src/gromacs/gpu_utils/gpu_utils.h

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 1991-2000, University of Groningen, The Netherlands.
   5  * Copyright (c) 2001-2010, The GROMACS development team.
   6  * Copyright (c) 2012,2013,2014,2015,2016,2017, by the GROMACS development team, led by
   7  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   8  * and including many others, as listed in the AUTHORS file in the
   9  * top-level source directory and at http://www.gromacs.org.
  10  *
  11  * GROMACS is free software; you can redistribute it and/or
  12  * modify it under the terms of the GNU Lesser General Public License
  13  * as published by the Free Software Foundation; either version 2.1
  14  * of the License, or (at your option) any later version.
  15  *
  16  * GROMACS is distributed in the hope that it will be useful,
  17  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  18  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  19  * Lesser General Public License for more details.
  20  *
  21  * You should have received a copy of the GNU Lesser General Public
  22  * License along with GROMACS; if not, see
  23  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  24  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  25  *
  26  * If you want to redistribute modifications to GROMACS, please
  27  * consider that scientific software is very special. Version
  28  * control is crucial - bugs must be traceable. We will be happy to
  29  * consider code for inclusion in the official distribution, but
  30  * derived work must not be called official GROMACS. Details are found
  31  * in the README & COPYING files - if they are missing, get the
  32  * official version at http://www.gromacs.org.
  33  *
  34  * To help us fund GROMACS development, we humbly ask that you cite
  35  * the research papers on the package. Check out http://www.gromacs.org.
  36  */
  37 /*! \libinternal \file
  38  *  \brief Declare functions for detection and initialization for GPU devices.
  39  *
  40  *  \author Szilard Pall <pall.szilard@gmail.com>
  41  *  \author Mark Abraham <mark.j.abraham@gmail.com>
  42  *
  43  *  \inlibraryapi
  44  */
  45 #ifndef GMX_GPU_UTILS_GPU_UTILS_H
  46 #define GMX_GPU_UTILS_GPU_UTILS_H
  47
  48 #include <cstdio>
  49
  50 #include "gromacs/gpu_utils/gpu_macros.h"
  51 #include "gromacs/utility/basedefinitions.h"
  52
  53 struct gmx_device_info_t;
  54 struct gmx_gpu_info_t;
  55
  56 namespace gmx
  57 {
  58 class MDLogger;
  59 }
  60
  61 /*! \brief Detect all GPUs in the system.
  62  *
  63  *  Will detect every GPU supported by the device driver in use. Also
  64  *  check for the compatibility of each and fill the gpu_info->gpu_dev array
  65  *  with the required information on each the device: ID, device properties,
  66  *  status.
  67  *
  68  *  \param[in] gpu_info    pointer to structure holding GPU information.
  69  *  \param[out] err_str    The error message of any GPU API error that caused
  70  *                         the detection to fail (if there was any). The memory
  71  *                         the pointer points to should be managed externally.
  72  *  \returns               non-zero if the detection encountered a failure, zero otherwise.
  73  */
  74 GPU_FUNC_QUALIFIER
  75 int detect_gpus(struct gmx_gpu_info_t *GPU_FUNC_ARGUMENT(gpu_info), char *GPU_FUNC_ARGUMENT(err_str)) GPU_FUNC_TERM_WITH_RETURN(-1)
  76
  77 /*! \brief Return whether the GPU with given \c index is compatible, ie suitable for use.
  78  *
  79  * \param[in]   gpu_info    Information about detected GPUs
  80  * \param[in]   index       index of GPU to ask about
  81  * \returns                 Whether the GPU is compatible.
  82  */
  83 GPU_FUNC_QUALIFIER
  84 bool isGpuCompatible(const gmx_gpu_info_t &GPU_FUNC_ARGUMENT(gpu_info),
  85                      int GPU_FUNC_ARGUMENT(index)) GPU_FUNC_TERM_WITH_RETURN(false)
  86
  87 /*! \brief Return a string describing how compatible the GPU with given \c index is.
  88  *
  89  * \param[in]   gpu_info    Information about detected GPUs
  90  * \param[in]   index       index of GPU to ask about
  91  * \returns                 A null-terminated C string describing the compatibility status, useful for error messages.
  92  */
  93 GPU_FUNC_QUALIFIER
  94 const char *getGpuCompatibilityDescription(const gmx_gpu_info_t &GPU_FUNC_ARGUMENT(gpu_info),
  95                                            int GPU_FUNC_ARGUMENT(index)) GPU_FUNC_TERM_WITH_RETURN("")
  96
  97 /*! \brief Frees the gpu_dev and dev_use array fields of \p gpu_info.
  98  *
  99  * \param[in]    gpu_info    pointer to structure holding GPU information
 100  */
 101 GPU_FUNC_QUALIFIER
 102 void free_gpu_info(const struct gmx_gpu_info_t *GPU_FUNC_ARGUMENT(gpu_info)) GPU_FUNC_TERM
 103
 104 /*! \brief Initializes the GPU described by \c deviceInfo.
 105  *
 106  * TODO Doxygen complains about these - probably a Doxygen bug, since
 107  * the patterns here are the same as elsewhere in this header.
 108  *
 109  *  param[in]    mdlog        log file to write to
 110  *  param[in]    rank         MPI rank of this process (for error output)
 111  * \param[inout] deviceInfo   device info of the GPU to initialize
 112  *
 113  * Issues a fatal error for any critical errors that occur during
 114  * initialization.
 115  */
 116 GPU_FUNC_QUALIFIER
 117 void init_gpu(const gmx::MDLogger &GPU_FUNC_ARGUMENT(mdlog),
 118               int GPU_FUNC_ARGUMENT(rank),
 119               gmx_device_info_t *GPU_FUNC_ARGUMENT(deviceInfo)) GPU_FUNC_TERM
 120
 121 /*! \brief Frees up the CUDA GPU used by the active context at the time of calling.
