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 PluginManager *PM; 35 36 #if OMPTARGET_PROFILE_ENABLED 37 static char *ProfileTraceFile = nullptr; 38 #endif 39 40 __attribute__((constructor(101))) void init() { 41 DP("Init target library!\n"); 42 PM = new PluginManager(); 43 44 #ifdef OMPTARGET_PROFILE_ENABLED 45 ProfileTraceFile = getenv("LIBOMPTARGET_PROFILE"); 46 // TODO: add a configuration option for time granularity 47 if (ProfileTraceFile) 48 llvm::timeTraceProfilerInitialize(500 /* us */, "libomptarget"); 49 #endif 50 } 51 52 __attribute__((destructor(101))) void deinit() { 53 DP("Deinit target library!\n"); 54 delete PM; 55 56 #ifdef OMPTARGET_PROFILE_ENABLED 57 if (ProfileTraceFile) { 58 // TODO: add env var for file output 59 if (auto E = llvm::timeTraceProfilerWrite(ProfileTraceFile, "-")) 60 fprintf(stderr, "Error writing out the time trace\n"); 61 62 llvm::timeTraceProfilerCleanup(); 63 } 64 #endif 65 } 66 67 void RTLsTy::LoadRTLs() { 68 // Parse environment variable OMP_TARGET_OFFLOAD (if set) 69 PM->TargetOffloadPolicy = 70 (kmp_target_offload_kind_t)__kmpc_get_target_offload(); 71 if (PM->TargetOffloadPolicy == tgt_disabled) { 72 return; 73 } 74 75 DP("Loading RTLs...\n"); 76 77 // Attempt to open all the plugins and, if they exist, check if the interface 78 // is correct and if they are supporting any devices. 79 for (auto *Name : RTLNames) { 80 DP("Loading library '%s'...\n", Name); 81 void *dynlib_handle = dlopen(Name, RTLD_NOW); 82 83 if (!dynlib_handle) { 84 // Library does not exist or cannot be found. 85 DP("Unable to load library '%s': %s!\n", Name, dlerror()); 86 continue; 87 } 88 89 DP("Successfully loaded library '%s'!\n", Name); 90 91 // Retrieve the RTL information from the runtime library. 92 RTLInfoTy R; 93 94 R.LibraryHandler = dynlib_handle; 95 R.isUsed = false; 96 97 #ifdef OMPTARGET_DEBUG 98 R.RTLName = Name; 99 #endif 100 101 if (!(*((void **)&R.is_valid_binary) = 102 dlsym(dynlib_handle, "__tgt_rtl_is_valid_binary"))) 103 continue; 104 if (!(*((void **)&R.number_of_devices) = 105 dlsym(dynlib_handle, "__tgt_rtl_number_of_devices"))) 106 continue; 107 if (!(*((void **)&R.init_device) = 108 dlsym(dynlib_handle, "__tgt_rtl_init_device"))) 109 continue; 110 if (!(*((void **)&R.load_binary) = 111 dlsym(dynlib_handle, "__tgt_rtl_load_binary"))) 112 continue; 113 if (!(*((void **)&R.data_alloc) = 114 dlsym(dynlib_handle, "__tgt_rtl_data_alloc"))) 115 continue; 116 if (!(*((void **)&R.data_submit) = 117 dlsym(dynlib_handle, "__tgt_rtl_data_submit"))) 118 continue; 119 if (!(*((void **)&R.data_retrieve) = 120 dlsym(dynlib_handle, "__tgt_rtl_data_retrieve"))) 121 continue; 122 if (!(*((void **)&R.data_delete) = 123 dlsym(dynlib_handle, "__tgt_rtl_data_delete"))) 124 continue; 125 if (!(*((void **)&R.run_region) = 126 dlsym(dynlib_handle, "__tgt_rtl_run_target_region"))) 127 continue; 128 if (!(*((void **)&R.run_team_region) = 129 dlsym(dynlib_handle, "__tgt_rtl_run_target_team_region"))) 130 continue; 131 132 // Optional functions 133 *((void **)&R.init_requires) = 134 dlsym(dynlib_handle, "__tgt_rtl_init_requires"); 135 *((void **)&R.data_submit_async) = 136 dlsym(dynlib_handle, "__tgt_rtl_data_submit_async"); 137 *((void **)&R.data_retrieve_async) = 138 dlsym(dynlib_handle, "__tgt_rtl_data_retrieve_async"); 139 *((void **)&R.run_region_async) = 140 dlsym(dynlib_handle, "__tgt_rtl_run_target_region_async"); 141 *((void **)&R.run_team_region_async) = 142 dlsym(dynlib_handle, "__tgt_rtl_run_target_team_region_async"); 143 *((void **)&R.synchronize) = dlsym(dynlib_handle, "__tgt_rtl_synchronize"); 144 *((void **)&R.data_exchange) = 145 dlsym(dynlib_handle, "__tgt_rtl_data_exchange"); 146 *((void **)&R.data_exchange_async) = 147 dlsym(dynlib_handle, "__tgt_rtl_data_exchange_async"); 148 *((void **)&R.is_data_exchangable) = 149 dlsym(dynlib_handle, "__tgt_rtl_is_data_exchangable"); 150 151 // No devices are supported by this RTL? 152 if (!