1 //===----- CGOpenMPRuntime.h - Interface to OpenMP Runtimes -----*- C++ -*-===// 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 // This provides a class for OpenMP runtime code generation. 10 // 11 //===----------------------------------------------------------------------===// 12 13 #ifndef LLVM_CLANG_LIB_CODEGEN_CGOPENMPRUNTIME_H 14 #define LLVM_CLANG_LIB_CODEGEN_CGOPENMPRUNTIME_H 15 16 #include "CGValue.h" 17 #include "clang/AST/DeclOpenMP.h" 18 #include "clang/AST/Type.h" 19 #include "clang/Basic/OpenMPKinds.h" 20 #include "clang/Basic/SourceLocation.h" 21 #include "llvm/ADT/DenseMap.h" 22 #include "llvm/ADT/StringMap.h" 23 #include "llvm/ADT/StringSet.h" 24 #include "llvm/IR/Function.h" 25 #include "llvm/IR/ValueHandle.h" 26 27 namespace llvm { 28 class ArrayType; 29 class Constant; 30 class FunctionType; 31 class GlobalVariable; 32 class StructType; 33 class Type; 34 class Value; 35 } // namespace llvm 36 37 namespace clang { 38 class Expr; 39 class GlobalDecl; 40 class OMPDependClause; 41 class OMPExecutableDirective; 42 class OMPLoopDirective; 43 class VarDecl; 44 class OMPDeclareReductionDecl; 45 class IdentifierInfo; 46 47 namespace CodeGen { 48 class Address; 49 class CodeGenFunction; 50 class CodeGenModule; 51 52 /// A basic class for pre|post-action for advanced codegen sequence for OpenMP 53 /// region. 54 class PrePostActionTy { 55 public: 56 explicit PrePostActionTy() {} 57 virtual void Enter(CodeGenFunction &CGF) {} 58 virtual void Exit(CodeGenFunction &CGF) {} 59 virtual ~PrePostActionTy() {} 60 }; 61 62 /// Class provides a way to call simple version of codegen for OpenMP region, or 63 /// an advanced with possible pre|post-actions in codegen. 64 class RegionCodeGenTy final { 65 intptr_t CodeGen; 66 typedef void (*CodeGenTy)(intptr_t, CodeGenFunction &, PrePostActionTy &); 67 CodeGenTy Callback; 68 mutable PrePostActionTy *PrePostAction; 69 RegionCodeGenTy() = delete; 70 RegionCodeGenTy &operator=(const RegionCodeGenTy &) = delete; 71 template <typename Callable> 72 static void CallbackFn(intptr_t CodeGen, CodeGenFunction &CGF, 73 PrePostActionTy &Action) { 74 return (*reinterpret_cast<Callable *>(CodeGen))(CGF, Action); 75 } 76 77 public: 78 template <typename Callable> 79 RegionCodeGenTy( 80 Callable &&CodeGen, 81 typename std::enable_if< 82 !std::is_same<typename std::remove_reference<Callable>::type, 83 RegionCodeGenTy>::value>::type * = nullptr) 84 : CodeGen(reinterpret_cast<intptr_t>(&CodeGen)), 85 Callback(CallbackFn<typename std::remove_reference<Callable>::type>), 86 PrePostAction(nullptr) {} 87 void setAction(PrePostActionTy &Action) const { PrePostAction = &Action; } 88 void operator()(CodeGenFunction &CGF) const; 89 }; 90 91 struct OMPTaskDataTy final { 92 SmallVector<const Expr *, 4> PrivateVars; 93 SmallVector<const Expr *, 4> PrivateCopies; 94 SmallVector<const Expr *, 4> FirstprivateVars; 95 SmallVector<const Expr *, 4> FirstprivateCopies; 96 SmallVector<const Expr *, 4> FirstprivateInits; 97 SmallVector<const Expr *, 4> LastprivateVars; 98 SmallVector<const Expr *, 4> LastprivateCopies; 99 SmallVector<const Expr *, 4> ReductionVars; 100 SmallVector<const Expr *, 4> ReductionCopies; 101 SmallVector<const Expr *, 4> ReductionOps; 102 SmallVector<std::pair<OpenMPDependClauseKind, const Expr *>, 4> Dependences; 103 llvm::PointerIntPair<llvm::Value *, 1, bool> Final; 104 llvm::PointerIntPair<llvm::Value *, 1, bool> Schedule; 105 llvm::PointerIntPair<llvm::Value *, 1, bool> Priority; 106 llvm::Value *Reductions = nullptr; 107 unsigned NumberOfParts = 0; 108 bool Tied = true; 109 bool Nogroup = false; 110 }; 111 112 /// Class intended to support codegen of all kind of the reduction clauses. 113 class ReductionCodeGen { 114 private: 115 /// Data required for codegen of reduction clauses. 116 struct ReductionData { 117 /// Reference to the original shared item. 118 const Expr *Ref = nullptr; 119 /// Helper expression for generation of private copy. 120 const Expr *Private = nullptr; 121 /// Helper expression for generation reduction operation. 122 const Expr *ReductionOp = nullptr; 123 ReductionData(const Expr *Ref, const Expr *Private, const Expr *ReductionOp) 124 : Ref(Ref), Private(Private), ReductionOp(ReductionOp) {} 125 }; 126 /// List of reduction-based clauses. 127 SmallVector<ReductionData, 4> ClausesData; 128 129 /// List of addresses of original shared variables/expressions. 130 SmallVector<std::pair<LValue, LValue>, 4> SharedAddresses; 131 /// Sizes of the reduction items in chars. 132 SmallVector<std::pair<llvm::Value *, llvm::Value *>, 4> Sizes; 133 /// Base declarations for the reduction items. 134 SmallVector<const VarDecl *, 4> BaseDecls; 135 136 /// Emits lvalue for shared expression. 137 LValue emitSharedLValue(CodeGenFunction &CGF, const Expr *E); 138 /// Emits upper bound for shared expression (if array section). 139 LValue emitSharedLValueUB(CodeGenFunction &CGF, const Expr *E); 140 /// Performs aggregate initialization. 141 /// \param N Number of reduction item in the common list. 142 /// \param PrivateAddr Address of the corresponding private item. 143 /// \param SharedLVal Address of the original shared variable. 144 /// \param DRD Declare reduction construct used for reduction item. 145 void emitAggregateInitialization(CodeGenFunction &CGF, unsigned N, 146 Address PrivateAddr, LValue SharedLVal, 147 const OMPDeclareReductionDecl *DRD); 148 149 public: 150 ReductionCodeGen(ArrayRef<const Expr *> Shareds, 151 ArrayRef<const Expr *> Privates, 152 ArrayRef<const Expr *> ReductionOps); 153 /// Emits lvalue for a reduction item. 154 /// \param N Number of the reduction item. 155 void emitSharedLValue(CodeGenFunction &CGF, unsigned N); 156 /// Emits the code for the variable-modified type, if required. 157 /// \param N Number of the reduction item. 158 void emitAggregateType(CodeGenFunction &CGF, unsigned N); 159 /// Emits the code for the variable-modified type, if required. 160 /// \param N Number of the reduction item. 161 /// \param Size Size of the type in chars. 162 void emitAggregateType(CodeGenFunction &CGF, unsigned N, llvm::Value *Size); 163 /// Performs initialization of the private copy for the reduction item. 164 /// \param N Number of the reduction item. 165 /// \param PrivateAddr Address of the corresponding private item. 166 /// \param DefaultInit Default initialization sequence that should be 167 /// performed if no reduction specific initialization is found. 168 /// \param SharedLVal Address of the original shared variable. 169 void 170 emitInitialization(CodeGenFunction &CGF, unsigned N, Address PrivateAddr, 171 LValue SharedLVal, 172 llvm::function_ref<bool(CodeGenFunction &)> DefaultInit); 173 /// Returns true if the private copy requires cleanups. 174 bool needCleanups(unsigned N); 175 /// Emits cleanup code for the reduction item. 176 /// \param N Number of the reduction item. 177 /// \param PrivateAddr Address of the corresponding private item. 178 void emitCleanups(CodeGenFunction &CGF, unsigned N, Address PrivateAddr); 179 /// Adjusts \p PrivatedAddr for using instead of the original variable 180 /// address in normal operations. 181 /// \param N Number of the reduction item. 182 /// \param PrivateAddr Address of the corresponding private item. 183 Address adjustPrivateAddress(CodeGenFunction &CGF, unsigned N, 184 Address PrivateAddr); 185 /// Returns LValue for the reduction item. 186 LValue getSharedLValue(unsigned N) const { return SharedAddresses[N].first; } 187 /// Returns the size of the reduction item (in chars and total number of 188 /// elements in the item), or nullptr, if the size is a constant. 189 std::pair<llvm::Value *, llvm::Value *> getSizes(unsigned N) const { 190 return Sizes[N]; 191 } 192 /// Returns the base declaration of the reduction item. 193 const VarDecl *getBaseDecl(unsigned N) const { return BaseDecls[N]; } 194 /// Returns the base declaration of the reduction item. 195 const Expr *getRefExpr(unsigned N) const { return ClausesData[N].Ref; } 196 /// Returns true if the initialization of the reduction item uses initializer 197 /// from declare reduction construct. 198 bool usesReductionInitializer(unsigned N) const; 199 }; 200 201 class CGOpenMPRuntime { 202 public: 203 /// Allows to disable automatic handling of functions used in target regions 204 /// as those marked as `omp declare target`. 205 class DisableAutoDeclareTargetRAII { 206 CodeGenModule &CGM; 207 bool SavedShouldMarkAsGlobal; 208 209 public: 210 DisableAutoDeclareTargetRAII(CodeGenModule &CGM); 211 ~DisableAutoDeclareTargetRAII(); 212 }; 213 214 protected: 215 CodeGenModule &CGM; 216 StringRef FirstSeparator, Separator; 217 218 /// Constructor allowing to redefine the name separator for the variables. 219 explicit CGOpenMPRuntime(CodeGenModule &CGM, StringRef FirstSeparator, 220 StringRef Separator); 221 222 /// Creates offloading entry for the provided entry ID \a ID, 223 /// address \a Addr, size \a Size, and flags \a Flags. 224 virtual void createOffloadEntry(llvm::Constant *ID, llvm::Constant *Addr, 225 uint64_t Size, int32_t Flags, 226 llvm::GlobalValue::LinkageTypes Linkage); 227 228 /// Helper to emit outlined function for 'target' directive. 