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, 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_gpu_info_t;
  54 struct gmx_gpu_opt_t;
  55
  56 /*! \brief Detect all GPUs in the system.
  57  *
  58  *  Will detect every GPU supported by the device driver in use. Also
  59  *  check for the compatibility of each and fill the gpu_info->gpu_dev array
  60  *  with the required information on each the device: ID, device properties,
  61  *  status.
  62  *
  63  *  \param[in] gpu_info    pointer to structure holding GPU information.
  64  *  \param[out] err_str    The error message of any GPU API error that caused
  65  *                         the detection to fail (if there was any). The memory
  66  *                         the pointer points to should be managed externally.
  67  *  \returns               non-zero if the detection encountered a failure, zero otherwise.
  68  */
  69 GPU_FUNC_QUALIFIER
  70 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)
  71
  72 /*! \brief Select the compatible GPUs
  73  *
  74  * This function selects the compatible gpus and initializes
  75  * gpu_info->dev_use and gpu_info->n_dev_use.
  76  *
  77  * Given the list of GPUs available in the system check each device in
  78  * gpu_info->gpu_dev and place the indices of the compatible GPUs into
  79  * dev_use with this marking the respective GPUs as "available for use."
  80  * Note that \p detect_gpus must have been called before.
  81  *
  82  * \param[in]     gpu_info    pointer to structure holding GPU information
  83  * \param[in,out] gpu_opt     pointer to structure holding GPU options
  84  */
  85 GPU_FUNC_QUALIFIER
  86 void pick_compatible_gpus(const struct gmx_gpu_info_t *GPU_FUNC_ARGUMENT(gpu_info),
  87                           gmx_gpu_opt_t *GPU_FUNC_ARGUMENT(gpu_opt)) GPU_FUNC_TERM
  88
  89 /*! \brief Check the existence/compatibility of a set of GPUs specified by their device IDs.
  90  *
  91  * Given the a list of gpu_opt->n_dev_use GPU device IDs stored in
  92  * gpu_opt->dev_use check the existence and compatibility
  93  * of the respective GPUs. Also provide the caller with an array containing
  94  * the result of checks in \p checkres.
  95  *
  96  * \param[out]  checkres    check result for each ID passed in requested_devs
  97  * \param[in]   gpu_info    pointer to structure holding GPU information
  98  * \param[out]  gpu_opt     pointer to structure holding GPU options
  99  * \returns                 TRUE if every the requested GPUs are compatible
 100  */
 101 GPU_FUNC_QUALIFIER
 102 gmx_bool check_selected_gpus(int *GPU_FUNC_ARGUMENT(checkres),
 103                              const struct gmx_gpu_info_t *GPU_FUNC_ARGUMENT(gpu_info),
 104                              gmx_gpu_opt_t *GPU_FUNC_ARGUMENT(gpu_opt)) GPU_FUNC_TERM_WITH_RETURN(-1)
 105
 106 /*! \brief Frees the gpu_dev and dev_use array fields of \p gpu_info.
 107  *
 108  * \param[in]    gpu_info    pointer to structure holding GPU information
 109  */
 110 GPU_FUNC_QUALIFIER
 111 void free_gpu_info(const struct gmx_gpu_info_t *GPU_FUNC_ARGUMENT(gpu_info)) GPU_FUNC_TERM
 112
 113 /*! \brief Initializes the GPU with the given index.
 114  *
 115  * The varible \p mygpu is the index of the GPU to initialize in the
 116  * gpu_info.gpu_dev array.
 117  *
 118  * \param[out] fplog        log file to write to
 119  * \param[in]  mygpu        index of the GPU to initialize
 120  * \param[out] result_str   the message related to the error that occurred
 121  *                          during the initialization (if there was any).
 122  * \param[in] gpu_info      GPU info of all detected devices in the system.
 123  * \param[in] gpu_opt       options for using the GPUs in gpu_info
 124  * \returns                 true if no error occurs during initialization.
