpu-types-host.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
  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 (?)
  43  *
  44  * \author Aleksei Iupinov <a.yupinov@gmail.com>
  45  * \ingroup module_ewald
  46  */
  47
  48 #ifndef GMX_EWALD_PME_GPU_TYPES_HOST_H
  49 #define GMX_EWALD_PME_GPU_TYPES_HOST_H
  50
  51 #include "config.h"
  52
  53 #include <memory>
  54 #include <vector>
  55
  56 #include "gromacs/ewald/pme.h"
  57 #include "gromacs/gpu_utils/gpu_utils.h"      // for GpuApiCallBehavior
  58 #include "gromacs/gpu_utils/hostallocator.h"
  59 #include "gromacs/math/vectypes.h"
  60
  61 #if GMX_GPU == GMX_GPU_CUDA
  62
  63 struct PmeGpuCuda;
  64 /*! \brief A typedef for including the GPU host data by pointer */
  65 typedef PmeGpuCuda PmeGpuSpecific;
  66
  67 struct PmeGpuCudaKernelParams;
  68 /*! \brief A typedef for including the GPU kernel arguments data by pointer */
  69 typedef PmeGpuCudaKernelParams PmeGpuKernelParams;
  70
  71 #else
  72
  73 /*! \brief A dummy typedef for the GPU host data placeholder on non-GPU builds */
  74 typedef int PmeGpuSpecific;
  75 /*! \brief A dummy typedef for the GPU kernel arguments data placeholder on non-GPU builds */
  76 typedef int PmeGpuKernelParams;
  77
  78 #endif
  79
  80 struct gmx_device_info_t;
  81
  82 /*! \internal \brief
  83  * The PME GPU settings structure, included in the main PME GPU structure by value.
  84  */
  85 struct PmeGpuSettings
  86 {
  87     /* Permanent settings set on initialization */
  88     /*! \brief A boolean which tells if the solving is performed on GPU. Currently always true */
  89     bool performGPUSolve;
  90     /*! \brief A boolean which tells if the gathering is performed on GPU. Currently always true */
  91     bool performGPUGather;
  92     /*! \brief A boolean which tells if the FFT is performed on GPU. Currently true for a single MPI rank. */
  93     bool performGPUFFT;
  94     /*! \brief A convenience boolean which tells if PME decomposition is used. */
  95     bool useDecomposition;
  96     /*! \brief A boolean which tells if any PME GPU stage should copy all of its outputs to the host.
  97      * Only intended to be used by the test framework.
  98      */
  99     bool               copyAllOutputs;
 100     /*! \brief An enum which tells whether most PME GPU D2H/H2D data transfers should be synchronous. */
 101     GpuApiCallBehavior transferKind;
 102     /*! \brief Various flags for the current PME computation, corresponding to the GMX_PME_ flags in pme.h. */
 103     int                currentFlags;
 104 };
 105
 106 /*! \internal \brief
 107  * The PME GPU intermediate buffers structure, included in the main PME GPU structure by value.
 108  * Buffers are managed by the PME GPU module.
 109  */
 110 struct PmeGpuStaging
 111 {
 112     //! Host-side force buffer
 113     std::vector < gmx::RVec, gmx::HostAllocator < gmx::RVec>> h_forces;
 114
 115     /*! \brief Virial and energy intermediate host-side buffer. Size is PME_GPU_VIRIAL_AND_ENERGY_COUNT. */
 116     float  *h_virialAndEnergy;
 117     /*! \brief B-spline values intermediate host-side buffer. */
 118     float  *h_splineModuli;
 119
 120     /*! \brief Pointer to the host memory with B-spline values. Only used for host-side gather, or unit tests */
 121     float  *h_theta;
 122     /*! \brief Pointer to the host memory with B-spline derivative values. Only used for host-side gather, or unit tests */
 123     float  *h_dtheta;
 124     /*! \brief Pointer to the host memory with ivec atom gridline indices. Only used for host-side gather, or unit tests */
 125     int    *h_gridlineIndices;
 126 };
 127
 128 /*! \internal \brief
 129  * The PME GPU structure for all the data copied directly from the CPU PME structure.
 130  * The copying is done when the CPU PME structure is already (re-)initialized
 131  * (pme_gpu_reinit is called at the end of gmx_pme_init).
