src/gromacs/gpu_utils/ocl_compiler.h

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2016 by the GROMACS development team.
   5  * Copyright (c) 2017,2018,2019,2020, by the GROMACS development team, led by
   6  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   7  * and including many others, as listed in the AUTHORS file in the
   8  * top-level source directory and at http://www.gromacs.org.
   9  *
  10  * GROMACS is free software; you can redistribute it and/or
  11  * modify it under the terms of the GNU Lesser General Public License
  12  * as published by the Free Software Foundation; either version 2.1
  13  * of the License, or (at your option) any later version.
  14  *
  15  * GROMACS is distributed in the hope that it will be useful,
  16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  18  * Lesser General Public License for more details.
  19  *
  20  * You should have received a copy of the GNU Lesser General Public
  21  * License along with GROMACS; if not, see
  22  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  23  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  24  *
  25  * If you want to redistribute modifications to GROMACS, please
  26  * consider that scientific software is very special. Version
  27  * control is crucial - bugs must be traceable. We will be happy to
  28  * consider code for inclusion in the official distribution, but
  29  * derived work must not be called official GROMACS. Details are found
  30  * in the README & COPYING files - if they are missing, get the
  31  * official version at http://www.gromacs.org.
  32  *
  33  * To help us fund GROMACS development, we humbly ask that you cite
  34  * the research papers on the package. Check out http://www.gromacs.org.
  35  */
  36 /*! \libinternal \file
  37  *  \brief Declare infrastructure for OpenCL JIT compilation
  38  *
  39  *  \author Dimitrios Karkoulis <dimitris.karkoulis@gmail.com>
  40  *  \author Anca Hamuraru <anca@streamcomputing.eu>
  41  *  \author Teemu Virolainen <teemu@streamcomputing.eu>
  42  *  \author Mark Abraham <mark.j.abraham@gmail.com>
  43  *  \inlibraryapi
  44  */
  45 #ifndef GMX_GPU_UTILS_OCL_COMPILER_H
  46 #define GMX_GPU_UTILS_OCL_COMPILER_H
  47
  48 #include <string>
  49
  50 #include "gromacs/gpu_utils/oclutils.h"
  51 #include "gromacs/hardware/device_information.h"
  52
  53 namespace gmx
  54 {
  55 namespace ocl
  56 {
  57
  58 /*! \brief Get the device-specific warp size
  59  *
  60  *  This is platform implementation dependent and seems to only work on the Nvidia and AMD
  61  * platforms! Nvidia reports 32, AMD for GPU 64. Intel seems to report 16, but that is not correct,
  62  *  as it execution width can be between 8-32 and it's picked per-kernel at compile-time.
  63  *  Therefore, for Intel it should actually be queried separately for each kernel (Issue #2520).
  64  *
  65  *  \param  context   Current OpenCL context
  66  *  \param  deviceId OpenCL device with the context
  67  *  \return cl_int value of the warp size
  68  *
  69  * \throws InternalError if an OpenCL error was encountered
  70  */
  71 size_t getDeviceWarpSize(cl_context context, cl_device_id deviceId);
  72
  73
  74 /*! \brief Get the kernel-specific warp size
  75  *
  76  *  \param  kernel   THe OpenCL kernel object
  77  *  \param  deviceId OpenCL device for which the kernel warp size is queried
  78  *  \return cl_int value of the warp size
  79  *
  80  * \throws InternalError if an OpenCL error was encountered
  81  */
  82 size_t getKernelWarpSize(cl_kernel kernel, cl_device_id deviceId);
  83
  84 /*! \brief Compile the specified kernel for the context and device.
  85  *
  86  * \param[out] fplog                 Open file pointer for log output
  87  * \param[in]  kernelRelativePath    Relative path to the kernel in the source tree,
  88  *                                   e.g. "src/gromacs/mdlib/nbnxn_ocl" for NB kernels.
  89  * \param[in]  kernelBaseFilename    The name of the kernel source file to compile, e.g.
  90  *                                   "nbnxn_ocl_kernels.cl"
  91  * \param[in]  extraDefines          Preprocessor defines required by the calling code,
  92  *                                   e.g. for configuring the kernels
  93  * \param[in]  context               OpenCL context on the device to compile for
  94  * \param[in]  deviceId              OpenCL device id of the device to compile for
  95  * \param[in]  deviceVendor          Enumerator of the device vendor to compile for
  96  *
  97  * \returns The compiled OpenCL program
  98  *
  99  * \todo Consider whether we can parallelize the compilation of all
 100  * the kernels by compiling them in separate programs - but since the
 101  * resulting programs can't refer to each other, that might lead to
 102  * bloat of util code?
 103  *
 104  * \throws std::bad_alloc  if out of memory.
 105  *         FileIOError     if a file I/O error prevents returning a valid compiled program.
 106  *         InternalError   if an OpenCL API error prevents returning a valid compiled program. */
 107 cl_program compileProgram(FILE*              fplog,
 108                           const std::string& kernelRelativePath,
 109                           const std::string& kernelBaseFilename,
 110                           const std::string& extraDefines,
 111                           cl_context         context,
 112                           cl_device_id       deviceId,
 113                           DeviceVendor       deviceVendor);
 114
 115 } // namespace ocl
 116 } // namespace gmx
 117
 118 #endif