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.
   7  * Copyright (c) 2017,2018,2019,2020, by the GROMACS development team, led by
   8  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   9  * and including many others, as listed in the AUTHORS file in the
  10  * top-level source directory and at http://www.gromacs.org.
  11  *
  12  * GROMACS is free software; you can redistribute it and/or
  13  * modify it under the terms of the GNU Lesser General Public License
  14  * as published by the Free Software Foundation; either version 2.1
  15  * of the License, or (at your option) any later version.
  16  *
  17  * GROMACS is distributed in the hope that it will be useful,
  18  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  19  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  20  * Lesser General Public License for more details.
  21  *
  22  * You should have received a copy of the GNU Lesser General Public
  23  * License along with GROMACS; if not, see
  24  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  25  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  26  *
  27  * If you want to redistribute modifications to GROMACS, please
  28  * consider that scientific software is very special. Version
  29  * control is crucial - bugs must be traceable. We will be happy to
  30  * consider code for inclusion in the official distribution, but
  31  * derived work must not be called official GROMACS. Details are found
  32  * in the README & COPYING files - if they are missing, get the
  33  * official version at http://www.gromacs.org.
  34  *
  35  * To help us fund GROMACS development, we humbly ask that you cite
  36  * the research papers on the package. Check out http://www.gromacs.org.
  37  */
  38 /*! \libinternal \file
  39  *  \brief Declare functions for detection and initialization for GPU devices.
  40  *
  41  *  \author Szilard Pall <pall.szilard@gmail.com>
  42  *  \author Mark Abraham <mark.j.abraham@gmail.com>
  43  *
  44  *  \inlibraryapi
  45  */
  46 #ifndef GMX_GPU_UTILS_GPU_UTILS_H
  47 #define GMX_GPU_UTILS_GPU_UTILS_H
  48
  49 #include <cstdio>
  50
  51 #include <string>
  52 #include <vector>
  53
  54 #include "gromacs/gpu_utils/gpu_macros.h"
  55 #include "gromacs/utility/basedefinitions.h"
  56
  57 struct DeviceInformation;
  58 struct gmx_gpu_info_t;
  59
  60 namespace gmx
  61 {
  62 class MDLogger;
  63 }
  64
  65 //! Enum which is only used to describe transfer calls at the moment
  66 enum class GpuApiCallBehavior
  67 {
  68     Sync,
  69     Async
  70 };
  71
  72 //! Types of actions associated to waiting or checking the completion of GPU tasks
  73 enum class GpuTaskCompletion
  74 {
  75     Wait, /*<< Issue a blocking wait for the task */
  76     Check /*<< Only check whether the task has completed */
  77 };
  78
  79 /*! \brief Return whether GPUs can be detected
  80  *
  81  * Returns true when this is a build of \Gromacs configured to support
  82  * GPU usage, GPU detection is not disabled by an environment variable
  83  * and a valid device driver, ICD, and/or runtime was detected.
  84  * Does not throw. */
  85 bool canPerformGpuDetection();
  86
  87 /*! \brief Return whether GPU detection is functioning correctly
  88  *
  89  * Returns true when this is a build of \Gromacs configured to support
  90  * GPU usage, and a valid device driver, ICD, and/or runtime was detected.
  91  *
  92  * This function is not intended to be called from build
  93  * configurations that do not support GPUs, and there will be no
  94  * descriptive message in that case.
  95  *
  96  * \param[out] errorMessage  When returning false on a build configured with
  97  *                           GPU support and non-nullptr was passed,
  98  *                           the string contains a descriptive message about
  99  *                           why GPUs cannot be detected.
 100  *
 101  * Does not throw. */
 102 GPU_FUNC_QUALIFIER
 103 bool isGpuDetectionFunctional(std::string* GPU_FUNC_ARGUMENT(errorMessage))
 104         GPU_FUNC_TERM_WITH_RETURN(false);
 105
 106 /*! \brief Find all GPUs in the system.
 107  *
 108  *  Will detect every GPU supported by the device driver in use.
 109  *  Must only be called if canPerformGpuDetection() has returned true.
 110  *  This routine also checks for the compatibility of each and fill the
 111  *  gpu_info->deviceInfo array with the required information on each the
 112  *  device: ID, device properties, status.
