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_gpu_info_t;
  54 struct gmx_gpu_opt_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    pointer to structure holding GPU information
  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    pointer to structure holding GPU information
  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 with the given index.
 105  *
 106  * The varible \p mygpu is the index of the GPU to initialize in the
 107  * gpu_info.gpu_dev array.
 108  *
 109  * \param      mdlog        log file to write to
 110  * \param[in]  mygpu        index of the GPU to initialize
 111  * \param[out] result_str   the message related to the error that occurred
 112  *                          during the initialization (if there was any).
 113  * \param[in] gpu_info      GPU info of all detected devices in the system.
 114  * \param[in] gpu_opt       options for using the GPUs in gpu_info
 115  * \returns                 true if no error occurs during initialization.
 116  */
 117 GPU_FUNC_QUALIFIER
 118 gmx_bool init_gpu(const gmx::MDLogger &GPU_FUNC_ARGUMENT(mdlog),
 119                   int GPU_FUNC_ARGUMENT(mygpu),
 120                   char *GPU_FUNC_ARGUMENT(result_str),
 121                   const struct gmx_gpu_info_t *GPU_FUNC_ARGUMENT(gpu_info),
 122                   const gmx_gpu_opt_t *GPU_FUNC_ARGUMENT(gpu_opt)) GPU_FUNC_TERM_WITH_RETURN(-1)
 123
 124 /*! \brief Frees up the CUDA GPU used by the active context at the time of calling.
 125  *
 126  * The context is explicitly destroyed and therefore all data uploaded to the GPU
 127  * is lost. This should only be called when none of this data is required anymore.
 128  *
 129  * \param[in]  mygpu        index of the GPU clean up for
 130  * \param[out] result_str   the message related to the error that occurred
 131  *                          during the initialization (if there was any).
 132  * \param[in] gpu_info      GPU info of all detected devices in the system.
 133  * \param[in] gpu_opt       options for using the GPUs in gpu_info
 134  * \returns                 true if no error occurs during the freeing.
 135  */
 136 CUDA_FUNC_QUALIFIER
 137 gmx_bool free_cuda_gpu(int CUDA_FUNC_ARGUMENT(mygpu),
 138                        char *CUDA_FUNC_ARGUMENT(result_str),
 139                        const gmx_gpu_info_t *CUDA_FUNC_ARGUMENT(gpu_info),
 140                        const gmx_gpu_opt_t *CUDA_FUNC_ARGUMENT(gpu_opt)) CUDA_FUNC_TERM_WITH_RETURN(TRUE)
 141
 142 /*! \brief Returns the device ID of the CUDA GPU currently in use.
 143  *
 144  * The GPU used is the one that is active at the time of the call in the active context.
 145  *
 146  * \returns                 device ID of the GPU in use at the time of the call
 147  */
 148 CUDA_FUNC_QUALIFIER
 149 int get_current_cuda_gpu_device_id(void) CUDA_FUNC_TERM_WITH_RETURN(-1)
 150
 151 /*! \brief Returns an identifier for the GPU with a given index into the array of used GPUs.
 152  *
 153  * Getter function which, given an index into the array of GPUs in use
 154  * (dev_use) -- typically an MPI rank --, returns an identifier of the
 155  * respective GPU.
 156  *
 157  * \param[in]    gpu_info   Pointer to structure holding GPU information
 158  * \param[in]    gpu_opt    Pointer to structure holding GPU options
 159  * \param[in]    idx        Index into the array of used GPUs
 160  * \returns                 device ID of the requested GPU
 161  */
 162 GPU_FUNC_QUALIFIER
 163 int get_gpu_device_id(const struct gmx_gpu_info_t *GPU_FUNC_ARGUMENT(gpu_info),
 164                       const gmx_gpu_opt_t *GPU_FUNC_ARGUMENT(gpu_opt),
 165                       int GPU_FUNC_ARGUMENT(idx)) GPU_FUNC_TERM_WITH_RETURN(-1)
 166
 167 /*! \brief Returns the name for the OpenCL GPU with a given index into the array of used GPUs.
 168  *
 169  * Getter function which, given an index into the array of GPUs in use
 170  * (dev_use) -- typically a tMPI/MPI rank --, returns the device name for the
 171  * respective OpenCL GPU.
