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