2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2018,2019,2020,2021, 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.
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.
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.
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.
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.
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.
35 /*! \libinternal \file
36 * \brief Implements a GpuEventSynchronizer class for OpenCL.
38 * \author Aleksei Iupinov <a.yupinov@gmail.com>
41 #ifndef GMX_GPU_UTILS_GPUEVENTSYNCHRONIZER_OCL_H
42 #define GMX_GPU_UTILS_GPUEVENTSYNCHRONIZER_OCL_H
46 # include "gromacs/gpu_utils/gputraits_ocl.h"
47 # include "gromacs/gpu_utils/oclutils.h"
48 # include "gromacs/utility/exceptions.h"
49 # include "gromacs/utility/gmxassert.h"
51 /*! \libinternal \brief
52 * A class which allows for CPU thread to mark and wait for certain GPU stream execution point.
53 * The event can be put into the stream with markEvent() and then later waited on with waitForEvent().
54 * This can be repeated as necessary, but the current implementation does not allow waiting on
55 * completed event more than once, expecting only exact pairs of markEvent(stream); waitForEvent().
56 * The class generally attempts to track the correctness of its state transitions, but
57 * please note that calling waitForEvent() right after the construction will fail with OpenCL but succeed with CUDA.
59 * Another possible mode of operation can be implemented if needed:
60 * multiple calls to waitForEvent() after a single markEvent(). For this, clReleaseEvent() call
61 * from waitForEvent() should instead happen conditionally at the beginning of markEvent(), replacing
62 * the GMX_ASSERT(). This was tested to work both with CUDA and NVidia OpenCL, but not with AMD/Intel OpenCl.
64 class GpuEventSynchronizer
68 GpuEventSynchronizer() : event_(nullptr) {}
70 ~GpuEventSynchronizer()
72 // This additional code only prevents cl_event leak in an unlikely situation of destructor
73 // being called after markEvent() but before waitForEvent() / enqueueWaitEvent().
76 clReleaseEvent(event_);
80 GpuEventSynchronizer(const GpuEventSynchronizer&) = delete;
82 GpuEventSynchronizer& operator=(GpuEventSynchronizer&&) = delete;
83 //! Moving is disabled but can be considered in the future if needed
84 GpuEventSynchronizer(GpuEventSynchronizer&&) = delete;
86 /*! \brief Marks the synchronization point in the \p stream.
87 * Should be called first and then followed by waitForEvent().
89 inline void markEvent(const DeviceStream& deviceStream)
91 GMX_ASSERT(nullptr == event_, "Do not call markEvent more than once!");
92 cl_int clError = clEnqueueMarkerWithWaitList(deviceStream.stream(), 0, nullptr, &event_);
93 if (CL_SUCCESS != clError)
95 GMX_THROW(gmx::InternalError("Failed to enqueue the GPU synchronization event: "
96 + ocl_get_error_string(clError)));
99 /*! \brief Synchronizes the host thread on the marked event. */
100 inline void waitForEvent()
102 cl_int clError = clWaitForEvents(1, &event_);
103 if (CL_SUCCESS != clError)
105 GMX_THROW(gmx::InternalError("Failed to synchronize on the GPU event: "
106 + ocl_get_error_string(clError)));
111 /*! \brief Checks the completion of the underlying event and resets the object if it was. */
112 inline bool isReady()
115 cl_int clError = clGetEventInfo(
116 event_, CL_EVENT_COMMAND_EXECUTION_STATUS, sizeof(cl_int), &result, nullptr);
117 if (CL_SUCCESS != clError)
119 GMX_THROW(gmx::InternalError("Failed to retrieve event info: " + ocl_get_error_string(clError)));
121 bool hasTriggered = (result == CL_COMPLETE);
128 /*! \brief Enqueues a wait for the recorded event in stream \p stream
130 * After enqueue, the associated event is released, so this method should
131 * be only called once per markEvent() call.
133 inline void enqueueWaitEvent(const DeviceStream& deviceStream)
135 cl_int clError = clEnqueueBarrierWithWaitList(deviceStream.stream(), 1, &event_, nullptr);
136 if (CL_SUCCESS != clError)
138 GMX_THROW(gmx::InternalError("Failed to enqueue device barrier for the GPU event: "
139 + ocl_get_error_string(clError)));
145 //! Reset (release) the event to unmarked state.
148 cl_int clError = clReleaseEvent(event_);
149 if (CL_SUCCESS != clError)
151 GMX_THROW(gmx::InternalError("Failed to release the GPU event: "
152 + ocl_get_error_string(clError)));