 113  *
 114  *  Note that this function leaves the GPU runtime API error state clean;
 115  *  this is implemented ATM in the CUDA flavor.
 116  *  TODO: check if errors do propagate in OpenCL as they do in CUDA and
 117  *  whether there is a mechanism to "clear" them.
 118  *
 119  *  \param[in] gpu_info    pointer to structure holding GPU information.
 120  *
 121  *  \throws                InternalError if a GPU API returns an unexpected failure (because
 122  *                         the call to canDetectGpus() should always prevent this occuring)
 123  */
 124 GPU_FUNC_QUALIFIER
 125 void findGpus(gmx_gpu_info_t* GPU_FUNC_ARGUMENT(gpu_info)) GPU_FUNC_TERM;
 126
 127 /*! \brief Return a container of the detected GPUs that are compatible.
 128  *
 129  * This function filters the result of the detection for compatible
 130  * GPUs, based on the previously run compatibility tests.
 131  *
 132  * \param[in]     gpu_info    Information detected about GPUs, including compatibility.
 133  * \return                    vector of IDs of GPUs already recorded as compatible */
 134 std::vector<int> getCompatibleGpus(const gmx_gpu_info_t& gpu_info);
 135
 136 /*! \brief Return a string describing how compatible the GPU with given \c index is.
 137  *
 138  * \param[in]   gpu_info    Information about detected GPUs
 139  * \param[in]   index       index of GPU to ask about
 140  * \returns                 A null-terminated C string describing the compatibility status, useful for error messages.
 141  */
 142 const char* getGpuCompatibilityDescription(const gmx_gpu_info_t& gpu_info, int index);
 143
 144 /*! \brief Frees the gpu_dev and dev_use array fields of \p gpu_info.
 145  *
 146  * \param[in]    gpu_info    pointer to structure holding GPU information
 147  */
 148 void free_gpu_info(const gmx_gpu_info_t* gpu_info);
 149
 150 /*! \brief Initializes the GPU described by \c deviceInfo.
 151  *
 152  * TODO Doxygen complains about these - probably a Doxygen bug, since
 153  * the patterns here are the same as elsewhere in this header.
 154  *
 155  * \param[in]    deviceInfo   device info of the GPU to initialize
 156  *
 157  * Issues a fatal error for any critical errors that occur during
 158  * initialization.
 159  */
 160 GPU_FUNC_QUALIFIER
 161 void init_gpu(const DeviceInformation* GPU_FUNC_ARGUMENT(deviceInfo)) GPU_FUNC_TERM;
 162
 163 /*! \brief Frees up the CUDA GPU used by the active context at the time of calling.
 164  *
 165  * If \c deviceInfo is nullptr, then it is understood that no device
 166  * was selected so no context is active to be freed. Otherwise, the
 167  * context is explicitly destroyed and therefore all data uploaded to
 168  * the GPU is lost. This must only be called when none of this data is
 169  * required anymore, because subsequent attempts to free memory
 170  * associated with the context will otherwise fail.
 171  *
 172  * Calls gmx_warning upon errors.
 173  *
 174  * \param[in]  deviceInfo   device info of the GPU to clean up for
 175  *
 176  * \returns                 true if no error occurs during the freeing.
 177  */
 178 CUDA_FUNC_QUALIFIER
 179 void free_gpu(const DeviceInformation* CUDA_FUNC_ARGUMENT(deviceInfo)) CUDA_FUNC_TERM;
 180
 181 /*! \brief Return a pointer to the device info for \c deviceId
 182  *
 183  * \param[in] gpu_info      GPU info of all detected devices in the system.
 184  * \param[in] deviceId      ID for the GPU device requested.
 185  *
 186  * \returns                 Pointer to the device info for \c deviceId.
 187  */
 188 GPU_FUNC_QUALIFIER
 189 DeviceInformation* getDeviceInfo(const gmx_gpu_info_t& GPU_FUNC_ARGUMENT(gpu_info),
 190                                  int GPU_FUNC_ARGUMENT(deviceId)) GPU_FUNC_TERM_WITH_RETURN(nullptr);
 191
 192 /*! \brief Returns the device ID of the CUDA GPU currently in use.
 193  *
 194  * The GPU used is the one that is active at the time of the call in the active context.
