src/gromacs/gpu_utils/gpuregiontimer_ocl.h

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2016,2017, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35
  36 /*! \libinternal \file
  37  *  \brief Implements the GPU region timer for OpenCL.
  38  *
  39  *  \author Aleksei Iupinov <a.yupinov@gmail.com>
  40  *
  41  *  \inlibraryapi
  42  */
  43
  44 #ifndef GMX_GPU_UTILS_GPUREGIONTIMER_OCL_H
  45 #define GMX_GPU_UTILS_GPUREGIONTIMER_OCL_H
  46
  47 #include <array>
  48
  49 #include "gromacs/gpu_utils/oclutils.h"
  50
  51 #include "gpuregiontimer.h"
  52
  53 template <> struct GpuTraits<GpuFramework::OpenCL>
  54 {
  55     using CommandStream = cl_command_queue;
  56     using CommandEvent  = cl_event;
  57 };
  58
  59 //! Short-hand for external use
  60 using GpuRegionTimer = GpuRegionTimerWrapper<GpuFramework::OpenCL>;
  61
  62 // cppcheck-suppress noConstructor
  63 template <> class GpuRegionTimerImpl<GpuFramework::OpenCL>
  64 {
  65     //! Short-hands
  66     using CommandStream = typename GpuTraits<GpuFramework::OpenCL>::CommandStream;
  67     using CommandEvent  = typename GpuTraits<GpuFramework::OpenCL>::CommandEvent;
  68
  69     /*! \brief The underlying individual timing events array.
  70      * The maximum size is chosen arbitrarily to work with current code, and can be changed.
  71      * There is simply no need for run-time resizing, and it's unlikely we'll ever need more than 10.
  72      */
  73     std::array<cl_event, 10> events_ = {{nullptr}};
  74     //! Index of the active event
  75     size_t                   currentEvent_ = 0;
  76
  77     public:
  78         GpuRegionTimerImpl()  = default;
  79         ~GpuRegionTimerImpl() = default;
  80         inline void openTimingRegion(CommandStream){}
  81         inline void closeTimingRegion(CommandStream){}
  82
  83         inline double getLastRangeTime()
  84         {
  85             double milliseconds = 0.0;
  86             for (size_t i = 0; i < currentEvent_; i++)
  87             {
  88                 if (events_[i]) // This conditional is ugly, but is required to make some tests (e.g. empty domain) pass
  89                 {
  90                     cl_ulong          start_ns, end_ns;
  91                     cl_int gmx_unused cl_error;
  92
  93                     cl_error = clGetEventProfilingInfo(events_[i], CL_PROFILING_COMMAND_START,
  94                                                        sizeof(cl_ulong), &start_ns, nullptr);
  95                     GMX_ASSERT(CL_SUCCESS == cl_error, "GPU timing update failure");
  96                     cl_error = clGetEventProfilingInfo(events_[i], CL_PROFILING_COMMAND_END,
  97                                                        sizeof(cl_ulong), &end_ns, nullptr);
  98                     GMX_ASSERT(CL_SUCCESS == cl_error, "GPU timing update failure");
  99                     milliseconds += (end_ns - start_ns) / 1000000.0;
 100                 }
 101             }
 102             reset();
 103             return milliseconds;
 104         }
 105
 106         inline void reset()
 107         {
 108             for (size_t i = 0; i < currentEvent_; i++)
 109             {
 110                 if (events_[i]) // This conditional is ugly, but is required to make some tests (e.g. empty domain) pass
 111                 {
 112                     GMX_ASSERT(CL_SUCCESS == clReleaseEvent(events_[i]), "OpenCL event release failure");
 113                 }
 114             }
 115             currentEvent_ = 0;
 116             // As long as we're doing nullptr checks, we might want to be extra cautious.
 117             events_.fill(nullptr);
 118         }
 119
 120         inline CommandEvent *fetchNextEvent()
 121         {
 122             GMX_ASSERT(currentEvent_ < events_.size(), "Increase c_maxEventNumber_ if needed");
 123             cl_event *result = &events_[currentEvent_];
 124             currentEvent_++;
 125             return result;
 126         }
 127 };
 128
 129 #endif