(R.NumberOfDevices = R.number_of_devices())) { 153 DP("No devices supported in this RTL\n"); 154 continue; 155 } 156 157 DP("Registering RTL %s supporting %d devices!\n", R.RTLName.c_str(), 158 R.NumberOfDevices); 159 160 // The RTL is valid! Will save the information in the RTLs list. 161 AllRTLs.push_back(R); 162 } 163 164 DP("RTLs loaded!\n"); 165 166 return; 167 } 168 169 //////////////////////////////////////////////////////////////////////////////// 170 // Functionality for registering libs 171 172 static void RegisterImageIntoTranslationTable(TranslationTable &TT, 173 RTLInfoTy &RTL, __tgt_device_image *image) { 174 175 // same size, as when we increase one, we also increase the other. 176 assert(TT.TargetsTable.size() == TT.TargetsImages.size() && 177 "We should have as many images as we have tables!"); 178 179 // Resize the Targets Table and Images to accommodate the new targets if 180 // required 181 unsigned TargetsTableMinimumSize = RTL.Idx + RTL.NumberOfDevices; 182 183 if (TT.TargetsTable.size() < TargetsTableMinimumSize) { 184 TT.TargetsImages.resize(TargetsTableMinimumSize, 0); 185 TT.TargetsTable.resize(TargetsTableMinimumSize, 0); 186 } 187 188 // Register the image in all devices for this target type. 189 for (int32_t i = 0; i < RTL.NumberOfDevices; ++i) { 190 // If we are changing the image we are also invalidating the target table. 191 if (TT.TargetsImages[RTL.Idx + i] != image) { 192 TT.TargetsImages[RTL.Idx + i] = image; 193 TT.TargetsTable[RTL.Idx + i] = 0; // lazy initialization of target table. 194 } 195 } 196 } 197 198 //////////////////////////////////////////////////////////////////////////////// 199 // Functionality for registering Ctors/Dtors 200 201 static void RegisterGlobalCtorsDtorsForImage(__tgt_bin_desc *desc, 202 __tgt_device_image *img, RTLInfoTy *RTL) { 203 204 for (int32_t i = 0; i < RTL->NumberOfDevices; ++i) { 205 DeviceTy &Device = PM->Devices[RTL->Idx + i]; 206 Device.PendingGlobalsMtx.lock(); 207 Device.HasPendingGlobals = true; 208 for (__tgt_offload_entry *entry = img->EntriesBegin; 209 entry != img->EntriesEnd; ++entry) { 210 if (entry->flags & OMP_DECLARE_TARGET_CTOR) { 211 DP("Adding ctor " DPxMOD " to the pending list.\n", 212 DPxPTR(entry->addr)); 213 Device.PendingCtorsDtors[desc].PendingCtors.push_back(entry->addr); 214 } else if (entry->flags & OMP_DECLARE_TARGET_DTOR) { 215 // Dtors are pushed in reverse order so they are executed from end 216 // to beginning when unregistering the library! 217 DP("Adding dtor " DPxMOD " to the pending list.\n", 218 DPxPTR(entry->addr)); 219 Device.PendingCtorsDtors[desc].PendingDtors.push_front(entry->addr); 220 } 221 222 if (entry->flags & OMP_DECLARE_TARGET_LINK) { 223 DP("The \"link\" attribute is not yet supported!\n"); 224 } 225 } 226 Device.PendingGlobalsMtx.unlock(); 227 } 228 } 229 230 void RTLsTy::RegisterRequires(int64_t flags) { 231 // TODO: add more elaborate check. 232 // Minimal check: only set requires flags if previous value 233 // is undefined. This ensures that only the first call to this 234 // function will set the requires flags. All subsequent calls 235 // will be checked for compatibility. 