2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2018,2019, 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 #ifndef GMX_GPU_UTILS_DEVICEBUFFER_OCL_H
36 #define GMX_GPU_UTILS_DEVICEBUFFER_OCL_H
38 /*! \libinternal \file
39 * \brief Implements the DeviceBuffer type and routines for OpenCL.
40 * Should only be included directly by the main DeviceBuffer file devicebuffer.h.
41 * TODO: the intent is for DeviceBuffer to become a class.
43 * \author Aleksei Iupinov <a.yupinov@gmail.com>
48 #include "gromacs/gpu_utils/devicebuffer_datatype.h"
49 #include "gromacs/gpu_utils/gpu_utils.h" //only for GpuApiCallBehavior
50 #include "gromacs/gpu_utils/gputraits_ocl.h"
51 #include "gromacs/utility/gmxassert.h"
53 /*! \libinternal \brief
54 * Allocates a device-side buffer.
55 * It is currently a caller's responsibility to call it only on not-yet allocated buffers.
57 * \tparam ValueType Raw value type of the \p buffer.
58 * \param[in,out] buffer Pointer to the device-side buffer.
59 * \param[in] numValues Number of values to accomodate.
60 * \param[in] context The buffer's context-to-be.
62 template <typename ValueType>
63 void allocateDeviceBuffer(DeviceBuffer<ValueType> *buffer,
67 GMX_ASSERT(buffer, "needs a buffer pointer");
68 void *hostPtr = nullptr;
70 *buffer = clCreateBuffer(context, CL_MEM_READ_WRITE, numValues * sizeof(ValueType), hostPtr, &clError);
71 GMX_RELEASE_ASSERT(clError == CL_SUCCESS, "clCreateBuffer failure");
75 * Frees a device-side buffer.
76 * This does not reset separately stored size/capacity integers,
77 * as this is planned to be a destructor of DeviceBuffer as a proper class,
78 * and no calls on \p buffer should be made afterwards.
80 * \param[in] buffer Pointer to the buffer to free.
82 template <typename DeviceBuffer>
83 void freeDeviceBuffer(DeviceBuffer *buffer)
85 GMX_ASSERT(buffer, "needs a buffer pointer");
88 GMX_RELEASE_ASSERT(clReleaseMemObject(*buffer) == CL_SUCCESS, "clReleaseMemObject failed");
93 * Performs the host-to-device data copy, synchronous or asynchronously on request.
95 * TODO: This is meant to gradually replace cu/ocl_copy_h2d.
97 * \tparam ValueType Raw value type of the \p buffer.
98 * \param[in,out] buffer Pointer to the device-side buffer
99 * \param[in] hostBuffer Pointer to the raw host-side memory, also typed \p ValueType
100 * \param[in] startingOffset Offset (in values) at the device-side buffer to copy into.
101 * \param[in] numValues Number of values to copy.
102 * \param[in] stream GPU stream to perform asynchronous copy in.
103 * \param[in] transferKind Copy type: synchronous or asynchronous.
104 * \param[out] timingEvent A pointer to the H2D copy timing event to be filled in.
105 * If the pointer is not null, the event can further be used
106 * to queue a wait for this operation or to query profiling information.
108 template <typename ValueType>
109 void copyToDeviceBuffer(DeviceBuffer<ValueType> *buffer,
110 const ValueType *hostBuffer,
111 size_t startingOffset,
113 CommandStream stream,
114 GpuApiCallBehavior transferKind,
115 CommandEvent *timingEvent)
119 return; // such calls are actually made with empty domains
121 GMX_ASSERT(buffer, "needs a buffer pointer");
122 GMX_ASSERT(hostBuffer, "needs a host buffer pointer");
124 const size_t offset = startingOffset * sizeof(ValueType);
125 const size_t bytes = numValues * sizeof(ValueType);
126 switch (transferKind)
128 case GpuApiCallBehavior::Async:
129 clError = clEnqueueWriteBuffer(stream, *buffer, CL_FALSE, offset, bytes, hostBuffer, 0, nullptr, timingEvent);
130 GMX_RELEASE_ASSERT(clError == CL_SUCCESS, "Asynchronous H2D copy failed");
133 case GpuApiCallBehavior::Sync:
134 clError = clEnqueueWriteBuffer(stream, *buffer, CL_TRUE, offset, bytes, hostBuffer, 0, nullptr, timingEvent);
135 GMX_RELEASE_ASSERT(clError == CL_SUCCESS, "Synchronous H2D copy failed");
144 * Performs the device-to-host data copy, synchronous or asynchronously on request.