229 /// \param D Directive to emit. 230 /// \param ParentName Name of the function that encloses the target region. 231 /// \param OutlinedFn Outlined function value to be defined by this call. 232 /// \param OutlinedFnID Outlined function ID value to be defined by this call. 233 /// \param IsOffloadEntry True if the outlined function is an offload entry. 234 /// \param CodeGen Lambda codegen specific to an accelerator device. 235 /// An outlined function may not be an entry if, e.g. the if clause always 236 /// evaluates to false. 237 virtual void emitTargetOutlinedFunctionHelper(const OMPExecutableDirective &D, 238 StringRef ParentName, 239 llvm::Function *&OutlinedFn, 240 llvm::Constant *&OutlinedFnID, 241 bool IsOffloadEntry, 242 const RegionCodeGenTy &CodeGen); 243 244 /// Emits code for OpenMP 'if' clause using specified \a CodeGen 245 /// function. Here is the logic: 246 /// if (Cond) { 247 /// ThenGen(); 248 /// } else { 249 /// ElseGen(); 250 /// } 251 void emitOMPIfClause(CodeGenFunction &CGF, const Expr *Cond, 252 const RegionCodeGenTy &ThenGen, 253 const RegionCodeGenTy &ElseGen); 254 255 /// Emits object of ident_t type with info for source location. 256 /// \param Flags Flags for OpenMP location. 257 /// 258 llvm::Value *emitUpdateLocation(CodeGenFunction &CGF, SourceLocation Loc, 259 unsigned Flags = 0); 260 261 /// Returns pointer to ident_t type. 262 llvm::Type *getIdentTyPointerTy(); 263 264 /// Gets thread id value for the current thread. 265 /// 266 llvm::Value *getThreadID(CodeGenFunction &CGF, SourceLocation Loc); 267 268 /// Get the function name of an outlined region. 269 // The name can be customized depending on the target. 270 // 271 virtual StringRef getOutlinedHelperName() const { return ".omp_outlined."; } 272 273 /// Emits \p Callee function call with arguments \p Args with location \p Loc. 274 void emitCall(CodeGenFunction &CGF, SourceLocation Loc, 275 llvm::FunctionCallee Callee, 276 ArrayRef<llvm::Value *> Args = llvm::None) const; 277 278 /// Emits address of the word in a memory where current thread id is 279 /// stored. 280 virtual Address emitThreadIDAddress(CodeGenFunction &CGF, SourceLocation Loc); 281 282 void setLocThreadIdInsertPt(CodeGenFunction &CGF, 283 bool AtCurrentPoint = false); 284 void clearLocThreadIdInsertPt(CodeGenFunction &CGF); 285 286 /// Check if the default location must be constant. 287 /// Default is false to support OMPT/OMPD. 288 virtual bool isDefaultLocationConstant() const { return false; } 289 290 /// Returns additional flags that can be stored in reserved_2 field of the 291 /// default location. 292 virtual unsigned getDefaultLocationReserved2Flags() const { return 0; } 293 294 /// Returns default flags for the barriers depending on the directive, for 295 /// which this barier is going to be emitted. 296 static unsigned getDefaultFlagsForBarriers(OpenMPDirectiveKind Kind); 297 298 /// Get the LLVM type for the critical name. 299 llvm::ArrayType *getKmpCriticalNameTy() const {return KmpCriticalNameTy;} 300 301 /// Returns corresponding lock object for the specified critical region 302 /// name. If the lock object does not exist it is created, otherwise the 303 /// reference to the existing copy is returned. 304 /// \param CriticalName Name of the critical region. 305 /// 306 llvm::Value *getCriticalRegionLock(StringRef CriticalName); 307 308 private: 309 /// Default const ident_t object used for initialization of all other 310 /// ident_t objects. 311 llvm::Constant *DefaultOpenMPPSource = nullptr; 312 using FlagsTy = std::pair<unsigned, unsigned>; 313 /// Map of flags and corresponding default locations. 314 using OpenMPDefaultLocMapTy = llvm::DenseMap<FlagsTy, llvm::Value *>; 315 OpenMPDefaultLocMapTy OpenMPDefaultLocMap; 316 Address getOrCreateDefaultLocation(unsigned Flags); 317 318 QualType IdentQTy; 319 llvm::StructType *IdentTy = nullptr; 320 /// Map for SourceLocation and OpenMP runtime library debug locations. 321 typedef llvm::DenseMap<unsigned, llvm::Value *> OpenMPDebugLocMapTy; 322 OpenMPDebugLocMapTy OpenMPDebugLocMap; 323 /// The type for a microtask which gets passed to __kmpc_fork_call(). 324 /// Original representation is: 325 /// typedef void (kmpc_micro)(kmp_int32 global_tid, kmp_int32 bound_tid,...); 326 llvm::FunctionType *Kmpc_MicroTy = nullptr; 327 /// Stores debug location and ThreadID for the function. 328 struct DebugLocThreadIdTy { 329 llvm::Value *DebugLoc; 330 llvm::Value *ThreadID; 331 /// Insert point for the service instructions. 332 llvm::AssertingVH<llvm::Instruction> ServiceInsertPt = nullptr; 333 }; 334 /// Map of local debug location, ThreadId and functions. 335 typedef llvm::DenseMap<llvm::Function *, DebugLocThreadIdTy> 336 OpenMPLocThreadIDMapTy; 337 OpenMPLocThreadIDMapTy OpenMPLocThreadIDMap; 338 /// Map of UDRs and corresponding combiner/initializer. 339 typedef llvm::DenseMap<const OMPDeclareReductionDecl *, 340 std::pair<llvm::Function *, llvm::Function *>> 341 UDRMapTy; 342 UDRMapTy UDRMap; 343 /// Map of functions and locally defined UDRs. 344 typedef llvm::DenseMap<llvm::Function *, 345 SmallVector<const OMPDeclareReductionDecl *, 4>> 346 FunctionUDRMapTy; 347 FunctionUDRMapTy FunctionUDRMap; 348 /// Type kmp_critical_name, originally defined as typedef kmp_int32 349 /// kmp_critical_name[8]; 350 llvm::ArrayType *KmpCriticalNameTy; 351 /// An ordered map of auto-generated variables to their unique names. 352 /// It stores variables with the following names: 1) ".gomp_critical_user_" + 353 /// <critical_section_name> + ".var" for "omp critical" directives; 2) 354 /// <mangled_name_for_global_var> + ".cache." for cache for threadprivate 355 /// variables. 356 llvm::StringMap<llvm::AssertingVH<llvm::Constant>, llvm::BumpPtrAllocator> 357 InternalVars; 358 /// Type typedef kmp_int32 (* kmp_routine_entry_t)(kmp_int32, void *); 359 llvm::Type *KmpRoutineEntryPtrTy = nullptr; 360 QualType KmpRoutineEntryPtrQTy; 361 /// Type typedef struct kmp_task { 362 /// void * shareds; /**< pointer to block of pointers to 363 /// shared vars */ 364 /// kmp_routine_entry_t routine; /**< pointer to routine to call for 365 /// executing task */ 366 /// kmp_int32 part_id; /**< part id for the task */ 367 /// kmp_routine_entry_t destructors; /* pointer to function to invoke 368 /// deconstructors of firstprivate C++ objects */ 369 /// } kmp_task_t; 370 QualType KmpTaskTQTy; 371 /// Saved kmp_task_t for task directive. 372 QualType SavedKmpTaskTQTy; 373 /// Saved kmp_task_t for taskloop-based directive. 374 QualType SavedKmpTaskloopTQTy; 375 /// Type typedef struct kmp_depend_info { 376 /// kmp_intptr_t base_addr; 377 /// size_t len; 378 /// struct { 379 /// bool in:1; 380 /// bool out:1; 381 /// } flags; 382 /// } kmp_depend_info_t; 383 QualType KmpDependInfoTy; 384 /// struct kmp_dim { // loop bounds info casted to kmp_int64 385 /// kmp_int64 lo; // lower 386 /// kmp_int64 up; // upper 387 /// kmp_int64 st; // stride 388 /// }; 389 QualType KmpDimTy; 390 /// Type struct __tgt_offload_entry{ 391 /// void *addr; // Pointer to the offload entry info. 392 /// // (function or global) 393 /// char *name; // Name of the function or global. 394 /// size_t size; // Size of the entry info (0 if it a function). 395 /// }; 396 QualType TgtOffloadEntryQTy; 397 /// struct __tgt_device_image{ 398 /// void *ImageStart; // Pointer to the target code start. 399 /// void *ImageEnd; // Pointer to the target code end. 400 /// // We also add the host entries to the device image, as it may be useful 401 /// // for the target runtime to have access to that information. 402 /// __tgt_offload_entry *EntriesBegin; // Begin of the table with all 403 /// // the entries. 404 /// __tgt_offload_entry *EntriesEnd; // End of the table with all the 405 /// // entries (non inclusive). 406 /// }; 407 QualType TgtDeviceImageQTy; 408 /// struct __tgt_bin_desc{ 409 /// int32_t NumDevices; // Number of devices supported. 410 /// __tgt_device_image *DeviceImages; // Arrays of device images 411 /// // (one per device). 412 /// __tgt_offload_entry *EntriesBegin; // Begin of the table with all the 413 /// // entries. 414 /// __tgt_offload_entry *EntriesEnd; // End of the table with all the 415 /// // entries (non inclusive). 416 /// }; 417 QualType TgtBinaryDescriptorQTy; 418 /// Entity that registers the offloading constants that were emitted so 419 /// far. 420 class OffloadEntriesInfoManagerTy { 421 CodeGenModule &CGM; 422 423 /// Number of entries registered so far. 424 unsigned OffloadingEntriesNum = 0; 425 426 public: 427 /// Base class of the entries info. 428 class OffloadEntryInfo { 429 public: 430 /// Kind of a given entry. 431 enum OffloadingEntryInfoKinds : unsigned { 432 /// Entry is a target region. 433 OffloadingEntryInfoTargetRegion = 0, 434 /// Entry is a declare target variable. 