236 assert(flags != OMP_REQ_UNDEFINED && 237 "illegal undefined flag for requires directive!"); 238 if (RequiresFlags == OMP_REQ_UNDEFINED) { 239 RequiresFlags = flags; 240 return; 241 } 242 243 // If multiple compilation units are present enforce 244 // consistency across all of them for require clauses: 245 // - reverse_offload 246 // - unified_address 247 // - unified_shared_memory 248 if ((RequiresFlags & OMP_REQ_REVERSE_OFFLOAD) != 249 (flags & OMP_REQ_REVERSE_OFFLOAD)) { 250 FATAL_MESSAGE0(1, 251 "'#pragma omp requires reverse_offload' not used consistently!"); 252 } 253 if ((RequiresFlags & OMP_REQ_UNIFIED_ADDRESS) != 254 (flags & OMP_REQ_UNIFIED_ADDRESS)) { 255 FATAL_MESSAGE0(1, 256 "'#pragma omp requires unified_address' not used consistently!"); 257 } 258 if ((RequiresFlags & OMP_REQ_UNIFIED_SHARED_MEMORY) != 259 (flags & OMP_REQ_UNIFIED_SHARED_MEMORY)) { 260 FATAL_MESSAGE0(1, 261 "'#pragma omp requires unified_shared_memory' not used consistently!"); 262 } 263 264 // TODO: insert any other missing checks 265 266 DP("New requires flags %" PRId64 " compatible with existing %" PRId64 "!\n", 267 flags, RequiresFlags); 268 } 269 270 void RTLsTy::RegisterLib(__tgt_bin_desc *desc) { 271 // Attempt to load all plugins available in the system. 272 std::call_once(initFlag, &RTLsTy::LoadRTLs, this); 273 274 PM->RTLsMtx.lock(); 275 // Register the images with the RTLs that understand them, if any. 276 for (int32_t i = 0; i < desc->NumDeviceImages; ++i) { 277 // Obtain the image. 278 __tgt_device_image *img = &desc->DeviceImages[i]; 279 280 RTLInfoTy *FoundRTL = NULL; 281 282 // Scan the RTLs that have associated images until we find one that supports 283 // the current image. 284 for (auto &R : AllRTLs) { 285 if (!R.is_valid_binary(img)) { 286 DP("Image " DPxMOD " is NOT compatible with RTL %s!\n", 287 DPxPTR(img->ImageStart), R.RTLName.c_str()); 288 continue; 289 } 290 291 DP("Image " DPxMOD " is compatible with RTL %s!\n", 292 DPxPTR(img->ImageStart), R.RTLName.c_str()); 293 294 // If this RTL is not already in use, initialize it. 295 if (!R.isUsed) { 296 // Initialize the device information for the RTL we are about to use. 297 DeviceTy device(&R); 298 size_t Start = PM->Devices.size(); 299 PM->Devices.resize(Start + R.NumberOfDevices, device); 300 for (int32_t device_id = 0; device_id < R.NumberOfDevices; 301 device_id++) { 302 // global device ID 303 PM->Devices[Start + device_id].DeviceID = Start + device_id; 304 // RTL local device ID 305 PM->Devices[Start + device_id].RTLDeviceID = device_id; 306 } 307 308 // Initialize the index of this RTL and save it in the used RTLs. 309 R.Idx = (UsedRTLs.empty()) 310 ? 0 311 : UsedRTLs.back()->Idx + UsedRTLs.back()->NumberOfDevices; 312 assert((size_t) R.Idx == Start && 313 "RTL index should equal the number of devices used so far."); 314 R.isUsed = true; 315 UsedRTLs.push_back(&R); 316 317 DP("RTL " DPxMOD " has index %d!\n", DPxPTR(R.LibraryHandler), R.Idx); 318 } 319 320 // Initialize (if necessary) translation table for this library. 321 PM->TrlTblMtx.lock(); 322 if (!PM->HostEntriesBeginToTransTable.count(desc->HostEntriesBegin)) { 323 TranslationTable &TransTable = 324 (PM->HostEntriesBeginToTransTable)[desc->HostEntriesBegin]; 325 TransTable.HostTable.EntriesBegin = desc->HostEntriesBegin; 326 TransTable.HostTable.EntriesEnd = desc->HostEntriesEnd; 327 } 328 329 // Retrieve translation table for this library. 330 TranslationTable &TransTable = 331 (PM->HostEntriesBeginToTransTable)[desc->HostEntriesBegin]; 332 333 DP("Registering image " DPxMOD " with RTL %s!