2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2014,2015,2016,2017,2018,2019, 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.
36 * \brief Define utility routines for OpenCL
38 * \author Anca Hamuraru <anca@streamcomputing.eu>
51 #include "gromacs/gpu_utils/gpu_utils.h"
52 #include "gromacs/utility/fatalerror.h"
53 #include "gromacs/utility/smalloc.h"
55 int ocl_copy_H2D(cl_mem d_dest,
59 GpuApiCallBehavior transferKind,
60 cl_command_queue command_queue,
63 cl_int gmx_unused cl_error;
65 if (d_dest == nullptr || h_src == nullptr || bytes == 0)
72 case GpuApiCallBehavior::Async:
73 cl_error = clEnqueueWriteBuffer(command_queue, d_dest, CL_FALSE, offset, bytes, h_src,
74 0, nullptr, copy_event);
75 assert(cl_error == CL_SUCCESS);
76 // TODO: handle errors
79 case GpuApiCallBehavior::Sync:
80 cl_error = clEnqueueWriteBuffer(command_queue, d_dest, CL_TRUE, offset, bytes, h_src, 0,
82 assert(cl_error == CL_SUCCESS);
83 // TODO: handle errors
92 /*! \brief Launches asynchronous host to device memory copy.
94 * If copy_event is not nullptr, on return it will contain an event object
95 * identifying this particular host to device operation. The event can further
96 * be used to queue a wait for this operation or to query profiling information.
98 int ocl_copy_H2D_async(cl_mem d_dest,
102 cl_command_queue command_queue,
103 cl_event* copy_event)
105 return ocl_copy_H2D(d_dest, h_src, offset, bytes, GpuApiCallBehavior::Async, command_queue, copy_event);
108 /*! \brief Launches synchronous host to device memory copy.
110 int ocl_copy_H2D_sync(cl_mem d_dest, const void* h_src, size_t offset, size_t bytes, cl_command_queue command_queue)
112 return ocl_copy_H2D(d_dest, h_src, offset, bytes, GpuApiCallBehavior::Sync, command_queue, nullptr);
115 int ocl_copy_D2H(void* h_dest,
119 GpuApiCallBehavior transferKind,
120 cl_command_queue command_queue,
121 cl_event* copy_event)
123 cl_int gmx_unused cl_error;
125 if (h_dest == nullptr || d_src == nullptr || bytes == 0)
130 switch (transferKind)
132 case GpuApiCallBehavior::Async:
133 cl_error = clEnqueueReadBuffer(command_queue, d_src, CL_FALSE, offset, bytes, h_dest, 0,
134 nullptr, copy_event);
135 assert(cl_error == CL_SUCCESS);
136 // TODO: handle errors
139 case GpuApiCallBehavior::Sync:
140 cl_error = clEnqueueReadBuffer(command_queue, d_src, CL_TRUE, offset, bytes, h_dest, 0,
141 nullptr, copy_event);
142 assert(cl_error == CL_SUCCESS);
143 // TODO: handle errors
152 /*! \brief Launches asynchronous device to host memory copy.
154 * If copy_event is not nullptr, on return it will contain an event object
155 * identifying this particular host to device operation. The event can further
156 * be used to queue a wait for this operation or to query profiling information.
158 int ocl_copy_D2H_async(void* h_dest,
162 cl_command_queue command_queue,
163 cl_event* copy_event)
165 return ocl_copy_D2H(h_dest, d_src, offset, bytes, GpuApiCallBehavior::Async, command_queue, copy_event);
168 /*! \brief \brief Allocates nbytes of host memory. Use ocl_free to free memory allocated with this function.
171 * This function should allocate page-locked memory to help reduce D2H and H2D
172 * transfer times, similar with pmalloc from pmalloc_cuda.cu.
174 * \param[in,out] h_ptr Pointer where to store the address of the newly allocated buffer.
175 * \param[in] nbytes Size in bytes of the buffer to be allocated.
177 void pmalloc(void** h_ptr, size_t nbytes)
179 /* Need a temporary type whose size is 1 byte, so that the
180 * implementation of snew_aligned can cope without issuing
182 char** temporary = reinterpret_cast<char**>(h_ptr);
184 /* 16-byte alignment is required by the neighbour-searching code,
185 * because it uses four-wide SIMD for bounding-box calculation.
186 * However, when we organize using page-locked memory for
187 * device-host transfers, it will probably need to be aligned to a
188 * 4kb page, like CUDA does. */
189 snew_aligned(*temporary, nbytes, 16);
192 /*! \brief Frees memory allocated with pmalloc.
194 * \param[in] h_ptr Buffer allocated with pmalloc that needs to be freed.