435 OffloadingEntryInfoDeviceGlobalVar = 1, 436 /// Invalid entry info. 437 OffloadingEntryInfoInvalid = ~0u 438 }; 439 440 protected: 441 OffloadEntryInfo() = delete; 442 explicit OffloadEntryInfo(OffloadingEntryInfoKinds Kind) : Kind(Kind) {} 443 explicit OffloadEntryInfo(OffloadingEntryInfoKinds Kind, unsigned Order, 444 uint32_t Flags) 445 : Flags(Flags), Order(Order), Kind(Kind) {} 446 ~OffloadEntryInfo() = default; 447 448 public: 449 bool isValid() const { return Order != ~0u; } 450 unsigned getOrder() const { return Order; } 451 OffloadingEntryInfoKinds getKind() const { return Kind; } 452 uint32_t getFlags() const { return Flags; } 453 void setFlags(uint32_t NewFlags) { Flags = NewFlags; } 454 llvm::Constant *getAddress() const { 455 return cast_or_null<llvm::Constant>(Addr); 456 } 457 void setAddress(llvm::Constant *V) { 458 assert(!Addr.pointsToAliveValue() && "Address has been set before!"); 459 Addr = V; 460 } 461 static bool classof(const OffloadEntryInfo *Info) { return true; } 462 463 private: 464 /// Address of the entity that has to be mapped for offloading. 465 llvm::WeakTrackingVH Addr; 466 467 /// Flags associated with the device global. 468 uint32_t Flags = 0u; 469 470 /// Order this entry was emitted. 471 unsigned Order = ~0u; 472 473 OffloadingEntryInfoKinds Kind = OffloadingEntryInfoInvalid; 474 }; 475 476 /// Return true if a there are no entries defined. 477 bool empty() const; 478 /// Return number of entries defined so far. 479 unsigned size() const { return OffloadingEntriesNum; } 480 OffloadEntriesInfoManagerTy(CodeGenModule &CGM) : CGM(CGM) {} 481 482 // 483 // Target region entries related. 484 // 485 486 /// Kind of the target registry entry. 487 enum OMPTargetRegionEntryKind : uint32_t { 488 /// Mark the entry as target region. 489 OMPTargetRegionEntryTargetRegion = 0x0, 490 /// Mark the entry as a global constructor. 491 OMPTargetRegionEntryCtor = 0x02, 492 /// Mark the entry as a global destructor. 493 OMPTargetRegionEntryDtor = 0x04, 494 }; 495 496 /// Target region entries info. 497 class OffloadEntryInfoTargetRegion final : public OffloadEntryInfo { 498 /// Address that can be used as the ID of the entry. 499 llvm::Constant *ID = nullptr; 500 501 public: 502 OffloadEntryInfoTargetRegion() 503 : OffloadEntryInfo(OffloadingEntryInfoTargetRegion) {} 504 explicit OffloadEntryInfoTargetRegion(unsigned Order, 505 llvm::Constant *Addr, 506 llvm::Constant *ID, 507 OMPTargetRegionEntryKind Flags) 508 : OffloadEntryInfo(OffloadingEntryInfoTargetRegion, Order, Flags), 509 ID(ID) { 510 setAddress(Addr); 511 } 512 513 llvm::Constant *getID() const { return ID; } 514 void setID(llvm::Constant *V) { 515 assert(!ID && "ID has been set before!"); 516 ID = V; 517 } 518 static bool classof(const OffloadEntryInfo *Info) { 519 return Info->getKind() == OffloadingEntryInfoTargetRegion; 520 } 521 }; 522 523 /// Initialize target region entry. 524 void initializeTargetRegionEntryInfo(unsigned DeviceID, unsigned FileID, 525 StringRef ParentName, unsigned LineNum, 526 unsigned Order); 527 /// Register target region entry. 528 void registerTargetRegionEntryInfo(unsigned DeviceID, unsigned FileID, 529 StringRef ParentName, unsigned LineNum, 530 llvm::Constant *Addr, llvm::Constant *ID, 531 OMPTargetRegionEntryKind Flags); 532 /// Return true if a target region entry with the provided information 533 /// exists. 534 bool hasTargetRegionEntryInfo(unsigned DeviceID, unsigned FileID, 535 StringRef ParentName, unsigned LineNum) const; 536 /// brief Applies action \a Action on all registered entries. 537 typedef llvm::function_ref<void(unsigned, unsigned, StringRef, unsigned, 538 const OffloadEntryInfoTargetRegion &)> 539 OffloadTargetRegionEntryInfoActTy; 540 void actOnTargetRegionEntriesInfo( 541 const OffloadTargetRegionEntryInfoActTy &Action); 542 543 // 544 // Device global variable entries related. 545 // 546 547 /// Kind of the global variable entry.. 548 enum OMPTargetGlobalVarEntryKind : uint32_t { 549 /// Mark the entry as a to declare target. 550 OMPTargetGlobalVarEntryTo = 0x0, 551 /// Mark the entry as a to declare target link. 552 OMPTargetGlobalVarEntryLink = 0x1, 553 }; 554 555 /// Device global variable entries info. 556 class OffloadEntryInfoDeviceGlobalVar final : public OffloadEntryInfo { 557 /// Type of the global variable. 558 CharUnits VarSize; 559 llvm::GlobalValue::LinkageTypes Linkage; 560 561 public: 562 OffloadEntryInfoDeviceGlobalVar() 563 : OffloadEntryInfo(OffloadingEntryInfoDeviceGlobalVar) {} 564 explicit OffloadEntryInfoDeviceGlobalVar(unsigned Order, 565 OMPTargetGlobalVarEntryKind Flags) 566 : OffloadEntryInfo(OffloadingEntryInfoDeviceGlobalVar, Order, Flags) {} 567 explicit OffloadEntryInfoDeviceGlobalVar( 568 unsigned Order, llvm::Constant *Addr, CharUnits VarSize, 569 OMPTargetGlobalVarEntryKind Flags, 570 llvm::GlobalValue::LinkageTypes Linkage) 571 : OffloadEntryInfo(OffloadingEntryInfoDeviceGlobalVar, Order, Flags), 572 VarSize(VarSize), Linkage(Linkage) { 573 setAddress(Addr); 574 } 575 576 CharUnits getVarSize() const { return VarSize; } 577 void setVarSize(CharUnits Size) { VarSize = Size; } 578 llvm::GlobalValue::LinkageTypes getLinkage() const { return Linkage; } 579 void setLinkage(llvm::GlobalValue::LinkageTypes LT) { Linkage = LT; } 580 static bool classof(const OffloadEntryInfo *Info) { 581 return Info->getKind() == OffloadingEntryInfoDeviceGlobalVar; 582 } 583 }; 584 585 /// Initialize device global variable entry. 586 void initializeDeviceGlobalVarEntryInfo(StringRef Name, 587 OMPTargetGlobalVarEntryKind Flags, 588 unsigned Order); 589 590 /// Register device global variable entry. 591 void 592 registerDeviceGlobalVarEntryInfo(StringRef VarName, llvm::Constant *Addr, 593 CharUnits VarSize, 594 OMPTargetGlobalVarEntryKind Flags, 595 llvm::GlobalValue::LinkageTypes Linkage); 596 /// Checks if the variable with the given name has been registered already. 597 bool hasDeviceGlobalVarEntryInfo(StringRef VarName) const { 598 return OffloadEntriesDeviceGlobalVar.count(VarName) > 0; 599 } 600 /// Applies action \a Action on all registered entries. 601 typedef llvm::function_ref<void(StringRef, 602 const OffloadEntryInfoDeviceGlobalVar &)> 603 OffloadDeviceGlobalVarEntryInfoActTy; 604 void actOnDeviceGlobalVarEntriesInfo( 605 const OffloadDeviceGlobalVarEntryInfoActTy &Action); 606 607 private: 608 // Storage for target region entries kind. The storage is to be indexed by 609 // file ID, device ID, parent function name and line number. 610 typedef llvm::DenseMap<unsigned, OffloadEntryInfoTargetRegion> 611 OffloadEntriesTargetRegionPerLine; 612 typedef llvm::StringMap<OffloadEntriesTargetRegionPerLine> 613 OffloadEntriesTargetRegionPerParentName; 614 typedef llvm::DenseMap<unsigned, OffloadEntriesTargetRegionPerParentName> 615 OffloadEntriesTargetRegionPerFile; 616 typedef llvm::DenseMap<unsigned, OffloadEntriesTargetRegionPerFile> 617 OffloadEntriesTargetRegionPerDevice; 618 typedef OffloadEntriesTargetRegionPerDevice OffloadEntriesTargetRegionTy; 619 OffloadEntriesTargetRegionTy OffloadEntriesTargetRegion; 620 /// Storage for device global variable entries kind. The storage is to be 621 /// indexed by mangled name. 622 typedef llvm::StringMap<OffloadEntryInfoDeviceGlobalVar> 623 OffloadEntriesDeviceGlobalVarTy; 624 OffloadEntriesDeviceGlobalVarTy OffloadEntriesDeviceGlobalVar; 625 }; 626 OffloadEntriesInfoManagerTy OffloadEntriesInfoManager; 627 628 bool ShouldMarkAsGlobal = true; 629 /// List of the emitted functions. 630 llvm::StringSet<> AlreadyEmittedTargetFunctions; 631 /// List of the global variables with their addresses that should not be 632 /// emitted for the target. 633 llvm::StringMap<llvm::WeakTrackingVH> EmittedNonTargetVariables; 634 635 /// List of variables that can become declare target implicitly and, thus, 636 /// must be emitted. 637 llvm::SmallDenseSet<const VarDecl *> DeferredGlobalVariables; 638 639 /// Creates and registers offloading binary descriptor for the current 640 /// compilation unit. The function that does the registration is returned. 641 llvm::Function *createOffloadingBinaryDescriptorRegistration(); 642 643 /// Creates all the offload entries in the current compilation unit 644 /// along with the associated metadata. 645 void createOffloadEntriesAndInfoMetadata(); 646 647 /// Loads all the offload entries information from the host IR 648 /// metadata. 649 void loadOffloadInfoMetadata(); 650 651 /// Returns __tgt_offload_entry type. 652 QualType getTgtOffloadEntryQTy(); 653 654 /// Returns __tgt_device_image type. 655 QualType getTgtDeviceImageQTy(); 656 657 /// Returns __tgt_bin_desc type. 658 QualType getTgtBinaryDescriptorQTy(); 659 660 /// Start scanning from statement \a S and and emit all target regions 661 /// found along the way. 662 /// \param S Starting statement. 663 /// \param ParentName Name of the function declaration that is being scanned. 664 void scanForTargetRegionsFunctions(const Stmt *S, StringRef ParentName); 665 666 /// Build type kmp_routine_entry_t (if not built yet). 667 void emitKmpRoutineEntryT(QualType KmpInt32Ty); 668 669 /// Returns pointer to kmpc_micro type. 670 llvm::Type *getKmpc_MicroPointerTy(); 671 672 /// Returns specified OpenMP runtime function. 673 /// \param Function OpenMP runtime function. 674 /// \return Specified function. 675 llvm::FunctionCallee createRuntimeFunction(unsigned Function); 676 677 /// Returns __kmpc_for_static_init_* runtime function for the specified 678 /// size \a IVSize and sign \a IVSigned. 679 llvm::FunctionCallee createForStaticInitFunction(unsigned IVSize, 680 bool IVSigned); 681 682 /// Returns __kmpc_dispatch_init_* runtime function for the specified 683 /// size \a IVSize and sign \a IVSigned. 