1 //===----------- rtl.cpp - Target independent OpenMP target RTL -----------===// 2 // 3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 // See https://llvm.org/LICENSE.txt for license information. 5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 // 7 //===----------------------------------------------------------------------===// 8 // 9 // Functionality for handling RTL plugins. 10 // 11 //===----------------------------------------------------------------------===// 12 13 #include "device.h" 14 #include "private.h" 15 #include "rtl.h" 16 17 #include <cassert> 18 #include <cstdlib> 19 #include <cstring> 20 #include <dlfcn.h> 21 #include <mutex> 22 #include <string> 23 24 // List of all plugins that can support offloading. 25 static const char *RTLNames[] = { 26 /* PowerPC target */ "libomptarget.rtl.ppc64.so", 27 /* x86_64 target */ "libomptarget.rtl.x86_64.so", 28 /* CUDA target */ "libomptarget.rtl.cuda.so", 29 /* AArch64 target */ "libomptarget.rtl.aarch64.so", 30 /* SX-Aurora VE target */ "libomptarget.rtl.ve.so", 31 /* AMDGPU target */ "libomptarget.rtl.amdgpu.so", 32 }; 33 34 RTLsTy *RTLs; 35 std::mutex *RTLsMtx; 36 37 HostEntriesBeginToTransTableTy *HostEntriesBeginToTransTable; 38 std::mutex *TrlTblMtx; 39 40 HostPtrToTableMapTy *HostPtrToTableMap; 41 std::mutex *TblMapMtx; 42 43 __attribute__((constructor(101))) void init() { 44 DP("Init target library!\n"); 45 RTLs = new RTLsTy(); 46 RTLsMtx = new std::mutex(); 47 HostEntriesBeginToTransTable = new HostEntriesBeginToTransTableTy(); 48 TrlTblMtx = new std::mutex(); 49 HostPtrToTableMap = new HostPtrToTableMapTy(); 50 TblMapMtx = new std::mutex(); 51 } 52 53 __attribute__((destructor(101))) void deinit() { 54 DP("Deinit target library!\n"); 55 delete RTLs; 56 delete RTLsMtx; 57 delete HostEntriesBeginToTransTable; 58 delete TrlTblMtx; 59 delete HostPtrToTableMap; 60 delete TblMapMtx; 61 } 62 63 void RTLsTy::LoadRTLs() { 64 65 if (char *envStr = getenv("LIBOMPTARGET_INFO")) { 66 InfoLevel = std::stoi(envStr); 67 } 68 #ifdef OMPTARGET_DEBUG 69 if (char *envStr = getenv("LIBOMPTARGET_DEBUG")) { 70 DebugLevel = std::stoi(envStr); 71 } 72 #endif // OMPTARGET_DEBUG 73 74 // Parse environment variable OMP_TARGET_OFFLOAD (if set) 75 TargetOffloadPolicy = (kmp_target_offload_kind_t) __kmpc_get_target_offload(); 76 if (TargetOffloadPolicy == tgt_disabled) { 77 return; 78 } 79 80 DP("Loading RTLs...\n"); 81 82 // Attempt to open all the plugins and, if they exist, check if the interface 83 // is correct and if they are supporting any devices. 84 for (auto *Name : RTLNames) { 85 DP("Loading library '%s'...\n", Name); 86 void *dynlib_handle = dlopen(Name, RTLD_NOW); 87 88 if (!dynlib_handle) { 89 // Library does not exist or cannot be found. 90 DP("Unable to load library '%s': %s!\n", Name, dlerror()); 91 continue; 92 } 93 94 DP("Successfully loaded library '%s'!\n", Name); 95 96 // Retrieve the RTL information from the runtime library. 97 RTLInfoTy R; 98 99 R.LibraryHandler = dynlib_handle; 100 R.isUsed = false; 101 102 #ifdef OMPTARGET_DEBUG 103 R.RTLName = Name; 104 #endif 105 106 if (!