\n", 334 DPxPTR(img->ImageStart), R.RTLName.c_str()); 335 RegisterImageIntoTranslationTable(TransTable, R, img); 336 PM->TrlTblMtx.unlock(); 337 FoundRTL = &R; 338 339 // Load ctors/dtors for static objects 340 RegisterGlobalCtorsDtorsForImage(desc, img, FoundRTL); 341 342 // if an RTL was found we are done - proceed to register the next image 343 break; 344 } 345 346 if (!FoundRTL) { 347 DP("No RTL found for image " DPxMOD "!\n", DPxPTR(img->ImageStart)); 348 } 349 } 350 PM->RTLsMtx.unlock(); 351 352 DP("Done registering entries!\n"); 353 } 354 355 void RTLsTy::UnregisterLib(__tgt_bin_desc *desc) { 356 DP("Unloading target library!\n"); 357 358 PM->RTLsMtx.lock(); 359 // Find which RTL understands each image, if any. 360 for (int32_t i = 0; i < desc->NumDeviceImages; ++i) { 361 // Obtain the image. 362 __tgt_device_image *img = &desc->DeviceImages[i]; 363 364 RTLInfoTy *FoundRTL = NULL; 365 366 // Scan the RTLs that have associated images until we find one that supports 367 // the current image. We only need to scan RTLs that are already being used. 368 for (auto *R : UsedRTLs) { 369 370 assert(R->isUsed && "Expecting used RTLs."); 371 372 if (!R->is_valid_binary(img)) { 373 DP("Image " DPxMOD " is NOT compatible with RTL " DPxMOD "!\n", 374 DPxPTR(img->ImageStart), DPxPTR(R->LibraryHandler)); 375 continue; 376 } 377 378 DP("Image " DPxMOD " is compatible with RTL " DPxMOD "!\n", 379 DPxPTR(img->ImageStart), DPxPTR(R->LibraryHandler)); 380 381 FoundRTL = R; 382 383 // Execute dtors for static objects if the device has been used, i.e. 384 // if its PendingCtors list has been emptied. 385 for (int32_t i = 0; i < FoundRTL->NumberOfDevices; ++i) { 386 DeviceTy &Device = PM->Devices[FoundRTL->Idx + i]; 387 Device.PendingGlobalsMtx.lock(); 388 if (Device.PendingCtorsDtors[desc].PendingCtors.empty()) { 389 for (auto &dtor : Device.PendingCtorsDtors[desc].PendingDtors) { 390 int rc = target(Device.DeviceID, dtor, 0, nullptr, nullptr, nullptr, 391 nullptr, nullptr, nullptr, 1, 1, true /*team*/); 392 if (rc != OFFLOAD_SUCCESS) { 393 DP("Running destructor " DPxMOD " failed.\n", DPxPTR(dtor)); 394 } 395 } 396 // Remove this library's entry from PendingCtorsDtors 397 Device.PendingCtorsDtors.erase(desc); 398 } 399 Device.PendingGlobalsMtx.unlock(); 400 } 401 402 DP("Unregistered image " DPxMOD " from RTL " DPxMOD "!\n", 403 DPxPTR(img->ImageStart), DPxPTR(R->LibraryHandler)); 404 405 break; 406 } 407 408 // if no RTL was found proceed to unregister the next image 409 if (!FoundRTL){ 410 DP("No RTLs in use support the image " DPxMOD "!\n", 411 DPxPTR(img->ImageStart)); 412 } 413 } 414 PM->RTLsMtx.unlock(); 415 DP("Done unregistering images!\n"); 416 417 // Remove entries from PM->HostPtrToTableMap 418 PM->TblMapMtx.lock(); 419 for (__tgt_offload_entry *cur = desc->HostEntriesBegin; 420 cur < desc->HostEntriesEnd; ++cur) { 421 PM->HostPtrToTableMap.erase(cur->addr); 422 } 423 424 // Remove translation table for this descriptor. 425 auto TransTable = PM->HostEntriesBeginToTransTable.find(desc->HostEntriesBegin); 426 if (TransTable != PM->HostEntriesBeginToTransTable.end()) { 427 DP("Removing translation table for descriptor " DPxMOD "\n", 428 DPxPTR(desc->HostEntriesBegin)); 429 PM->HostEntriesBeginToTransTable.erase(TransTable); 430 } else { 431 DP("Translation table for descriptor " DPxMOD " cannot be found, probably " 432 "it has been already removed.\n", DPxPTR(desc->HostEntriesBegin)); 433 } 434 435 PM->TblMapMtx.unlock(); 436 437 // TODO: Remove RTL and the devices it manages if it's not used anymore? 438 // TODO: Write some RTL->unload_image(...) function? 439 440 DP("Done unregistering library!\n"); 441 } 442