 125  */
 126 GPU_FUNC_QUALIFIER
 127 gmx_bool init_gpu(FILE *GPU_FUNC_ARGUMENT(fplog),
 128                   int GPU_FUNC_ARGUMENT(mygpu),
 129                   char *GPU_FUNC_ARGUMENT(result_str),
 130                   const struct gmx_gpu_info_t *GPU_FUNC_ARGUMENT(gpu_info),
 131                   const gmx_gpu_opt_t *GPU_FUNC_ARGUMENT(gpu_opt)) GPU_FUNC_TERM_WITH_RETURN(-1)
 132
 133 /*! \brief Frees up the CUDA GPU used by the active context at the time of calling.
 134  *
 135  * The context is explicitly destroyed and therefore all data uploaded to the GPU
 136  * is lost. This should only be called when none of this data is required anymore.
 137  *
 138  * \param[in]  mygpu        index of the GPU clean up for
 139  * \param[out] result_str   the message related to the error that occurred
 140  *                          during the initialization (if there was any).
 141  * \param[in] gpu_info      GPU info of all detected devices in the system.
 142  * \param[in] gpu_opt       options for using the GPUs in gpu_info
 143  * \returns                 true if no error occurs during the freeing.
 144  */
 145 CUDA_FUNC_QUALIFIER
 146 gmx_bool free_cuda_gpu(int CUDA_FUNC_ARGUMENT(mygpu),
 147                        char *CUDA_FUNC_ARGUMENT(result_str),
 148                        const gmx_gpu_info_t *CUDA_FUNC_ARGUMENT(gpu_info),
 149                        const gmx_gpu_opt_t *CUDA_FUNC_ARGUMENT(gpu_opt)) CUDA_FUNC_TERM_WITH_RETURN(TRUE)
 150
 151 /*! \brief Returns the device ID of the CUDA GPU currently in use.
 152  *
 153  * The GPU used is the one that is active at the time of the call in the active context.
 154  *
 155  * \returns                 device ID of the GPU in use at the time of the call
 156  */
 157 CUDA_FUNC_QUALIFIER
 158 int get_current_cuda_gpu_device_id(void) CUDA_FUNC_TERM_WITH_RETURN(-1)
 159
 160 /*! \brief Returns an identifier for the GPU with a given index into the array of used GPUs.
 161  *
 162  * Getter function which, given an index into the array of GPUs in use
 163  * (dev_use) -- typically an MPI rank --, returns an identifier of the
 164  * respective GPU.
 165  *
 166  * \param[in]    gpu_info   Pointer to structure holding GPU information
 167  * \param[in]    gpu_opt    Pointer to structure holding GPU options
 168  * \param[in]    idx        Index into the array of used GPUs
 169  * \returns                 device ID of the requested GPU
 170  */
 171 GPU_FUNC_QUALIFIER
 172 int get_gpu_device_id(const struct gmx_gpu_info_t *GPU_FUNC_ARGUMENT(gpu_info),
 173                       const gmx_gpu_opt_t *GPU_FUNC_ARGUMENT(gpu_opt),
 174                       int GPU_FUNC_ARGUMENT(idx)) GPU_FUNC_TERM_WITH_RETURN(-1)
 175
 176 /*! \brief Returns the name for the OpenCL GPU with a given index into the array of used GPUs.
 177  *
 178  * Getter function which, given an index into the array of GPUs in use
 179  * (dev_use) -- typically a tMPI/MPI rank --, returns the device name for the
 180  * respective OpenCL GPU.
 181  *
 182  * \param[in]    gpu_info   Pointer to structure holding GPU information
 183  * \param[in]    gpu_opt    Pointer to structure holding GPU options
 184  * \param[in]    idx        Index into the array of used GPUs
 185  * \returns                 A string with the name of the requested OpenCL GPU
 186  */
 187 OPENCL_FUNC_QUALIFIER
 188 char* get_ocl_gpu_device_name(const struct gmx_gpu_info_t *OPENCL_FUNC_ARGUMENT(gpu_info),
 189                               const gmx_gpu_opt_t  *OPENCL_FUNC_ARGUMENT(gpu_opt),
 190                               int                  OPENCL_FUNC_ARGUMENT(idx)) OPENCL_FUNC_TERM_WITH_RETURN(NULL)
 191
 192 /*! \brief Formats and returns a device information string for a given GPU.