(*((void **)&R.is_valid_binary) = 107 dlsym(dynlib_handle, "__tgt_rtl_is_valid_binary"))) 108 continue; 109 if (!(*((void **)&R.number_of_devices) = 110 dlsym(dynlib_handle, "__tgt_rtl_number_of_devices"))) 111 continue; 112 if (!(*((void **)&R.init_device) = 113 dlsym(dynlib_handle, "__tgt_rtl_init_device"))) 114 continue; 115 if (!(*((void **)&R.load_binary) = 116 dlsym(dynlib_handle, "__tgt_rtl_load_binary"))) 117 continue; 118 if (!(*((void **)&R.data_alloc) = 119 dlsym(dynlib_handle, "__tgt_rtl_data_alloc"))) 120 continue; 121 if (!(*((void **)&R.data_submit) = 122 dlsym(dynlib_handle, "__tgt_rtl_data_submit"))) 123 continue; 124 if (!(*((void **)&R.data_retrieve) = 125 dlsym(dynlib_handle, "__tgt_rtl_data_retrieve"))) 126 continue; 127 if (!(*((void **)&R.data_delete) = 128 dlsym(dynlib_handle, "__tgt_rtl_data_delete"))) 129 continue; 130 if (!(*((void **)&R.run_region) = 131 dlsym(dynlib_handle, "__tgt_rtl_run_target_region"))) 132 continue; 133 if (!(*((void **)&R.run_team_region) = 134 dlsym(dynlib_handle, "__tgt_rtl_run_target_team_region"))) 135 continue; 136 137 // Optional functions 138 *((void **)&R.init_requires) = 139 dlsym(dynlib_handle, "__tgt_rtl_init_requires"); 140 *((void **)&R.data_submit_async) = 141 dlsym(dynlib_handle, "__tgt_rtl_data_submit_async"); 142 *((void **)&R.data_retrieve_async) = 143 dlsym(dynlib_handle, "__tgt_rtl_data_retrieve_async"); 144 *((void **)&R.run_region_async) = 145 dlsym(dynlib_handle, "__tgt_rtl_run_target_region_async"); 146 *((void **)&R.run_team_region_async) = 147 dlsym(dynlib_handle, "__tgt_rtl_run_target_team_region_async"); 148 *((void **)&R.synchronize) = dlsym(dynlib_handle, "__tgt_rtl_synchronize"); 149 *((void **)&R.data_exchange) = 150 dlsym(dynlib_handle, "__tgt_rtl_data_exchange"); 151 *((void **)&R.data_exchange_async) = 152 dlsym(dynlib_handle, "__tgt_rtl_data_exchange_async"); 153 *((void **)&R.is_data_exchangable) = 154 dlsym(dynlib_handle, "__tgt_rtl_is_data_exchangable"); 155 156 // No devices are supported by this RTL? 157 if (!(R.NumberOfDevices = R.number_of_devices())) { 158 DP("No devices supported in this RTL\n"); 159 continue; 160 } 161 162 DP("Registering RTL %s supporting %d devices!\n", R.RTLName.c_str(), 163 R.NumberOfDevices); 164 165 // The RTL is valid! Will save the information in the RTLs list. 166 AllRTLs.push_back(R); 167 } 168 169 DP("RTLs loaded!\n"); 170 171 return; 172 } 173 174 //////////////////////////////////////////////////////////////////////////////// 175 // Functionality for registering libs 176 177 static void RegisterImageIntoTranslationTable(TranslationTable &TT, 178 RTLInfoTy &RTL, __tgt_device_image *image) { 179 180 // same size, as when we increase one, we also increase the other. 