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