684 llvm::FunctionCallee createDispatchInitFunction(unsigned IVSize, 685 bool IVSigned); 686 687 /// Returns __kmpc_dispatch_next_* runtime function for the specified 688 /// size \a IVSize and sign \a IVSigned. 689 llvm::FunctionCallee createDispatchNextFunction(unsigned IVSize, 690 bool IVSigned); 691 692 /// Returns __kmpc_dispatch_fini_* runtime function for the specified 693 /// size \a IVSize and sign \a IVSigned. 694 llvm::FunctionCallee createDispatchFiniFunction(unsigned IVSize, 695 bool IVSigned); 696 697 /// If the specified mangled name is not in the module, create and 698 /// return threadprivate cache object. This object is a pointer's worth of 699 /// storage that's reserved for use by the OpenMP runtime. 700 /// \param VD Threadprivate variable. 701 /// \return Cache variable for the specified threadprivate. 702 llvm::Constant *getOrCreateThreadPrivateCache(const VarDecl *VD); 703 704 /// Gets (if variable with the given name already exist) or creates 705 /// internal global variable with the specified Name. The created variable has 706 /// linkage CommonLinkage by default and is initialized by null value. 707 /// \param Ty Type of the global variable. If it is exist already the type 708 /// must be the same. 709 /// \param Name Name of the variable. 710 llvm::Constant *getOrCreateInternalVariable(llvm::Type *Ty, 711 const llvm::Twine &Name, 712 unsigned AddressSpace = 0); 713 714 /// Set of threadprivate variables with the generated initializer. 715 llvm::StringSet<> ThreadPrivateWithDefinition; 716 717 /// Set of declare target variables with the generated initializer. 718 llvm::StringSet<> DeclareTargetWithDefinition; 719 720 /// Emits initialization code for the threadprivate variables. 721 /// \param VDAddr Address of the global variable \a VD. 722 /// \param Ctor Pointer to a global init function for \a VD. 723 /// \param CopyCtor Pointer to a global copy function for \a VD. 724 /// \param Dtor Pointer to a global destructor function for \a VD. 725 /// \param Loc Location of threadprivate declaration. 726 void emitThreadPrivateVarInit(CodeGenFunction &CGF, Address VDAddr, 727 llvm::Value *Ctor, llvm::Value *CopyCtor, 728 llvm::Value *Dtor, SourceLocation Loc); 729 730 struct TaskResultTy { 731 llvm::Value *NewTask = nullptr; 732 llvm::Function *TaskEntry = nullptr; 733 llvm::Value *NewTaskNewTaskTTy = nullptr; 734 LValue TDBase; 735 const RecordDecl *KmpTaskTQTyRD = nullptr; 736 llvm::Value *TaskDupFn = nullptr; 737 }; 738 /// Emit task region for the task directive. The task region is emitted in 739 /// several steps: 740 /// 1. Emit a call to kmp_task_t *__kmpc_omp_task_alloc(ident_t *, kmp_int32 741 /// gtid, kmp_int32 flags, size_t sizeof_kmp_task_t, size_t sizeof_shareds, 742 /// kmp_routine_entry_t *task_entry). Here task_entry is a pointer to the 743 /// function: 744 /// kmp_int32 .omp_task_entry.(kmp_int32 gtid, kmp_task_t *tt) { 745 /// TaskFunction(gtid, tt->part_id, tt->shareds); 746 /// return 0; 747 /// } 748 /// 2. Copy a list of shared variables to field shareds of the resulting 749 /// structure kmp_task_t returned by the previous call (if any). 750 /// 3. Copy a pointer to destructions function to field destructions of the 751 /// resulting structure kmp_task_t. 752 /// \param D Current task directive. 753 /// \param TaskFunction An LLVM function with type void (*)(i32 /*gtid*/, i32 754 /// /*part_id*/, captured_struct */*__context*/); 755 /// \param SharedsTy A type which contains references the shared variables. 756 /// \param Shareds Context with the list of shared variables from the \p 757 /// TaskFunction. 758 /// \param Data Additional data for task generation like tiednsee, final 759 /// state, list of privates etc. 760 TaskResultTy emitTaskInit(CodeGenFunction &CGF, SourceLocation Loc, 761 const OMPExecutableDirective &D, 762 llvm::Function *TaskFunction, QualType SharedsTy, 763 Address Shareds, const OMPTaskDataTy &Data); 764 765 /// Returns default address space for the constant firstprivates, 0 by 766 /// default. 767 virtual unsigned getDefaultFirstprivateAddressSpace() const { return 0; } 768 769 public: 770 explicit CGOpenMPRuntime(CodeGenModule &CGM) 771 : CGOpenMPRuntime(CGM, ".", ".") {} 772 virtual ~CGOpenMPRuntime() {} 773 virtual void clear(); 774 775 /// Checks if the \p Body is the \a CompoundStmt and returns its child 776 /// statement iff there is only one that is not evaluatable at the compile 777 /// time. 778 static const Stmt *getSingleCompoundChild(ASTContext &Ctx, const Stmt *Body); 779 780 /// Get the platform-specific name separator. 781 std::string getName(ArrayRef<StringRef> Parts) const; 782 783 /// Emit code for the specified user defined reduction construct. 784 virtual void emitUserDefinedReduction(CodeGenFunction *CGF, 785 const OMPDeclareReductionDecl *D); 786 /// Get combiner/initializer for the specified user-defined reduction, if any. 787 virtual std::pair<llvm::Function *, llvm::Function *> 788 getUserDefinedReduction(const OMPDeclareReductionDecl *D); 789 790 /// Emits outlined function for the specified OpenMP parallel directive 791 /// \a D. This outlined function has type void(*)(kmp_int32 *ThreadID, 792 /// kmp_int32 BoundID, struct context_vars*). 793 /// \param D OpenMP directive. 794 /// \param ThreadIDVar Variable for thread id in the current OpenMP region. 795 /// \param InnermostKind Kind of innermost directive (for simple directives it 796 /// is a directive itself, for combined - its innermost directive). 797 /// \param CodeGen Code generation sequence for the \a D directive. 798 virtual llvm::Function *emitParallelOutlinedFunction( 799 const OMPExecutableDirective &D, const VarDecl *ThreadIDVar, 800 OpenMPDirectiveKind InnermostKind, const RegionCodeGenTy &CodeGen); 801 802 /// Emits outlined function for the specified OpenMP teams directive 803 /// \a D. This outlined function has type void(*)(kmp_int32 *ThreadID, 804 /// kmp_int32 BoundID, struct context_vars*). 805 /// \param D OpenMP directive. 806 /// \param ThreadIDVar Variable for thread id in the current OpenMP region. 807 /// \param InnermostKind Kind of innermost directive (for simple directives it 808 /// is a directive itself, for combined - its innermost directive). 809 /// \param CodeGen Code generation sequence for the \a D directive. 810 virtual llvm::Function *emitTeamsOutlinedFunction( 811 const OMPExecutableDirective &D, const VarDecl *ThreadIDVar, 812 OpenMPDirectiveKind InnermostKind, const RegionCodeGenTy &CodeGen); 813 814 /// Emits outlined function for the OpenMP task directive \a D. This 815 /// outlined function has type void(*)(kmp_int32 ThreadID, struct task_t* 816 /// TaskT). 817 /// \param D OpenMP directive. 818 /// \param ThreadIDVar Variable for thread id in the current OpenMP region. 819 /// \param PartIDVar Variable for partition id in the current OpenMP untied 820 /// task region. 821 /// \param TaskTVar Variable for task_t argument. 822 /// \param InnermostKind Kind of innermost directive (for simple directives it 823 /// is a directive itself, for combined - its innermost directive). 824 /// \param CodeGen Code generation sequence for the \a D directive. 825 /// \param Tied true if task is generated for tied task, false otherwise. 826 /// \param NumberOfParts Number of parts in untied task. Ignored for tied 827 /// tasks. 828 /// 829 virtual llvm::Function *emitTaskOutlinedFunction( 830 const OMPExecutableDirective &D, const VarDecl *ThreadIDVar, 831 const VarDecl *PartIDVar, const VarDecl *TaskTVar, 832 OpenMPDirectiveKind InnermostKind, const RegionCodeGenTy &CodeGen, 833 bool Tied, unsigned &NumberOfParts); 834 835 /// Cleans up references to the objects in finished function. 836 /// 837 virtual void functionFinished(CodeGenFunction &CGF); 838 839 /// Emits code for parallel or serial call of the \a OutlinedFn with 840 /// variables captured in a record which address is stored in \a 841 /// CapturedStruct. 842 /// \param OutlinedFn Outlined function to be run in parallel threads. Type of 843 /// this function is void(*)(kmp_int32 *, kmp_int32, struct context_vars*). 844 /// \param CapturedVars A pointer to the record with the references to 845 /// variables used in \a OutlinedFn function. 846 /// \param IfCond Condition in the associated 'if' clause, if it was 847 /// specified, nullptr otherwise. 848 /// 849 virtual void emitParallelCall(CodeGenFunction &CGF, SourceLocation Loc, 850 llvm::Function *OutlinedFn, 851 ArrayRef<llvm::Value *> CapturedVars, 852 const Expr *IfCond); 853 854 /// Emits a critical region. 855 /// \param CriticalName Name of the critical region. 856 /// \param CriticalOpGen Generator for the statement associated with the given 857 /// critical region. 858 /// \param Hint Value of the 'hint' clause (optional). 859 virtual void emitCriticalRegion(CodeGenFunction &CGF, StringRef CriticalName, 860 const RegionCodeGenTy &CriticalOpGen, 861 SourceLocation Loc, 862 const Expr *Hint = nullptr); 863 864 /// Emits a master region. 865 /// \param MasterOpGen Generator for the statement associated with the given 866 /// master region. 