181 assert(TT.TargetsTable.size() == TT.TargetsImages.size() && 182 "We should have as many images as we have tables!"); 183 184 // Resize the Targets Table and Images to accommodate the new targets if 185 // required 186 unsigned TargetsTableMinimumSize = RTL.Idx + RTL.NumberOfDevices; 187 188 if (TT.TargetsTable.size() < TargetsTableMinimumSize) { 189 TT.TargetsImages.resize(TargetsTableMinimumSize, 0); 190 TT.TargetsTable.resize(TargetsTableMinimumSize, 0); 191 } 192 193 // Register the image in all devices for this target type. 194 for (int32_t i = 0; i < RTL.NumberOfDevices; ++i) { 195 // If we are changing the image we are also invalidating the target table. 196 if (TT.TargetsImages[RTL.Idx + i] != image) { 197 TT.TargetsImages[RTL.Idx + i] = image; 198 TT.TargetsTable[RTL.Idx + i] = 0; // lazy initialization of target table. 199 } 200 } 201 } 202 203 //////////////////////////////////////////////////////////////////////////////// 204 // Functionality for registering Ctors/Dtors 205 206 static void RegisterGlobalCtorsDtorsForImage(__tgt_bin_desc *desc, 207 __tgt_device_image *img, RTLInfoTy *RTL) { 208 209 for (int32_t i = 0; i < RTL->NumberOfDevices; ++i) { 210 DeviceTy &Device = Devices[RTL->Idx + i]; 211 Device.PendingGlobalsMtx.lock(); 212 Device.HasPendingGlobals = true; 213 for (__tgt_offload_entry *entry = img->EntriesBegin; 214 entry != img->EntriesEnd; ++entry) { 215 if (entry->flags & OMP_DECLARE_TARGET_CTOR) { 216 DP("Adding ctor " DPxMOD " to the pending list.\n", 217 DPxPTR(entry->addr)); 218 Device.PendingCtorsDtors[desc].PendingCtors.push_back(entry->addr); 219 } else if (entry->flags & OMP_DECLARE_TARGET_DTOR) { 220 // Dtors are pushed in reverse order so they are executed from end 221 // to beginning when unregistering the library! 222 DP("Adding dtor " DPxMOD " to the pending list.\n", 223 DPxPTR(entry->addr)); 224 Device.PendingCtorsDtors[desc].PendingDtors.push_front(entry->addr); 225 } 226 227 if (entry->flags & OMP_DECLARE_TARGET_LINK) { 228 DP("The \"link\" attribute is not yet supported!\n"); 229 } 230 } 231 Device.PendingGlobalsMtx.unlock(); 232 } 233 } 234 235 void RTLsTy::RegisterRequires(int64_t flags) { 236 // TODO: add more elaborate check. 237 // Minimal check: only set requires flags if previous value 238 // is undefined. This ensures that only the first call to this 239 // function will set the requires flags. All subsequent calls 240 // will be checked for compatibility. 241 assert(flags != OMP_REQ_UNDEFINED && 242 "illegal undefined flag for requires directive!"); 243 if (RequiresFlags == OMP_REQ_UNDEFINED) { 244 RequiresFlags = flags; 245 return; 246 } 247 248 // If multiple compilation units are present enforce 249 // consistency across all of them for require clauses: 250 // - reverse_offload 251 // - unified_address 252 // - unified_shared_memory 253 if ((RequiresFlags & OMP_REQ_REVERSE_OFFLOAD) != 254 (flags & OMP_REQ_REVERSE_OFFLOAD)) { 255 FATAL_MESSAGE0(1, 256 "'#pragma omp requires reverse_offload' not used consistently!"); 257 } 258 if ((RequiresFlags & OMP_REQ_UNIFIED_ADDRESS) != 259 (flags & OMP_REQ_UNIFIED_ADDRESS)) { 260 FATAL_MESSAGE0(1, 261 "'#pragma omp requires unified_address' not used consistently!"); 262 } 263 if ((RequiresFlags & OMP_REQ_UNIFIED_SHARED_MEMORY) != 264 (flags & OMP_REQ_UNIFIED_SHARED_MEMORY)) { 265 FATAL_MESSAGE0(1, 266 "'#pragma omp requires unified_shared_memory' not used consistently!"); 267 } 268 269 // TODO: insert any other missing checks 270 271 DP("New requires flags %" PRId64 " compatible with existing %" PRId64 "!