196 void pfree(void* h_ptr)
201 sfree_aligned(h_ptr);
205 /*! \brief Convert error code to diagnostic string */
206 std::string ocl_get_error_string(cl_int error)
210 // run-time and JIT compiler errors
211 case 0: return "CL_SUCCESS";
212 case -1: return "CL_DEVICE_NOT_FOUND";
213 case -2: return "CL_DEVICE_NOT_AVAILABLE";
214 case -3: return "CL_COMPILER_NOT_AVAILABLE";
215 case -4: return "CL_MEM_OBJECT_ALLOCATION_FAILURE";
216 case -5: return "CL_OUT_OF_RESOURCES";
217 case -6: return "CL_OUT_OF_HOST_MEMORY";
218 case -7: return "CL_PROFILING_INFO_NOT_AVAILABLE";
219 case -8: return "CL_MEM_COPY_OVERLAP";
220 case -9: return "CL_IMAGE_FORMAT_MISMATCH";
221 case -10: return "CL_IMAGE_FORMAT_NOT_SUPPORTED";
222 case -11: return "CL_BUILD_PROGRAM_FAILURE";
223 case -12: return "CL_MAP_FAILURE";
224 case -13: return "CL_MISALIGNED_SUB_BUFFER_OFFSET";
225 case -14: return "CL_EXEC_STATUS_ERROR_FOR_EVENTS_IN_WAIT_LIST";
226 case -15: return "CL_COMPILE_PROGRAM_FAILURE";
227 case -16: return "CL_LINKER_NOT_AVAILABLE";
228 case -17: return "CL_LINK_PROGRAM_FAILURE";
229 case -18: return "CL_DEVICE_PARTITION_FAILED";
230 case -19: return "CL_KERNEL_ARG_INFO_NOT_AVAILABLE";
232 // compile-time errors
233 case -30: return "CL_INVALID_VALUE";
234 case -31: return "CL_INVALID_DEVICE_TYPE";
235 case -32: return "CL_INVALID_PLATFORM";
236 case -33: return "CL_INVALID_DEVICE";
237 case -34: return "CL_INVALID_CONTEXT";
238 case -35: return "CL_INVALID_QUEUE_PROPERTIES";
239 case -36: return "CL_INVALID_COMMAND_QUEUE";
240 case -37: return "CL_INVALID_HOST_PTR";
241 case -38: return "CL_INVALID_MEM_OBJECT";
242 case -39: return "CL_INVALID_IMAGE_FORMAT_DESCRIPTOR";
243 case -40: return "CL_INVALID_IMAGE_SIZE";
244 case -41: return "CL_INVALID_SAMPLER";
245 case -42: return "CL_INVALID_BINARY";
246 case -43: return "CL_INVALID_BUILD_OPTIONS";
247 case -44: return "CL_INVALID_PROGRAM";
248 case -45: return "CL_INVALID_PROGRAM_EXECUTABLE";
249 case -46: return "CL_INVALID_KERNEL_NAME";
250 case -47: return "CL_INVALID_KERNEL_DEFINITION";
251 case -48: return "CL_INVALID_KERNEL";
252 case -49: return "CL_INVALID_ARG_INDEX";
253 case -50: return "CL_INVALID_ARG_VALUE";
254 case -51: return "CL_INVALID_ARG_SIZE";
255 case -52: return "CL_INVALID_KERNEL_ARGS";
256 case -53: return "CL_INVALID_WORK_DIMENSION";
257 case -54: return "CL_INVALID_WORK_GROUP_SIZE";
258 case -55: return "CL_INVALID_WORK_ITEM_SIZE";
259 case -56: return "CL_INVALID_GLOBAL_OFFSET";
260 case -57: return "CL_INVALID_EVENT_WAIT_LIST";
261 case -58: return "CL_INVALID_EVENT";
262 case -59: return "CL_INVALID_OPERATION";
263 case -60: return "CL_INVALID_GL_OBJECT";
264 case -61: return "CL_INVALID_BUFFER_SIZE";
265 case -62: return "CL_INVALID_MIP_LEVEL";
266 case -63: return "CL_INVALID_GLOBAL_WORK_SIZE";
267 case -64: return "CL_INVALID_PROPERTY";
268 case -65: return "CL_INVALID_IMAGE_DESCRIPTOR";
269 case -66: return "CL_INVALID_COMPILER_OPTIONS";
270 case -67: return "CL_INVALID_LINKER_OPTIONS";
271 case -68: return "CL_INVALID_DEVICE_PARTITION_COUNT";
274 case -1000: return "CL_INVALID_GL_SHAREGROUP_REFERENCE_KHR";
275 case -1001: return "CL_PLATFORM_NOT_FOUND_KHR";
276 case -1002: return "CL_INVALID_D3D10_DEVICE_KHR";
277 case -1003: return "CL_INVALID_D3D10_RESOURCE_KHR";
278 case -1004: return "CL_D3D10_RESOURCE_ALREADY_ACQUIRED_KHR";
279 case -1005: return "CL_D3D10_RESOURCE_NOT_ACQUIRED_KHR";
280 default: return "Unknown OpenCL error: " + std::to_string(static_cast<int32_t>(error));