867 virtual void emitMasterRegion(CodeGenFunction &CGF, 868 const RegionCodeGenTy &MasterOpGen, 869 SourceLocation Loc); 870 871 /// Emits code for a taskyield directive. 872 virtual void emitTaskyieldCall(CodeGenFunction &CGF, SourceLocation Loc); 873 874 /// Emit a taskgroup region. 875 /// \param TaskgroupOpGen Generator for the statement associated with the 876 /// given taskgroup region. 877 virtual void emitTaskgroupRegion(CodeGenFunction &CGF, 878 const RegionCodeGenTy &TaskgroupOpGen, 879 SourceLocation Loc); 880 881 /// Emits a single region. 882 /// \param SingleOpGen Generator for the statement associated with the given 883 /// single region. 884 virtual void emitSingleRegion(CodeGenFunction &CGF, 885 const RegionCodeGenTy &SingleOpGen, 886 SourceLocation Loc, 887 ArrayRef<const Expr *> CopyprivateVars, 888 ArrayRef<const Expr *> DestExprs, 889 ArrayRef<const Expr *> SrcExprs, 890 ArrayRef<const Expr *> AssignmentOps); 891 892 /// Emit an ordered region. 893 /// \param OrderedOpGen Generator for the statement associated with the given 894 /// ordered region. 895 virtual void emitOrderedRegion(CodeGenFunction &CGF, 896 const RegionCodeGenTy &OrderedOpGen, 897 SourceLocation Loc, bool IsThreads); 898 899 /// Emit an implicit/explicit barrier for OpenMP threads. 900 /// \param Kind Directive for which this implicit barrier call must be 901 /// generated. Must be OMPD_barrier for explicit barrier generation. 902 /// \param EmitChecks true if need to emit checks for cancellation barriers. 903 /// \param ForceSimpleCall true simple barrier call must be emitted, false if 904 /// runtime class decides which one to emit (simple or with cancellation 905 /// checks). 906 /// 907 virtual void emitBarrierCall(CodeGenFunction &CGF, SourceLocation Loc, 908 OpenMPDirectiveKind Kind, 909 bool EmitChecks = true, 910 bool ForceSimpleCall = false); 911 912 /// Check if the specified \a ScheduleKind is static non-chunked. 913 /// This kind of worksharing directive is emitted without outer loop. 914 /// \param ScheduleKind Schedule kind specified in the 'schedule' clause. 915 /// \param Chunked True if chunk is specified in the clause. 916 /// 917 virtual bool isStaticNonchunked(OpenMPScheduleClauseKind ScheduleKind, 918 bool Chunked) const; 919 920 /// Check if the specified \a ScheduleKind is static non-chunked. 921 /// This kind of distribute directive is emitted without outer loop. 922 /// \param ScheduleKind Schedule kind specified in the 'dist_schedule' clause. 923 /// \param Chunked True if chunk is specified in the clause. 924 /// 925 virtual bool isStaticNonchunked(OpenMPDistScheduleClauseKind ScheduleKind, 926 bool Chunked) const; 927 928 /// Check if the specified \a ScheduleKind is static chunked. 929 /// \param ScheduleKind Schedule kind specified in the 'schedule' clause. 930 /// \param Chunked True if chunk is specified in the clause. 931 /// 932 virtual bool isStaticChunked(OpenMPScheduleClauseKind ScheduleKind, 933 bool Chunked) const; 934 935 /// Check if the specified \a ScheduleKind is static non-chunked. 936 /// \param ScheduleKind Schedule kind specified in the 'dist_schedule' clause. 937 /// \param Chunked True if chunk is specified in the clause. 938 /// 939 virtual bool isStaticChunked(OpenMPDistScheduleClauseKind ScheduleKind, 940 bool Chunked) const; 941 942 /// Check if the specified \a ScheduleKind is dynamic. 943 /// This kind of worksharing directive is emitted without outer loop. 944 /// \param ScheduleKind Schedule Kind specified in the 'schedule' clause. 945 /// 946 virtual bool isDynamic(OpenMPScheduleClauseKind ScheduleKind) const; 947 948 /// struct with the values to be passed to the dispatch runtime function 949 struct DispatchRTInput { 950 /// Loop lower bound 951 llvm::Value *LB = nullptr; 952 /// Loop upper bound 953 llvm::Value *UB = nullptr; 954 /// Chunk size specified using 'schedule' clause (nullptr if chunk 955 /// was not specified) 956 llvm::Value *Chunk = nullptr; 957 DispatchRTInput() = default; 958 DispatchRTInput(llvm::Value *LB, llvm::Value *UB, llvm::Value *Chunk) 959 : LB(LB), UB(UB), Chunk(Chunk) {} 960 }; 961 962 /// Call the appropriate runtime routine to initialize it before start 963 /// of loop. 964 965 /// This is used for non static scheduled types and when the ordered 966 /// clause is present on the loop construct. 967 /// Depending on the loop schedule, it is necessary to call some runtime 968 /// routine before start of the OpenMP loop to get the loop upper / lower 969 /// bounds \a LB and \a UB and stride \a ST. 970 /// 971 /// \param CGF Reference to current CodeGenFunction. 972 /// \param Loc Clang source location. 973 /// \param ScheduleKind Schedule kind, specified by the 'schedule' clause. 974 /// \param IVSize Size of the iteration variable in bits. 975 /// \param IVSigned Sign of the iteration variable. 976 /// \param Ordered true if loop is ordered, false otherwise. 977 /// \param DispatchValues struct containing llvm values for lower bound, upper 978 /// bound, and chunk expression. 979 /// For the default (nullptr) value, the chunk 1 will be used. 980 /// 981 virtual void emitForDispatchInit(CodeGenFunction &CGF, SourceLocation Loc, 982 const OpenMPScheduleTy &ScheduleKind, 983 unsigned IVSize, bool IVSigned, bool Ordered, 984 const DispatchRTInput &DispatchValues); 985 986 /// Struct with the values to be passed to the static runtime function 987 struct StaticRTInput { 988 /// Size of the iteration variable in bits. 989 unsigned IVSize = 0; 990 /// Sign of the iteration variable. 991 bool IVSigned = false; 992 /// true if loop is ordered, false otherwise. 993 bool Ordered = false; 994 /// Address of the output variable in which the flag of the last iteration 995 /// is returned. 996 Address IL = Address::invalid(); 997 /// Address of the output variable in which the lower iteration number is 998 /// returned. 999 Address LB = Address::invalid(); 1000 /// Address of the output variable in which the upper iteration number is 1001 /// returned. 1002 Address UB = Address::invalid(); 1003 /// Address of the output variable in which the stride value is returned 1004 /// necessary to generated the static_chunked scheduled loop. 1005 Address ST = Address::invalid(); 1006 /// Value of the chunk for the static_chunked scheduled loop. For the 1007 /// default (nullptr) value, the chunk 1 will be used. 1008 llvm::Value *Chunk = nullptr; 1009 StaticRTInput(unsigned IVSize, bool IVSigned, bool Ordered, Address IL, 1010 Address LB, Address UB, Address ST, 1011 llvm::Value *Chunk = nullptr) 1012 : IVSize(IVSize), IVSigned(IVSigned), Ordered(Ordered), IL(IL), LB(LB), 1013 UB(UB), ST(ST), Chunk(Chunk) {} 1014 }; 1015 /// Call the appropriate runtime routine to initialize it before start 1016 /// of loop. 1017 /// 1018 /// This is used only in case of static schedule, when the user did not 1019 /// specify a ordered clause on the loop construct. 1020 /// Depending on the loop schedule, it is necessary to call some runtime 1021 /// routine before start of the OpenMP loop to get the loop upper / lower 1022 /// bounds LB and UB and stride ST. 1023 /// 1024 /// \param CGF Reference to current CodeGenFunction. 1025 /// \param Loc Clang source location. 1026 /// \param DKind Kind of the directive. 1027 /// \param ScheduleKind Schedule kind, specified by the 'schedule' clause. 1028 /// \param Values Input arguments for the construct. 1029 /// 1030 virtual void emitForStaticInit(CodeGenFunction &CGF, SourceLocation Loc, 1031 OpenMPDirectiveKind DKind, 1032 const OpenMPScheduleTy &ScheduleKind, 1033 const StaticRTInput &Values); 1034 1035 /// 1036 /// \param CGF Reference to current CodeGenFunction. 1037 /// \param Loc Clang source location. 1038 /// \param SchedKind Schedule kind, specified by the 'dist_schedule' clause. 1039 /// \param Values Input arguments for the construct. 1040 /// 1041 virtual void emitDistributeStaticInit(CodeGenFunction &CGF, 1042 SourceLocation Loc, 1043 OpenMPDistScheduleClauseKind SchedKind, 1044 const StaticRTInput &Values); 1045 1046 /// Call the appropriate runtime routine to notify that we finished 1047 /// iteration of the ordered loop with the dynamic scheduling. 1048 /// 1049 /// \param CGF Reference to current CodeGenFunction. 1050 /// \param Loc Clang source location. 1051 /// \param IVSize Size of the iteration variable in bits. 1052 /// \param IVSigned Sign of the iteration variable. 1053 /// 1054 virtual void emitForOrderedIterationEnd(CodeGenFunction &CGF, 1055 SourceLocation Loc, unsigned IVSize, 1056 bool IVSigned); 1057 1058 /// Call the appropriate runtime routine to notify that we finished 1059 /// all the work with current loop. 1060 /// 1061 /// \param CGF Reference to current CodeGenFunction. 1062 /// \param Loc Clang source location. 1063 /// \param DKind Kind of the directive for which the static finish is emitted. 1064 /// 1065 virtual void emitForStaticFinish(CodeGenFunction &CGF, SourceLocation Loc, 1066 OpenMPDirectiveKind DKind); 1067 1068 /// Call __kmpc_dispatch_next( 1069 /// ident_t *loc, kmp_int32 tid, kmp_int32 *p_lastiter, 1070 /// kmp_int[32|64] *p_lower, kmp_int[32|64] *p_upper, 1071 /// kmp_int[32|64] *p_stride); 1072 /// \param IVSize Size of the iteration variable in bits. 1073 /// \param IVSigned Sign of the iteration variable. 