\n", 272 flags, RequiresFlags); 273 } 274 275 void RTLsTy::RegisterLib(__tgt_bin_desc *desc) { 276 // Attempt to load all plugins available in the system. 277 std::call_once(initFlag, &RTLsTy::LoadRTLs, this); 278 279 RTLsMtx->lock(); 280 // Register the images with the RTLs that understand them, if any. 281 for (int32_t i = 0; i < desc->NumDeviceImages; ++i) { 282 // Obtain the image. 283 __tgt_device_image *img = &desc->DeviceImages[i]; 284 285 RTLInfoTy *FoundRTL = NULL; 286 287 // Scan the RTLs that have associated images until we find one that supports 288 // the current image. 289 for (auto &R : AllRTLs) { 290 if (!R.is_valid_binary(img)) { 291 DP("Image " DPxMOD " is NOT compatible with RTL %s!\n", 292 DPxPTR(img->ImageStart), R.RTLName.c_str()); 293 continue; 294 } 295 296 DP("Image " DPxMOD " is compatible with RTL %s!\n", 297 DPxPTR(img->ImageStart), R.RTLName.c_str()); 298 299 // If this RTL is not already in use, initialize it. 300 if (!R.isUsed) { 301 // Initialize the device information for the RTL we are about to use. 302 DeviceTy device(&R); 303 size_t start = Devices.size(); 304 Devices.resize(start + R.NumberOfDevices, device); 305 for (int32_t device_id = 0; device_id < R.NumberOfDevices; 306 device_id++) { 307 // global device ID 308 Devices[start + device_id].DeviceID = start + device_id; 309 // RTL local device ID 310 Devices[start + device_id].RTLDeviceID = device_id; 311 } 312 313 // Initialize the index of this RTL and save it in the used RTLs. 314 R.Idx = (UsedRTLs.empty()) 315 ? 0 316 : UsedRTLs.back()->Idx + UsedRTLs.back()->NumberOfDevices; 317 assert((size_t) R.Idx == start && 318 "RTL index should equal the number of devices used so far."); 319 R.isUsed = true; 320 UsedRTLs.push_back(&R); 321 322 DP("RTL " DPxMOD " has index %d!\n", DPxPTR(R.LibraryHandler), R.Idx); 323 } 324 325 // Initialize (if necessary) translation table for this library. 326 TrlTblMtx->lock(); 327 if(!HostEntriesBeginToTransTable->count(desc->HostEntriesBegin)){ 328 TranslationTable &tt = 329 (*HostEntriesBeginToTransTable)[desc->HostEntriesBegin]; 330 tt.HostTable.EntriesBegin = desc->HostEntriesBegin; 331 tt.HostTable.EntriesEnd = desc->HostEntriesEnd; 332 } 333 334 // Retrieve translation table for this library. 335 TranslationTable &TransTable = 336 (*HostEntriesBeginToTransTable)[desc->HostEntriesBegin]; 337 338 DP("Registering image " DPxMOD " with RTL %s!\n", 339 DPxPTR(img->ImageStart), R.RTLName.c_str()); 340 RegisterImageIntoTranslationTable(TransTable, R, img); 341 TrlTblMtx->unlock(); 342 FoundRTL = &R; 343 344 // Load ctors/dtors for static objects 345 RegisterGlobalCtorsDtorsForImage(desc, img, FoundRTL); 346 347 // if an RTL was found we are done - proceed to register the next image 348 break; 349 } 350 351 if (!FoundRTL) { 352 DP("No RTL found for image " DPxMOD "!\n", DPxPTR(img->ImageStart)); 353 } 354 } 355 RTLsMtx->unlock(); 356 357 358 DP("Done registering entries!