 172  *
 173  * \param[in]    gpu_info   Pointer to structure holding GPU information
 174  * \param[in]    gpu_opt    Pointer to structure holding GPU options
 175  * \param[in]    idx        Index into the array of used GPUs
 176  * \returns                 A string with the name of the requested OpenCL GPU
 177  */
 178 OPENCL_FUNC_QUALIFIER
 179 char* get_ocl_gpu_device_name(const struct gmx_gpu_info_t *OPENCL_FUNC_ARGUMENT(gpu_info),
 180                               const gmx_gpu_opt_t  *OPENCL_FUNC_ARGUMENT(gpu_opt),
 181                               int                  OPENCL_FUNC_ARGUMENT(idx)) OPENCL_FUNC_TERM_WITH_RETURN(NULL)
 182
 183 /*! \brief Formats and returns a device information string for a given GPU.
 184  *
 185  * Given an index *directly* into the array of available GPUs (gpu_dev)
 186  * returns a formatted info string for the respective GPU which includes
 187  * ID, name, compute capability, and detection status.
 188  *
 189  * \param[out]  s           pointer to output string (has to be allocated externally)
 190  * \param[in]   gpu_info    pointer to structure holding GPU information
 191  * \param[in]   index       an index *directly* into the array of available GPUs
 192  */
 193 GPU_FUNC_QUALIFIER
 194 void get_gpu_device_info_string(char *GPU_FUNC_ARGUMENT(s),
 195                                 const struct gmx_gpu_info_t *GPU_FUNC_ARGUMENT(gpu_info),
 196                                 int GPU_FUNC_ARGUMENT(index)) GPU_FUNC_TERM
 197
 198 /*! \brief Returns the size of the gpu_dev_info struct.
 199  *
 200  * The size of gpu_dev_info can be used for allocation and communication.
 201  *
 202  * \returns                 size in bytes of gpu_dev_info
 203  */
 204 GPU_FUNC_QUALIFIER
 205 size_t sizeof_gpu_dev_info(void) GPU_FUNC_TERM_WITH_RETURN(0)
 206
 207 /*! \brief Returns a pointer *ptr to page-locked memory of size nbytes.
 208  *
 209  * The allocated memory is suitable to be used for data transfers between host
 210  * and GPU.
 211  * Error handling should be done within this function.
 212  */
 213 typedef void gmx_host_alloc_t (void **ptr, size_t nbytes);
 214
 215 /*! \brief Frees page-locked memory pointed to by *ptr.
 216  *
 217  * NULL should not be passed to this function.
 218  */
 219 typedef void gmx_host_free_t (void *ptr);
 220
 221 /*! \brief Set page-locked memory allocation functions used by the GPU host. */
 222 void gpu_set_host_malloc_and_free(bool               bUseGpuKernels,
 223                                   gmx_host_alloc_t **nb_alloc,
 224                                   gmx_host_free_t  **nb_free);
 225
 226
 227
 228 /*! \brief Starts the GPU profiler if mdrun is being profiled.
 229  *
 230  *  When a profiler run is in progress (based on the presence of the NVPROF_ID
 231  *  env. var.), the profiler is started to begin collecting data during the
 232  *  rest of the run (or until stopGpuProfiler is called).
 233  *
 234  *  Note that this is implemented only for the CUDA API.
 235  */
 236 CUDA_FUNC_QUALIFIER
 237 void startGpuProfiler(void) CUDA_FUNC_TERM
 238
 239
 240 /*! \brief Resets the GPU profiler if mdrun is being profiled.
 241  *
 242  * When a profiler run is in progress (based on the presence of the NVPROF_ID
 243  * env. var.), the profiler data is restet in order to eliminate the data collected
 244  * from the preceding part fo the run.
 245  *
 246  * This function should typically be called at the mdrun counter reset time.
 247  *
 248  * Note that this is implemented only for the CUDA API.
 249  */
 250 CUDA_FUNC_QUALIFIER
 251 void resetGpuProfiler(void) CUDA_FUNC_TERM
 252
 253
 254 /*! \brief Stops the CUDA profiler if mdrun is being profiled.
 255  *
 256  *  This function can be called at cleanup when skipping recording
 257  *  recording subsequent API calls from being traces/profiled is desired,
 258  *  e.g. before uninitialization.
 259  *
 260  *  Note that this is implemented only for the CUDA API.
 261  */
 262 CUDA_FUNC_QUALIFIER
 263 void stopGpuProfiler(void) CUDA_FUNC_TERM
 264
 265
 266 #endif