1074 /// \param IL Address of the output variable in which the flag of the 1075 /// last iteration is returned. 1076 /// \param LB Address of the output variable in which the lower iteration 1077 /// number is returned. 1078 /// \param UB Address of the output variable in which the upper iteration 1079 /// number is returned. 1080 /// \param ST Address of the output variable in which the stride value is 1081 /// returned. 1082 virtual llvm::Value *emitForNext(CodeGenFunction &CGF, SourceLocation Loc, 1083 unsigned IVSize, bool IVSigned, 1084 Address IL, Address LB, 1085 Address UB, Address ST); 1086 1087 /// Emits call to void __kmpc_push_num_threads(ident_t *loc, kmp_int32 1088 /// global_tid, kmp_int32 num_threads) to generate code for 'num_threads' 1089 /// clause. 1090 /// \param NumThreads An integer value of threads. 1091 virtual void emitNumThreadsClause(CodeGenFunction &CGF, 1092 llvm::Value *NumThreads, 1093 SourceLocation Loc); 1094 1095 /// Emit call to void __kmpc_push_proc_bind(ident_t *loc, kmp_int32 1096 /// global_tid, int proc_bind) to generate code for 'proc_bind' clause. 1097 virtual void emitProcBindClause(CodeGenFunction &CGF, 1098 OpenMPProcBindClauseKind ProcBind, 1099 SourceLocation Loc); 1100 1101 /// Returns address of the threadprivate variable for the current 1102 /// thread. 1103 /// \param VD Threadprivate variable. 1104 /// \param VDAddr Address of the global variable \a VD. 1105 /// \param Loc Location of the reference to threadprivate var. 1106 /// \return Address of the threadprivate variable for the current thread. 1107 virtual Address getAddrOfThreadPrivate(CodeGenFunction &CGF, 1108 const VarDecl *VD, 1109 Address VDAddr, 1110 SourceLocation Loc); 1111 1112 /// Returns the address of the variable marked as declare target with link 1113 /// clause. 1114 virtual Address getAddrOfDeclareTargetLink(const VarDecl *VD); 1115 1116 /// Emit a code for initialization of threadprivate variable. It emits 1117 /// a call to runtime library which adds initial value to the newly created 1118 /// threadprivate variable (if it is not constant) and registers destructor 1119 /// for the variable (if any). 1120 /// \param VD Threadprivate variable. 1121 /// \param VDAddr Address of the global variable \a VD. 1122 /// \param Loc Location of threadprivate declaration. 1123 /// \param PerformInit true if initialization expression is not constant. 1124 virtual llvm::Function * 1125 emitThreadPrivateVarDefinition(const VarDecl *VD, Address VDAddr, 1126 SourceLocation Loc, bool PerformInit, 1127 CodeGenFunction *CGF = nullptr); 1128 1129 /// Emit a code for initialization of declare target variable. 1130 /// \param VD Declare target variable. 1131 /// \param Addr Address of the global variable \a VD. 1132 /// \param PerformInit true if initialization expression is not constant. 1133 virtual bool emitDeclareTargetVarDefinition(const VarDecl *VD, 1134 llvm::GlobalVariable *Addr, 1135 bool PerformInit); 1136 1137 /// Creates artificial threadprivate variable with name \p Name and type \p 1138 /// VarType. 1139 /// \param VarType Type of the artificial threadprivate variable. 1140 /// \param Name Name of the artificial threadprivate variable. 1141 virtual Address getAddrOfArtificialThreadPrivate(CodeGenFunction &CGF, 1142 QualType VarType, 1143 StringRef Name); 1144 1145 /// Emit flush of the variables specified in 'omp flush' directive. 1146 /// \param Vars List of variables to flush. 1147 virtual void emitFlush(CodeGenFunction &CGF, ArrayRef<const Expr *> Vars, 1148 SourceLocation Loc); 1149 1150 /// Emit task region for the task directive. The task region is 1151 /// emitted in several steps: 1152 /// 1. Emit a call to kmp_task_t *__kmpc_omp_task_alloc(ident_t *, kmp_int32 1153 /// gtid, kmp_int32 flags, size_t sizeof_kmp_task_t, size_t sizeof_shareds, 1154 /// kmp_routine_entry_t *task_entry). Here task_entry is a pointer to the 1155 /// function: 1156 /// kmp_int32 .omp_task_entry.(kmp_int32 gtid, kmp_task_t *tt) { 1157 /// TaskFunction(gtid, tt->part_id, tt->shareds); 1158 /// return 0; 1159 /// } 1160 /// 2. Copy a list of shared variables to field shareds of the resulting 1161 /// structure kmp_task_t returned by the previous call (if any). 1162 /// 3. Copy a pointer to destructions function to field destructions of the 1163 /// resulting structure kmp_task_t. 1164 /// 4. Emit a call to kmp_int32 __kmpc_omp_task(ident_t *, kmp_int32 gtid, 1165 /// kmp_task_t *new_task), where new_task is a resulting structure from 1166 /// previous items. 1167 /// \param D Current task directive. 1168 /// \param TaskFunction An LLVM function with type void (*)(i32 /*gtid*/, i32 1169 /// /*part_id*/, captured_struct */*__context*/); 1170 /// \param SharedsTy A type which contains references the shared variables. 1171 /// \param Shareds Context with the list of shared variables from the \p 1172 /// TaskFunction. 1173 /// \param IfCond Not a nullptr if 'if' clause was specified, nullptr 1174 /// otherwise. 1175 /// \param Data Additional data for task generation like tiednsee, final 1176 /// state, list of privates etc. 1177 virtual void emitTaskCall(CodeGenFunction &CGF, SourceLocation Loc, 1178 const OMPExecutableDirective &D, 1179 llvm::Function *TaskFunction, QualType SharedsTy, 1180 Address Shareds, const Expr *IfCond, 1181 const OMPTaskDataTy &Data); 1182 1183 /// Emit task region for the taskloop directive. The taskloop region is 1184 /// emitted in several steps: 1185 /// 1. Emit a call to kmp_task_t *__kmpc_omp_task_alloc(ident_t *, kmp_int32 1186 /// gtid, kmp_int32 flags, size_t sizeof_kmp_task_t, size_t sizeof_shareds, 1187 /// kmp_routine_entry_t *task_entry). Here task_entry is a pointer to the 1188 /// function: 1189 /// kmp_int32 .omp_task_entry.(kmp_int32 gtid, kmp_task_t *tt) { 1190 /// TaskFunction(gtid, tt->part_id, tt->shareds); 1191 /// return 0; 1192 /// } 1193 /// 2. Copy a list of shared variables to field shareds of the resulting 1194 /// structure kmp_task_t returned by the previous call (if any). 1195 /// 3. Copy a pointer to destructions function to field destructions of the 1196 /// resulting structure kmp_task_t. 1197 /// 4. Emit a call to void __kmpc_taskloop(ident_t *loc, int gtid, kmp_task_t 1198 /// *task, int if_val, kmp_uint64 *lb, kmp_uint64 *ub, kmp_int64 st, int 1199 /// nogroup, int sched, kmp_uint64 grainsize, void *task_dup ), where new_task 1200 /// is a resulting structure from 1201 /// previous items. 1202 /// \param D Current task directive. 1203 /// \param TaskFunction An LLVM function with type void (*)(i32 /*gtid*/, i32 1204 /// /*part_id*/, captured_struct */*__context*/); 1205 /// \param SharedsTy A type which contains references the shared variables. 1206 /// \param Shareds Context with the list of shared variables from the \p 1207 /// TaskFunction. 1208 /// \param IfCond Not a nullptr if 'if' clause was specified, nullptr 1209 /// otherwise. 1210 /// \param Data Additional data for task generation like tiednsee, final 1211 /// state, list of privates etc. 1212 virtual void emitTaskLoopCall(CodeGenFunction &CGF, SourceLocation Loc, 1213 const OMPLoopDirective &D, 1214 llvm::Function *TaskFunction, 1215 QualType SharedsTy, Address Shareds, 1216 const Expr *IfCond, const OMPTaskDataTy &Data); 1217 1218 /// Emit code for the directive that does not require outlining. 1219 /// 1220 /// \param InnermostKind Kind of innermost directive (for simple directives it 1221 /// is a directive itself, for combined - its innermost directive). 1222 /// \param CodeGen Code generation sequence for the \a D directive. 1223 /// \param HasCancel true if region has inner cancel directive, false 1224 /// otherwise. 1225 virtual void emitInlinedDirective(CodeGenFunction &CGF, 1226 OpenMPDirectiveKind InnermostKind, 1227 const RegionCodeGenTy &CodeGen, 1228 bool HasCancel = false); 1229 1230 /// Emits reduction function. 1231 /// \param ArgsType Array type containing pointers to reduction variables. 1232 /// \param Privates List of private copies for original reduction arguments. 1233 /// \param LHSExprs List of LHS in \a ReductionOps reduction operations. 1234 /// \param RHSExprs List of RHS in \a ReductionOps reduction operations. 1235 /// \param ReductionOps List of reduction operations in form 'LHS binop RHS' 1236 /// or 'operator binop(LHS, RHS)'. 1237 llvm::Function *emitReductionFunction(SourceLocation Loc, 1238 llvm::Type *ArgsType, 1239 ArrayRef<const Expr *> Privates, 1240 ArrayRef<const Expr *> LHSExprs, 1241 ArrayRef<const Expr *> RHSExprs, 1242 ArrayRef<const Expr *> ReductionOps); 1243 1244 /// Emits single reduction combiner 1245 void emitSingleReductionCombiner(CodeGenFunction &CGF, 1246 const Expr *ReductionOp, 1247 const Expr *PrivateRef, 1248 const DeclRefExpr *LHS, 1249 const DeclRefExpr *RHS); 1250 1251 struct ReductionOptionsTy { 1252 bool WithNowait; 1253 bool SimpleReduction; 1254 OpenMPDirectiveKind ReductionKind; 1255 }; 1256 /// Emit a code for reduction clause. Next code should be emitted for 1257 /// reduction: 1258 /// \code 1259 /// 1260 /// static kmp_critical_name lock = { 0 }; 1261 /// 1262 /// void reduce_func(void *lhs[<n>], void *rhs[<n>]) { 1263 /// ... 