 122  *
 123  * The context is explicitly destroyed and therefore all data uploaded to the GPU
 124  * is lost. This should only be called when none of this data is required anymore.
 125  *
 126  * \param[in]  deviceInfo   device info of the GPU to clean up for
 127  * \param[out] result_str   the message related to the error that occurred
 128  *                          during the initialization (if there was any).
 129  *
 130  * \returns                 true if no error occurs during the freeing.
 131  */
 132 CUDA_FUNC_QUALIFIER
 133 gmx_bool free_cuda_gpu(const gmx_device_info_t *CUDA_FUNC_ARGUMENT(deviceInfo),
 134                        char *CUDA_FUNC_ARGUMENT(result_str)) CUDA_FUNC_TERM_WITH_RETURN(TRUE)
 135
 136 /*! \brief Return a pointer to the device info for \c deviceId
 137  *
 138  * \param[in] gpu_info      GPU info of all detected devices in the system.
 139  * \param[in] deviceId      ID for the GPU device requested.
 140  *
 141  * \returns                 Pointer to the device info for \c deviceId.
 142  */
 143 GPU_FUNC_QUALIFIER
 144 gmx_device_info_t *getDeviceInfo(const gmx_gpu_info_t &GPU_FUNC_ARGUMENT(gpu_info),
 145                                  int GPU_FUNC_ARGUMENT(deviceId)) GPU_FUNC_TERM_WITH_RETURN(NULL)
 146
 147 /*! \brief Returns the device ID of the CUDA GPU currently in use.
 148  *
 149  * The GPU used is the one that is active at the time of the call in the active context.
 150  *
 151  * \returns                 device ID of the GPU in use at the time of the call
 152  */
 153 CUDA_FUNC_QUALIFIER
 154 int get_current_cuda_gpu_device_id(void) CUDA_FUNC_TERM_WITH_RETURN(-1)
 155
 156 /*! \brief Formats and returns a device information string for a given GPU.
 157  *
 158  * Given an index *directly* into the array of available GPUs (gpu_dev)
 159  * returns a formatted info string for the respective GPU which includes
 160  * ID, name, compute capability, and detection status.
 161  *
 162  * \param[out]  s           pointer to output string (has to be allocated externally)
 163  * \param[in]   gpu_info    Information about detected GPUs
 164  * \param[in]   index       an index *directly* into the array of available GPUs
 165  */
 166 GPU_FUNC_QUALIFIER
 167 void get_gpu_device_info_string(char *GPU_FUNC_ARGUMENT(s),
 168                                 const struct gmx_gpu_info_t &GPU_FUNC_ARGUMENT(gpu_info),
 169                                 int GPU_FUNC_ARGUMENT(index)) GPU_FUNC_TERM
 170
 171 /*! \brief Returns the size of the gpu_dev_info struct.
 172  *
 173  * The size of gpu_dev_info can be used for allocation and communication.
 174  *
 175  * \returns                 size in bytes of gpu_dev_info
 176  */
 177 GPU_FUNC_QUALIFIER
 178 size_t sizeof_gpu_dev_info(void) GPU_FUNC_TERM_WITH_RETURN(0)
 179
 180 /*! \brief Returns a pointer *ptr to page-locked memory of size nbytes.
 181  *
 182  * The allocated memory is suitable to be used for data transfers between host
 183  * and GPU.
 184  * Error handling should be done within this function.
 185  */
 186 typedef void gmx_host_alloc_t (void **ptr, size_t nbytes);
 187
 188 /*! \brief Frees page-locked memory pointed to by *ptr.
 189  *
 190  * NULL should not be passed to this function.
 191  */
 192 typedef void gmx_host_free_t (void *ptr);
 193
 194 /*! \brief Set page-locked memory allocation functions used by the GPU host. */
 195 void gpu_set_host_malloc_and_free(bool               bUseGpuKernels,
 196                                   gmx_host_alloc_t **nb_alloc,
 197                                   gmx_host_free_t  **nb_free);
 198
 199
 200
 201 /*! \brief Starts the GPU profiler if mdrun is being profiled.
 202  *
 203  *  When a profiler run is in progress (based on the presence of the NVPROF_ID
 204  *  env. var.), the profiler is started to begin collecting data during the
 205  *  rest of the run (or until stopGpuProfiler is called).
 206  *
 207  *  Note that this is implemented only for the CUDA API.
 208  */
 209 CUDA_FUNC_QUALIFIER
 210 void startGpuProfiler(void) CUDA_FUNC_TERM
 211
 212
 213 /*! \brief Resets the GPU profiler if mdrun is being profiled.
 214  *
 215  * When a profiler run is in progress (based on the presence of the NVPROF_ID
 216  * env. var.), the profiler data is restet in order to eliminate the data collected
 217  * from the preceding part fo the run.
 218  *
 219  * This function should typically be called at the mdrun counter reset time.
 220  *
 221  * Note that this is implemented only for the CUDA API.
 222  */
 223 CUDA_FUNC_QUALIFIER
 224 void resetGpuProfiler(void) CUDA_FUNC_TERM
 225
 226
 227 /*! \brief Stops the CUDA profiler if mdrun is being profiled.
 228  *
 229  *  This function can be called at cleanup when skipping recording
 230  *  recording subsequent API calls from being traces/profiled is desired,
 231  *  e.g. before uninitialization.
 232  *
 233  *  Note that this is implemented only for the CUDA API.
 234  */
 235 CUDA_FUNC_QUALIFIER
 236 void stopGpuProfiler(void) CUDA_FUNC_TERM
 237
 238
 239 #endif