rocprofiler-sdk/ompt/api_args.h Source File

ROCprofiler-SDK developer API 1.0.0 ROCm Profiling API and tools

• rocprofiler-sdk
• ompt

api_args.h

1// MIT License

2//

3// Copyright (c) 2023-2025 Advanced Micro Devices, Inc. All rights reserved.

4//

5// Permission is hereby granted, free of charge, to any person obtaining a copy

6// of this software and associated documentation files (the "Software"), to deal

7// in the Software without restriction, including without limitation the rights

8// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell

9// copies of the Software, and to permit persons to whom the Software is

10// furnished to do so, subject to the following conditions:

11//

12// The above copyright notice and this permission notice shall be included in

13// all copies or substantial portions of the Software.

14//

15// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR

16// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,

17// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE

18// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER

19// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,

20// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN

21// THE SOFTWARE.

22

23#pragma once

24

25#include <rocprofiler-sdk/defines.h>

26

27#include <rocprofiler-sdk/ompt/omp-tools.h>

28

29#include <stdint.h>

30

31ROCPROFILER_EXTERN_C_INIT

32

33// all the available callback interface runtime entry points

34typedef struct rocprofiler_ompt_callback_functions_t

35{

36 ompt_enumerate_states_t ompt_enumerate_states;

37 ompt_enumerate_mutex_impls_t ompt_enumerate_mutex_impls;

38 ompt_get_thread_data_t ompt_get_thread_data;

39 ompt_get_num_places_t ompt_get_num_places;

40 ompt_get_place_proc_ids_t ompt_get_place_proc_ids;

41 ompt_get_place_num_t ompt_get_place_num;

42 ompt_get_partition_place_nums_t ompt_get_partition_place_nums;

43 ompt_get_proc_id_t ompt_get_proc_id;

44 ompt_get_state_t ompt_get_state;

45 ompt_get_parallel_info_t ompt_get_parallel_info;

46 ompt_get_task_info_t ompt_get_task_info;

47 ompt_get_task_memory_t ompt_get_task_memory;

48 ompt_get_num_devices_t ompt_get_num_devices;

49 ompt_get_num_procs_t ompt_get_num_procs;

50 ompt_get_target_info_t ompt_get_target_info;

51 ompt_get_unique_id_t ompt_get_unique_id;

52} rocprofiler_ompt_callback_functions_t;

53

54// Empty struct has a size of 0 in C but size of 1 in C++.

55// This struct is added to the union members which represent

56// functions with no arguments to ensure ABI compatibility

57typedef struct rocprofiler_ompt_no_args

58{

59 char empty;

60} rocprofiler_ompt_no_args;

61

62typedef union rocprofiler_ompt_args_t

63{

64 // The ompt_data_t* values passed to the client tool are proxies.

65 // This allows the client tool to use them as it would in their own

66 // OMPT tool.

67 // We keepa a map from the address of the ompt_data_t passed to the SDK's

68 // callback to the proxy object and keep it in sync when a callback is done

69 // to the client tool.

70 struct

71 {

72 ompt_thread_t thread_type;

73 ompt_data_t* thread_data;

74 } thread_begin;

75

76 struct

77 {

78 ompt_data_t* thread_data;

79 } thread_end;

80

81 struct

82 {

83 ompt_data_t* encountering_task_data;

84 const ompt_frame_t* encountering_task_frame;

85 ompt_data_t* parallel_data;

86 unsigned int requested_parallelism;

87 int flags;

88 const void* codeptr_ra;

89 } parallel_begin;

90

91 struct

92 {

93 ompt_data_t* parallel_data;

94 ompt_data_t* encountering_task_data;

95 int flags;

96 const void* codeptr_ra;

97 } parallel_end;

98

99 struct

100 {

101 ompt_data_t* encountering_task_data;

102 const ompt_frame_t* encountering_task_frame;

103 ompt_data_t* new_task_data;

104 int flags;

105 int has_dependences;

106 const void* codeptr_ra;

107 } task_create;

108

109 struct

110 {

111 ompt_data_t* prior_task_data;

112 ompt_task_status_t prior_task_status;

113 ompt_data_t* next_task_data;

114 } task_schedule;

115

116 struct

117 {

118 ompt_scope_endpoint_t endpoint;

119 ompt_data_t* parallel_data;

120 ompt_data_t* task_data;

121 unsigned int actual_parallelism;

122 unsigned int index;

123 int flags;

124 } implicit_task;

125

126 struct

127 {

128 int device_num;

129 const char* type;

130 ompt_device_t* device;

131 ompt_function_lookup_t lookup;

132 const char* documentation;

133 } device_initialize;

134

135 struct

136 {

137 int device_num;

138 } device_finalize;

139

140 struct

141 {

142 int device_num;

143 const char* filename;

144 int64_t offset_in_file;

145 void* vma_in_file;

146 size_t bytes;

147 void* host_addr;

148 void* device_addr;

149 uint64_t module_id;