1264 /// *(Type<i>*)lhs[i] = RedOp<i>(*(Type<i>*)lhs[i], *(Type<i>*)rhs[i]); 1265 /// ... 1266 /// } 1267 /// 1268 /// ... 1269 /// void *RedList[<n>] = {&<RHSExprs>[0], ..., &<RHSExprs>[<n>-1]}; 1270 /// switch (__kmpc_reduce{_nowait}(<loc>, <gtid>, <n>, sizeof(RedList), 1271 /// RedList, reduce_func, &<lock>)) { 1272 /// case 1: 1273 /// ... 1274 /// <LHSExprs>[i] = RedOp<i>(*<LHSExprs>[i], *<RHSExprs>[i]); 1275 /// ... 1276 /// __kmpc_end_reduce{_nowait}(<loc>, <gtid>, &<lock>); 1277 /// break; 1278 /// case 2: 1279 /// ... 1280 /// Atomic(<LHSExprs>[i] = RedOp<i>(*<LHSExprs>[i], *<RHSExprs>[i])); 1281 /// ... 1282 /// break; 1283 /// default:; 1284 /// } 1285 /// \endcode 1286 /// 1287 /// \param Privates List of private copies for original reduction arguments. 1288 /// \param LHSExprs List of LHS in \a ReductionOps reduction operations. 1289 /// \param RHSExprs List of RHS in \a ReductionOps reduction operations. 1290 /// \param ReductionOps List of reduction operations in form 'LHS binop RHS' 1291 /// or 'operator binop(LHS, RHS)'. 1292 /// \param Options List of options for reduction codegen: 1293 /// WithNowait true if parent directive has also nowait clause, false 1294 /// otherwise. 1295 /// SimpleReduction Emit reduction operation only. Used for omp simd 1296 /// directive on the host. 1297 /// ReductionKind The kind of reduction to perform. 1298 virtual void emitReduction(CodeGenFunction &CGF, SourceLocation Loc, 1299 ArrayRef<const Expr *> Privates, 1300 ArrayRef<const Expr *> LHSExprs, 1301 ArrayRef<const Expr *> RHSExprs, 1302 ArrayRef<const Expr *> ReductionOps, 1303 ReductionOptionsTy Options); 1304 1305 /// Emit a code for initialization of task reduction clause. Next code 1306 /// should be emitted for reduction: 1307 /// \code 1308 /// 1309 /// _task_red_item_t red_data[n]; 1310 /// ... 1311 /// red_data[i].shar = &origs[i]; 1312 /// red_data[i].size = sizeof(origs[i]); 1313 /// red_data[i].f_init = (void*)RedInit<i>; 1314 /// red_data[i].f_fini = (void*)RedDest<i>; 1315 /// red_data[i].f_comb = (void*)RedOp<i>; 1316 /// red_data[i].flags = <Flag_i>; 1317 /// ... 1318 /// void* tg1 = __kmpc_task_reduction_init(gtid, n, red_data); 1319 /// \endcode 1320 /// 1321 /// \param LHSExprs List of LHS in \a Data.ReductionOps reduction operations. 1322 /// \param RHSExprs List of RHS in \a Data.ReductionOps reduction operations. 1323 /// \param Data Additional data for task generation like tiedness, final 1324 /// state, list of privates, reductions etc. 1325 virtual llvm::Value *emitTaskReductionInit(CodeGenFunction &CGF, 1326 SourceLocation Loc, 1327 ArrayRef<const Expr *> LHSExprs, 1328 ArrayRef<const Expr *> RHSExprs, 1329 const OMPTaskDataTy &Data); 1330 1331 /// Required to resolve existing problems in the runtime. Emits threadprivate 1332 /// variables to store the size of the VLAs/array sections for 1333 /// initializer/combiner/finalizer functions + emits threadprivate variable to 1334 /// store the pointer to the original reduction item for the custom 1335 /// initializer defined by declare reduction construct. 1336 /// \param RCG Allows to reuse an existing data for the reductions. 1337 /// \param N Reduction item for which fixups must be emitted. 1338 virtual void emitTaskReductionFixups(CodeGenFunction &CGF, SourceLocation Loc, 1339 ReductionCodeGen &RCG, unsigned N); 1340 1341 /// Get the address of `void *` type of the privatue copy of the reduction 1342 /// item specified by the \p SharedLVal. 1343 /// \param ReductionsPtr Pointer to the reduction data returned by the 1344 /// emitTaskReductionInit function. 1345 /// \param SharedLVal Address of the original reduction item. 1346 virtual Address getTaskReductionItem(CodeGenFunction &CGF, SourceLocation Loc, 1347 llvm::Value *ReductionsPtr, 1348 LValue SharedLVal); 1349 1350 /// Emit code for 'taskwait' directive. 1351 virtual void emitTaskwaitCall(CodeGenFunction &CGF, SourceLocation Loc); 1352 1353 /// Emit code for 'cancellation point' construct. 1354 /// \param CancelRegion Region kind for which the cancellation point must be 1355 /// emitted. 1356 /// 1357 virtual void emitCancellationPointCall(CodeGenFunction &CGF, 1358 SourceLocation Loc, 1359 OpenMPDirectiveKind CancelRegion); 1360 1361 /// Emit code for 'cancel' construct. 1362 /// \param IfCond Condition in the associated 'if' clause, if it was 1363 /// specified, nullptr otherwise. 1364 /// \param CancelRegion Region kind for which the cancel must be emitted. 1365 /// 1366 virtual void emitCancelCall(CodeGenFunction &CGF, SourceLocation Loc, 1367 const Expr *IfCond, 1368 OpenMPDirectiveKind CancelRegion); 1369 1370 /// Emit outilined function for 'target' directive. 1371 /// \param D Directive to emit. 1372 /// \param ParentName Name of the function that encloses the target region. 1373 /// \param OutlinedFn Outlined function value to be defined by this call. 1374 /// \param OutlinedFnID Outlined function ID value to be defined by this call. 1375 /// \param IsOffloadEntry True if the outlined function is an offload entry. 1376 /// \param CodeGen Code generation sequence for the \a D directive. 1377 /// An outlined function may not be an entry if, e.g. the if clause always 1378 /// evaluates to false. 1379 virtual void emitTargetOutlinedFunction(const OMPExecutableDirective &D, 1380 StringRef ParentName, 1381 llvm::Function *&OutlinedFn, 1382 llvm::Constant *&OutlinedFnID, 1383 bool IsOffloadEntry, 1384 const RegionCodeGenTy &CodeGen); 1385 1386 /// Emit code that pushes the trip count of loops associated with constructs 1387 /// 'target teams distribute' and 'teams distribute parallel for'. 1388 /// \param SizeEmitter Emits the int64 value for the number of iterations of 1389 /// the associated loop. 1390 virtual void emitTargetNumIterationsCall( 1391 CodeGenFunction &CGF, const OMPExecutableDirective &D, const Expr *Device, 1392 const llvm::function_ref<llvm::Value *( 1393 CodeGenFunction &CGF, const OMPLoopDirective &D)> &SizeEmitter); 1394 1395 /// Emit the target offloading code associated with \a D. The emitted 1396 /// code attempts offloading the execution to the device, an the event of 1397 /// a failure it executes the host version outlined in \a OutlinedFn. 1398 /// \param D Directive to emit. 1399 /// \param OutlinedFn Host version of the code to be offloaded. 1400 /// \param OutlinedFnID ID of host version of the code to be offloaded. 1401 /// \param IfCond Expression evaluated in if clause associated with the target 1402 /// directive, or null if no if clause is used. 1403 /// \param Device Expression evaluated in device clause associated with the 1404 /// target directive, or null if no device clause is used. 1405 virtual void emitTargetCall(CodeGenFunction &CGF, 1406 const OMPExecutableDirective &D, 1407 llvm::Function *OutlinedFn, 1408 llvm::Value *OutlinedFnID, const Expr *IfCond, 1409 const Expr *Device); 1410 1411 /// Emit the target regions enclosed in \a GD function definition or 1412 /// the function itself in case it is a valid device function. Returns true if 1413 /// \a GD was dealt with successfully. 1414 /// \param GD Function to scan. 1415 virtual bool emitTargetFunctions(GlobalDecl GD); 1416 1417 /// Emit the global variable if it is a valid device global variable. 1418 /// Returns true if \a GD was dealt with successfully. 1419 /// \param GD Variable declaration to emit. 1420 virtual bool emitTargetGlobalVariable(GlobalDecl GD); 1421 1422 /// Checks if the provided global decl \a GD is a declare target variable and 1423 /// registers it when emitting code for the host. 1424 virtual void registerTargetGlobalVariable(const VarDecl *VD, 1425 llvm::Constant *Addr); 1426 1427 /// Registers provided target firstprivate variable as global on the 1428 /// target. 1429 llvm::Constant *registerTargetFirstprivateCopy(CodeGenFunction &CGF, 1430 const VarDecl *VD); 1431 1432 /// Emit the global \a GD if it is meaningful for the target. Returns 1433 /// if it was emitted successfully. 1434 /// \param GD Global to scan. 1435 virtual bool emitTargetGlobal(GlobalDecl GD); 1436 1437 /// Creates the offloading descriptor in the event any target region 1438 /// was emitted in the current module and return the function that registers 1439 /// it. 1440 virtual llvm::Function *emitRegistrationFunction(); 1441 1442 /// Emits code for teams call of the \a OutlinedFn with 1443 /// variables captured in a record which address is stored in \a 1444 /// CapturedStruct. 1445 /// \param OutlinedFn Outlined function to be run by team masters. Type of 1446 /// this function is void(*)(kmp_int32 *, kmp_int32, struct context_vars*). 1447 /// \param CapturedVars A pointer to the record with the references to 1448 /// variables used in \a OutlinedFn function. 1449 /// 1450 virtual void emitTeamsCall(CodeGenFunction &CGF, 1451 const OMPExecutableDirective &D, 1452 SourceLocation Loc, llvm::Function *OutlinedFn, 1453 ArrayRef<llvm::Value *> CapturedVars); 1454 1455 /// Emits call to void __kmpc_push_num_teams(ident_t *loc, kmp_int32 1456 /// global_tid, kmp_int32 num_teams, kmp_int32 thread_limit) to generate code 1457 /// for num_teams clause. 1458 /// \param NumTeams An integer expression of teams. 