 193  *
 194  * Given an index *directly* into the array of available GPUs (gpu_dev)
 195  * returns a formatted info string for the respective GPU which includes
 196  * ID, name, compute capability, and detection status.
 197  *
 198  * \param[out]  s           pointer to output string (has to be allocated externally)
 199  * \param[in]   gpu_info    pointer to structure holding GPU information
 200  * \param[in]   index       an index *directly* into the array of available GPUs
 201  */
 202 GPU_FUNC_QUALIFIER
 203 void get_gpu_device_info_string(char *GPU_FUNC_ARGUMENT(s),
 204                                 const struct gmx_gpu_info_t *GPU_FUNC_ARGUMENT(gpu_info),
 205                                 int GPU_FUNC_ARGUMENT(index)) GPU_FUNC_TERM
 206
 207 /*! \brief Returns the size of the gpu_dev_info struct.
 208  *
 209  * The size of gpu_dev_info can be used for allocation and communication.
 210  *
 211  * \returns                 size in bytes of gpu_dev_info
 212  */
 213 GPU_FUNC_QUALIFIER
 214 size_t sizeof_gpu_dev_info(void) GPU_FUNC_TERM_WITH_RETURN(0)
 215
 216 /*! \brief Returns a pointer *ptr to page-locked memory of size nbytes.
 217  *
 218  * The allocated memory is suitable to be used for data transfers between host
 219  * and GPU.
 220  * Error handling should be done within this function.
 221  */
 222 typedef void gmx_host_alloc_t (void **ptr, size_t nbytes);
 223
 224 /*! \brief Frees page-locked memory pointed to by *ptr.
 225  *
 226  * NULL should not be passed to this function.
 227  */
 228 typedef void gmx_host_free_t (void *ptr);
 229
 230 /*! \brief Set page-locked memory allocation functions used by the GPU host. */
 231 void gpu_set_host_malloc_and_free(bool               bUseGpuKernels,
 232                                   gmx_host_alloc_t **nb_alloc,
 233                                   gmx_host_free_t  **nb_free);
 234
 235
 236
 237 /*! \brief Starts the GPU profiler if mdrun is being profiled.
 238  *
 239  *  When a profiler run is in progress (based on the presence of the NVPROF_ID
 240  *  env. var.), the profiler is started to begin collecting data during the
 241  *  rest of the run (or until stopGpuProfiler is called).
 242  *
 243  *  Note that this is implemented only for the CUDA API.
 244  */
 245 CUDA_FUNC_QUALIFIER
 246 void startGpuProfiler(void) CUDA_FUNC_TERM
 247
 248
 249 /*! \brief Resets the GPU profiler if mdrun is being profiled.
 250  *
 251  * When a profiler run is in progress (based on the presence of the NVPROF_ID
 252  * env. var.), the profiler data is restet in order to eliminate the data collected
 253  * from the preceding part fo the run.
 254  *
 255  * This function should typically be called at the mdrun counter reset time.
 256  *
 257  * Note that this is implemented only for the CUDA API.
 258  */
 259 CUDA_FUNC_QUALIFIER
 260 void resetGpuProfiler(void) CUDA_FUNC_TERM
 261
 262
 263 /*! \brief Stops the CUDA profiler if mdrun is being profiled.
 264  *
 265  *  This function can be called at cleanup when skipping recording
 266  *  recording subsequent API calls from being traces/profiled is desired,
 267  *  e.g. before uninitialization.
 268  *
 269  *  Note that this is implemented only for the CUDA API.
 270  */
 271 CUDA_FUNC_QUALIFIER
 272 void stopGpuProfiler(void) CUDA_FUNC_TERM
 273
 274
 275 #endif