146 * TODO: This is meant to gradually replace cu/ocl_copy_d2h.
148 * \tparam ValueType Raw value type of the \p buffer.
149 * \param[in,out] hostBuffer Pointer to the raw host-side memory, also typed \p ValueType
150 * \param[in] buffer Pointer to the device-side buffer
151 * \param[in] startingOffset Offset (in values) at the device-side buffer to copy from.
152 * \param[in] numValues Number of values to copy.
153 * \param[in] stream GPU stream to perform asynchronous copy in.
154 * \param[in] transferKind Copy type: synchronous or asynchronous.
155 * \param[out] timingEvent A pointer to the H2D copy timing event to be filled in.
156 * If the pointer is not null, the event can further be used
157 * to queue a wait for this operation or to query profiling information.
159 template <typename ValueType>
160 void copyFromDeviceBuffer(ValueType *hostBuffer,
161 DeviceBuffer<ValueType> *buffer,
162 size_t startingOffset,
164 CommandStream stream,
165 GpuApiCallBehavior transferKind,
166 CommandEvent *timingEvent)
168 GMX_ASSERT(buffer, "needs a buffer pointer");
169 GMX_ASSERT(hostBuffer, "needs a host buffer pointer");
171 const size_t offset = startingOffset * sizeof(ValueType);
172 const size_t bytes = numValues * sizeof(ValueType);
173 switch (transferKind)
175 case GpuApiCallBehavior::Async:
176 clError = clEnqueueReadBuffer(stream, *buffer, CL_FALSE, offset, bytes, hostBuffer, 0, nullptr, timingEvent);
177 GMX_RELEASE_ASSERT(clError == CL_SUCCESS, "Asynchronous D2H copy failed");
180 case GpuApiCallBehavior::Sync:
181 clError = clEnqueueReadBuffer(stream, *buffer, CL_TRUE, offset, bytes, hostBuffer, 0, nullptr, timingEvent);
182 GMX_RELEASE_ASSERT(clError == CL_SUCCESS, "Synchronous D2H copy failed");
191 * Clears the device buffer asynchronously.
193 * \tparam ValueType Raw value type of the \p buffer.
194 * \param[in,out] buffer Pointer to the device-side buffer
195 * \param[in] startingOffset Offset (in values) at the device-side buffer to start clearing at.
196 * \param[in] numValues Number of values to clear.
197 * \param[in] stream GPU stream.
199 template <typename ValueType>
200 void clearDeviceBufferAsync(DeviceBuffer<ValueType> *buffer,
201 size_t startingOffset,
203 CommandStream stream)
205 GMX_ASSERT(buffer, "needs a buffer pointer");
206 const size_t offset = startingOffset * sizeof(ValueType);
207 const size_t bytes = numValues * sizeof(ValueType);
208 const ValueType pattern = 0;
209 const cl_uint numWaitEvents = 0;
210 const cl_event *waitEvents = nullptr;
211 cl_event commandEvent;
212 cl_int clError = clEnqueueFillBuffer(stream, *buffer, &pattern, sizeof(pattern),
214 numWaitEvents, waitEvents, &commandEvent);
215 GMX_RELEASE_ASSERT(clError == CL_SUCCESS, "Couldn't clear the device buffer");