 195  *
 196  * \returns                 device ID of the GPU in use at the time of the call
 197  */
 198 CUDA_FUNC_QUALIFIER
 199 int get_current_cuda_gpu_device_id() CUDA_FUNC_TERM_WITH_RETURN(-1);
 200
 201 /*! \brief Formats and returns a device information string for a given GPU.
 202  *
 203  * Given an index *directly* into the array of available GPUs (gpu_dev)
 204  * returns a formatted info string for the respective GPU which includes
 205  * ID, name, compute capability, and detection status.
 206  *
 207  * \param[out]  s           pointer to output string (has to be allocated externally)
 208  * \param[in]   gpu_info    Information about detected GPUs
 209  * \param[in]   index       an index *directly* into the array of available GPUs
 210  */
 211 GPU_FUNC_QUALIFIER
 212 void get_gpu_device_info_string(char*                 GPU_FUNC_ARGUMENT(s),
 213                                 const gmx_gpu_info_t& GPU_FUNC_ARGUMENT(gpu_info),
 214                                 int                   GPU_FUNC_ARGUMENT(index)) GPU_FUNC_TERM;
 215
 216
 217 /*! \brief Returns the size of the gpu_dev_info struct.
 218  *
 219  * The size of gpu_dev_info can be used for allocation and communication.
 220  *
 221  * \returns                 size in bytes of gpu_dev_info
 222  */
 223 GPU_FUNC_QUALIFIER
 224 size_t sizeof_gpu_dev_info() GPU_FUNC_TERM_WITH_RETURN(0);
 225
 226 //! Get status of device with specified index
 227 int gpu_info_get_stat(const gmx_gpu_info_t& info, int index);
 228
 229 /*! \brief Check if GROMACS has been built with GPU support.
 230  *
 231  * \param[in] error Pointer to error string or nullptr.
 232  * \todo Move this to NB module once it exists.
 233  */
 234 bool buildSupportsNonbondedOnGpu(std::string* error);
 235
 236 /*! \brief Starts the GPU profiler if mdrun is being profiled.
 237  *
 238  *  When a profiler run is in progress (based on the presence of the NVPROF_ID
 239  *  env. var.), the profiler is started to begin collecting data during the
 240  *  rest of the run (or until stopGpuProfiler is called).
 241  *
 242  *  Note that this is implemented only for the CUDA API.
 243  */
 244 CUDA_FUNC_QUALIFIER
 245 void startGpuProfiler() CUDA_FUNC_TERM;
 246
 247
 248 /*! \brief Resets the GPU profiler if mdrun is being profiled.
 249  *
 250  * When a profiler run is in progress (based on the presence of the NVPROF_ID
 251  * env. var.), the profiler data is restet in order to eliminate the data collected
 252  * from the preceding part fo the run.
 253  *
 254  * This function should typically be called at the mdrun counter reset time.
 255  *
 256  * Note that this is implemented only for the CUDA API.
 257  */
 258 CUDA_FUNC_QUALIFIER
 259 void resetGpuProfiler() CUDA_FUNC_TERM;
 260
 261
 262 /*! \brief Stops the CUDA profiler if mdrun is being profiled.
 263  *
 264  *  This function can be called at cleanup when skipping recording
 265  *  recording subsequent API calls from being traces/profiled is desired,
 266  *  e.g. before uninitialization.
 267  *
 268  *  Note that this is implemented only for the CUDA API.
 269  */
 270 CUDA_FUNC_QUALIFIER
 271 void stopGpuProfiler() CUDA_FUNC_TERM;
 272
 273 //! Tells whether the host buffer was pinned for non-blocking transfers. Only implemented for CUDA.
 274 CUDA_FUNC_QUALIFIER
 275 bool isHostMemoryPinned(const void* CUDA_FUNC_ARGUMENT(h_ptr)) CUDA_FUNC_TERM_WITH_RETURN(false);
 276
 277 /*! \brief Enable peer access between GPUs where supported
 278  * \param[in] gpuIdsToUse   List of GPU IDs in use
 279  * \param[in] mdlog         Logger object
 280  */
 281 CUDA_FUNC_QUALIFIER
 282 void setupGpuDevicePeerAccess(const std::vector<int>& CUDA_FUNC_ARGUMENT(gpuIdsToUse),
 283                               const gmx::MDLogger&    CUDA_FUNC_ARGUMENT(mdlog)) CUDA_FUNC_TERM;
 284
 285 #endif