 132  * All the variables here are named almost the same way as in gmx_pme_t.
 133  * The types are different: pointers are replaced by vectors.
 134  * TODO: use the shared data with the PME CPU.
 135  * Included in the main PME GPU structure by value.
 136  */
 137 struct PmeShared
 138 {
 139     /*! \brief Grid count - currently always 1 on GPU */
 140     int                    ngrids;
 141     /*! \brief Grid dimensions - nkx, nky, nkz */
 142     int                    nk[DIM];
 143     /*! \brief PME interpolation order */
 144     int                    pme_order;
 145     /*! \brief Ewald splitting coefficient for Coulomb */
 146     real                   ewaldcoeff_q;
 147     /*! \brief Electrostatics parameter */
 148     real                   epsilon_r;
 149     /*! \brief Gridline indices - nnx, nny, nnz */
 150     std::vector<int>       nn;
 151     /*! \brief Fractional shifts - fshx, fshy, fshz */
 152     std::vector<real>      fsh;
 153     /*! \brief Precomputed B-spline values */
 154     std::vector<real>      bsp_mod[DIM];
 155     /*! \brief The PME codepath being taken */
 156     PmeRunMode             runMode;
 157     /*! \brief The box scaler based on inputrec - created in pme_init and managed by CPU structure */
 158     class EwaldBoxZScaler *boxScaler;
 159     /*! \brief The previous computation box to know if we even need to update the current box params.
 160      * \todo Manage this on higher level.
 161      * \todo Alternatively, when this structure is used by CPU PME code, make use of this field there as well.
 162      */
 163     matrix previousBox;
 164 };
 165
 166 /*! \internal \brief
 167  * The main PME GPU host structure, included in the PME CPU structure by pointer.
 168  */
 169 struct PmeGpu
 170 {
 171     /*! \brief The information copied once per reinit from the CPU structure. */
 172     std::shared_ptr<PmeShared> common; // TODO: make the CPU structure use the same type
 173
 174     /*! \brief The settings. */
 175     PmeGpuSettings settings;
 176
 177     /*! \brief The host-side buffers.
 178      * The device-side buffers are buried in kernelParams, but that will have to change.
 179      */
 180     PmeGpuStaging staging;
 181
 182     /*! \brief Number of local atoms, padded to be divisible by PME_ATOM_DATA_ALIGNMENT.
 183      * Used for kernel scheduling.
 184      * kernelParams.atoms.nAtoms is the actual atom count to be used for data copying.
 185      * TODO: this and the next member represent a memory allocation/padding properties -
 186      * what a container type should do ideally.
 187      */
 188     int nAtomsPadded;
 189     /*! \brief Number of local atoms, padded to be divisible by PME_ATOM_DATA_ALIGNMENT
 190      * if c_usePadding is true.
 191      * Used only as a basic size for almost all the atom data allocations
 192      * (spline parameter data is also aligned by PME_SPREADGATHER_PARTICLES_PER_WARP).
 193      * This should be the same as (c_usePadding ? nAtomsPadded : kernelParams.atoms.nAtoms).
 194      * kernelParams.atoms.nAtoms is the actual atom count to be used for most data copying.
 195      */
 196     int nAtomsAlloc;
 197
 198     /*! \brief A pointer to the device used during the execution. */
 199     gmx_device_info_t *deviceInfo;
 200
 201     /*! \brief Kernel scheduling grid width limit in X - derived from deviceinfo compute capability in CUDA.
 202      * Declared as very large int to make it useful in computations with type promotion, to avoid overflows.
 203      */
 204     std::intmax_t maxGridWidthX;
 205
 206     /*! \brief A single structure encompassing all the PME data used on GPU.
 207      * Its value is the only argument to all the PME GPU kernels.
 208      * \todo Test whether this should be copied to the constant GPU memory once for each computation
 209      * (or even less often with no box updates) instead of being an argument.
 210      */
 211     std::shared_ptr<PmeGpuKernelParams> kernelParams;
 212
 213     /*! \brief The pointer to GPU-framework specific host-side data, such as CUDA streams and events. */
 214     std::shared_ptr<PmeGpuSpecific> archSpecific; /* FIXME: make it an unique_ptr */
 215 };
 216
 217 #endif