150 } device_load;

151

152 // struct

153 // {

154 // int device_num;

155 // uint64_t module_id;

156 // } device_unload;

157

158 struct

159 {

160 ompt_sync_region_t kind;

161 ompt_scope_endpoint_t endpoint;

162 ompt_data_t* parallel_data;

163 ompt_data_t* task_data;

164 const void* codeptr_ra;

165 } sync_region_wait;

166

167 struct

168 {

169 ompt_mutex_t kind;

170 ompt_wait_id_t wait_id;

171 const void* codeptr_ra;

172 } mutex_released;

173

174 struct

175 {

176 ompt_data_t* task_data;

177 const ompt_dependence_t* deps;

178 int ndeps;

179 } dependences;

180

181 struct

182 {

183 ompt_data_t* src_task_data;

184 ompt_data_t* sink_task_data;

185 } task_dependence;

186

187 struct

188 {

189 ompt_work_t work_type;

190 ompt_scope_endpoint_t endpoint;

191 ompt_data_t* parallel_data;

192 ompt_data_t* task_data;

193 uint64_t count;

194 const void* codeptr_ra;

195 } work;

196

197 struct

198 {

199 ompt_scope_endpoint_t endpoint;

200 ompt_data_t* parallel_data;

201 ompt_data_t* task_data;

202 const void* codeptr_ra;

203 } masked;

204

205 struct

206 {

207 ompt_sync_region_t kind;

208 ompt_scope_endpoint_t endpoint;

209 ompt_data_t* parallel_data;

210 ompt_data_t* task_data;

211 const void* codeptr_ra;

212 } sync_region;

213

214 struct

215 {

216 ompt_mutex_t kind;

217 unsigned int hint;

218 unsigned int impl;

219 ompt_wait_id_t wait_id;

220 const void* codeptr_ra;

221 } lock_init;

222

223 struct

224 {

225 ompt_mutex_t kind;

226 ompt_wait_id_t wait_id;

227 const void* codeptr_ra;

228 } lock_destroy;

229

230 struct

231 {

232 ompt_mutex_t kind;

233 unsigned int hint;

234 unsigned int impl;

235 ompt_wait_id_t wait_id;

236 const void* codeptr_ra;

237 } mutex_acquire;

238

239 struct

240 {

241 ompt_mutex_t kind;

242 ompt_wait_id_t wait_id;

243 const void* codeptr_ra;

244 } mutex_acquired;

245

246 struct

247 {

248 ompt_scope_endpoint_t endpoint;

249 ompt_wait_id_t wait_id;

250 const void* codeptr_ra;

251 } nest_lock;

252

253 struct

254 {

255 ompt_data_t* thread_data;

256 const void* codeptr_ra;

257 } flush;

258

259 struct

260 {

261 ompt_data_t* task_data;

262 int flags;

263 const void* codeptr_ra;

264 } cancel;

265

266 struct

267 {

268 ompt_sync_region_t kind;

269 ompt_scope_endpoint_t endpoint;

270 ompt_data_t* parallel_data;

271 ompt_data_t* task_data;

272 const void* codeptr_ra;

273 } reduction;

274

275 struct

276 {

277 ompt_data_t* parallel_data;

278 ompt_data_t* task_data;

279 ompt_dispatch_t kind;

280 ompt_data_t instance;

281 } dispatch;

282

283 struct

284 {

285 ompt_target_t kind;

286 ompt_scope_endpoint_t endpoint;

287 int device_num;

288 ompt_data_t* task_data;

289 ompt_data_t* target_task_data;

290 ompt_data_t* target_data;

291 const void* codeptr_ra;

292 } target_emi;

293

294 struct

295 {

296 ompt_scope_endpoint_t endpoint;

297 ompt_data_t* target_task_data;

298 ompt_data_t* target_data;

299 ompt_data_t* host_op_id;

300 ompt_target_data_op_t optype;

301 void* src_address;

302 int src_device_num;

303 void* dst_address;

304 int dst_device_num;

305 size_t bytes;

306 const void* codeptr_ra;

307 } target_data_op_emi;

308

309 struct

310 {

311 ompt_scope_endpoint_t endpoint;

312 ompt_data_t* target_data;

313 ompt_data_t* host_op_id;

314 unsigned int requested_num_teams;

315 } target_submit_emi;

316

317 // struct

318 // {

319 // unsigned int nitems;

320 // void** host_addr;

321 // void** device_addr;

322 // size_t* bytes;

323 // unsigned int* mapping_flags;

324 // const void* codeptr_ra;

325 // } target_map_emi;

326

327 struct

328 {

329 ompt_severity_t severity;

330 const char* message;

331 size_t length;

332 const void* codeptr_ra;

333 } error;

334

335 rocprofiler_ompt_callback_functions_t callback_functions;

336

337} rocprofiler_ompt_args_t;

338

339ROCPROFILER_EXTERN_C_FINI

---

Generated by  1.9.8