\n"); 359 } 360 361 void RTLsTy::UnregisterLib(__tgt_bin_desc *desc) { 362 DP("Unloading target library!\n"); 363 364 RTLsMtx->lock(); 365 // Find which RTL understands each image, if any. 366 for (int32_t i = 0; i < desc->NumDeviceImages; ++i) { 367 // Obtain the image. 368 __tgt_device_image *img = &desc->DeviceImages[i]; 369 370 RTLInfoTy *FoundRTL = NULL; 371 372 // Scan the RTLs that have associated images until we find one that supports 373 // the current image. We only need to scan RTLs that are already being used. 374 for (auto *R : UsedRTLs) { 375 376 assert(R->isUsed && "Expecting used RTLs."); 377 378 if (!R->is_valid_binary(img)) { 379 DP("Image " DPxMOD " is NOT compatible with RTL " DPxMOD "!\n", 380 DPxPTR(img->ImageStart), DPxPTR(R->LibraryHandler)); 381 continue; 382 } 383 384 DP("Image " DPxMOD " is compatible with RTL " DPxMOD "!\n", 385 DPxPTR(img->ImageStart), DPxPTR(R->LibraryHandler)); 386 387 FoundRTL = R; 388 389 // Execute dtors for static objects if the device has been used, i.e. 390 // if its PendingCtors list has been emptied. 391 for (int32_t i = 0; i < FoundRTL->NumberOfDevices; ++i) { 392 DeviceTy &Device = Devices[FoundRTL->Idx + i]; 393 Device.PendingGlobalsMtx.lock(); 394 if (Device.PendingCtorsDtors[desc].PendingCtors.empty()) { 395 for (auto &dtor : Device.PendingCtorsDtors[desc].PendingDtors) { 396 int rc = target(Device.DeviceID, dtor, 0, NULL, NULL, NULL, NULL, 397 NULL, 1, 1, true /*team*/); 398 if (rc != OFFLOAD_SUCCESS) { 399 DP("Running destructor " DPxMOD " failed.\n", DPxPTR(dtor)); 400 } 401 } 402 // Remove this library's entry from PendingCtorsDtors 403 Device.PendingCtorsDtors.erase(desc); 404 } 405 Device.PendingGlobalsMtx.unlock(); 406 } 407 408 DP("Unregistered image " DPxMOD " from RTL " DPxMOD "!\n", 409 DPxPTR(img->ImageStart), DPxPTR(R->LibraryHandler)); 410 411 break; 412 } 413 414 // if no RTL was found proceed to unregister the next image 415 if (!FoundRTL){ 416 DP("No RTLs in use support the image " DPxMOD "!\n", 417 DPxPTR(img->ImageStart)); 418 } 419 } 420 RTLsMtx->unlock(); 421 DP("Done unregistering images!\n"); 422 423 // Remove entries from HostPtrToTableMap 424 TblMapMtx->lock(); 425 for (__tgt_offload_entry *cur = desc->HostEntriesBegin; 426 cur < desc->HostEntriesEnd; ++cur) { 427 HostPtrToTableMap->erase(cur->addr); 428 } 429 430 // Remove translation table for this descriptor. 431 auto tt = HostEntriesBeginToTransTable->find(desc->HostEntriesBegin); 432 if (tt != HostEntriesBeginToTransTable->end()) { 433 DP("Removing translation table for descriptor " DPxMOD "\n", 434 DPxPTR(desc->HostEntriesBegin)); 435 HostEntriesBeginToTransTable->erase(tt); 436 } else { 437 DP("Translation table for descriptor " DPxMOD " cannot be found, probably " 438 "it has been already removed.\n", DPxPTR(desc->HostEntriesBegin)); 439 } 440 441 TblMapMtx->unlock(); 442 443 // TODO: Remove RTL and the devices it manages if it's not used anymore? 444 // TODO: Write some RTL->unload_image(...) function? 445 446 DP("Done unregistering library!\n"); 447 } 448