1459 /// \param ThreadLimit An integer expression of threads. 1460 virtual void emitNumTeamsClause(CodeGenFunction &CGF, const Expr *NumTeams, 1461 const Expr *ThreadLimit, SourceLocation Loc); 1462 1463 /// Struct that keeps all the relevant information that should be kept 1464 /// throughout a 'target data' region. 1465 class TargetDataInfo { 1466 /// Set to true if device pointer information have to be obtained. 1467 bool RequiresDevicePointerInfo = false; 1468 1469 public: 1470 /// The array of base pointer passed to the runtime library. 1471 llvm::Value *BasePointersArray = nullptr; 1472 /// The array of section pointers passed to the runtime library. 1473 llvm::Value *PointersArray = nullptr; 1474 /// The array of sizes passed to the runtime library. 1475 llvm::Value *SizesArray = nullptr; 1476 /// The array of map types passed to the runtime library. 1477 llvm::Value *MapTypesArray = nullptr; 1478 /// The total number of pointers passed to the runtime library. 1479 unsigned NumberOfPtrs = 0u; 1480 /// Map between the a declaration of a capture and the corresponding base 1481 /// pointer address where the runtime returns the device pointers. 1482 llvm::DenseMap<const ValueDecl *, Address> CaptureDeviceAddrMap; 1483 1484 explicit TargetDataInfo() {} 1485 explicit TargetDataInfo(bool RequiresDevicePointerInfo) 1486 : RequiresDevicePointerInfo(RequiresDevicePointerInfo) {} 1487 /// Clear information about the data arrays. 1488 void clearArrayInfo() { 1489 BasePointersArray = nullptr; 1490 PointersArray = nullptr; 1491 SizesArray = nullptr; 1492 MapTypesArray = nullptr; 1493 NumberOfPtrs = 0u; 1494 } 1495 /// Return true if the current target data information has valid arrays. 1496 bool isValid() { 1497 return BasePointersArray && PointersArray && SizesArray && 1498 MapTypesArray && NumberOfPtrs; 1499 } 1500 bool requiresDevicePointerInfo() { return RequiresDevicePointerInfo; } 1501 }; 1502 1503 /// Emit the target data mapping code associated with \a D. 1504 /// \param D Directive to emit. 1505 /// \param IfCond Expression evaluated in if clause associated with the 1506 /// target directive, or null if no device clause is used. 1507 /// \param Device Expression evaluated in device clause associated with the 1508 /// target directive, or null if no device clause is used. 1509 /// \param Info A record used to store information that needs to be preserved 1510 /// until the region is closed. 1511 virtual void emitTargetDataCalls(CodeGenFunction &CGF, 1512 const OMPExecutableDirective &D, 1513 const Expr *IfCond, const Expr *Device, 1514 const RegionCodeGenTy &CodeGen, 1515 TargetDataInfo &Info); 1516 1517 /// Emit the data mapping/movement code associated with the directive 1518 /// \a D that should be of the form 'target [{enter|exit} data | update]'. 1519 /// \param D Directive to emit. 1520 /// \param IfCond Expression evaluated in if clause associated with the target 1521 /// directive, or null if no if clause is used. 1522 /// \param Device Expression evaluated in device clause associated with the 1523 /// target directive, or null if no device clause is used. 1524 virtual void emitTargetDataStandAloneCall(CodeGenFunction &CGF, 1525 const OMPExecutableDirective &D, 1526 const Expr *IfCond, 1527 const Expr *Device); 1528 1529 /// Marks function \a Fn with properly mangled versions of vector functions. 1530 /// \param FD Function marked as 'declare simd'. 1531 /// \param Fn LLVM function that must be marked with 'declare simd' 1532 /// attributes. 1533 virtual void emitDeclareSimdFunction(const FunctionDecl *FD, 1534 llvm::Function *Fn); 1535 1536 /// Emit initialization for doacross loop nesting support. 1537 /// \param D Loop-based construct used in doacross nesting construct. 1538 virtual void emitDoacrossInit(CodeGenFunction &CGF, const OMPLoopDirective &D, 1539 ArrayRef<Expr *> NumIterations); 1540 1541 /// Emit code for doacross ordered directive with 'depend' clause. 1542 /// \param C 'depend' clause with 'sink|source' dependency kind. 1543 virtual void emitDoacrossOrdered(CodeGenFunction &CGF, 1544 const OMPDependClause *C); 1545 1546 /// Translates the native parameter of outlined function if this is required 1547 /// for target. 1548 /// \param FD Field decl from captured record for the parameter. 1549 /// \param NativeParam Parameter itself. 1550 virtual const VarDecl *translateParameter(const FieldDecl *FD, 1551 const VarDecl *NativeParam) const { 1552 return NativeParam; 1553 } 1554 1555 /// Gets the address of the native argument basing on the address of the 1556 /// target-specific parameter. 1557 /// \param NativeParam Parameter itself. 1558 /// \param TargetParam Corresponding target-specific parameter. 1559 virtual Address getParameterAddress(CodeGenFunction &CGF, 1560 const VarDecl *NativeParam, 1561 const VarDecl *TargetParam) const; 1562 1563 /// Choose default schedule type and chunk value for the 1564 /// dist_schedule clause. 1565 virtual void getDefaultDistScheduleAndChunk(CodeGenFunction &CGF, 1566 const OMPLoopDirective &S, OpenMPDistScheduleClauseKind &ScheduleKind, 1567 llvm::Value *&Chunk) const {} 1568 1569 /// Choose default schedule type and chunk value for the 1570 /// schedule clause. 1571 virtual void getDefaultScheduleAndChunk(CodeGenFunction &CGF, 1572 const OMPLoopDirective &S, OpenMPScheduleClauseKind &ScheduleKind, 1573 const Expr *&ChunkExpr) const; 1574 1575 /// Emits call of the outlined function with the provided arguments, 1576 /// translating these arguments to correct target-specific arguments. 1577 virtual void 1578 emitOutlinedFunctionCall(CodeGenFunction &CGF, SourceLocation Loc, 1579 llvm::FunctionCallee OutlinedFn, 1580 ArrayRef<llvm::Value *> Args = llvm::None) const; 1581 1582 /// Emits OpenMP-specific function prolog. 1583 /// Required for device constructs. 1584 virtual void emitFunctionProlog(CodeGenFunction &CGF, const Decl *D) {} 1585 1586 /// Gets the OpenMP-specific address of the local variable. 1587 virtual Address getAddressOfLocalVariable(CodeGenFunction &CGF, 1588 const VarDecl *VD); 1589 1590 /// Marks the declaration as already emitted for the device code and returns 1591 /// true, if it was marked already, and false, otherwise. 1592 bool markAsGlobalTarget(GlobalDecl GD); 1593 1594 /// Emit deferred declare target variables marked for deferred emission. 1595 void emitDeferredTargetDecls() const; 1596 1597 /// Adjust some parameters for the target-based directives, like addresses of 1598 /// the variables captured by reference in lambdas. 1599 virtual void 1600 adjustTargetSpecificDataForLambdas(CodeGenFunction &CGF, 1601 const OMPExecutableDirective &D) const; 1602 1603 /// Perform check on requires decl to ensure that target architecture 1604 /// supports unified addressing 1605 virtual void checkArchForUnifiedAddressing(const OMPRequiresDecl *D) const {} 1606 1607 /// Checks if the variable has associated OMPAllocateDeclAttr attribute with 1608 /// the predefined allocator and translates it into the corresponding address 1609 /// space. 1610 virtual bool hasAllocateAttributeForGlobalVar(const VarDecl *VD, LangAS &AS); 1611 }; 1612 1613 /// Class supports emissionof SIMD-only code. 1614 class CGOpenMPSIMDRuntime final : public CGOpenMPRuntime { 1615 public: 1616 explicit CGOpenMPSIMDRuntime(CodeGenModule &CGM) : CGOpenMPRuntime(CGM) {} 1617 ~CGOpenMPSIMDRuntime() override {} 1618 1619 /// Emits outlined function for the specified OpenMP parallel directive 1620 /// \a D. This outlined function has type void(*)(kmp_int32 *ThreadID, 1621 /// kmp_int32 BoundID, struct context_vars*). 1622 /// \param D OpenMP directive. 1623 /// \param ThreadIDVar Variable for thread id in the current OpenMP region. 1624 /// \param InnermostKind Kind of innermost directive (for simple directives it 1625 /// is a directive itself, for combined - its innermost directive). 1626 /// \param CodeGen Code generation sequence for the \a D directive. 1627 llvm::Function * 1628 emitParallelOutlinedFunction(const OMPExecutableDirective &D, 1629 const VarDecl *ThreadIDVar, 1630 OpenMPDirectiveKind InnermostKind, 1631 const RegionCodeGenTy &CodeGen) override; 1632 1633 /// Emits outlined function for the specified OpenMP teams directive 1634 /// \a D. This outlined function has type void(*)(kmp_int32 *ThreadID, 1635 /// kmp_int32 BoundID, struct context_vars*). 1636 /// \param D OpenMP directive. 1637 /// \param ThreadIDVar Variable for thread id in the current OpenMP region. 1638 /// \param InnermostKind Kind of innermost directive (for simple directives it 1639 /// is a directive itself, for combined - its innermost directive). 1640 /// \param CodeGen Code generation sequence for the \a D directive. 1641 llvm::Function * 1642 emitTeamsOutlinedFunction(const OMPExecutableDirective &D, 1643 const VarDecl *ThreadIDVar, 1644 OpenMPDirectiveKind InnermostKind, 1645 const RegionCodeGenTy &CodeGen) override; 1646 1647 /// Emits outlined function for the OpenMP task directive \a D. This 1648 /// outlined function has type void(*)(kmp_int32 ThreadID, struct task_t* 1649 /// TaskT). 1650 /// \param D OpenMP directive. 1651 /// \param ThreadIDVar Variable for thread id in the current OpenMP region. 1652 /// \param PartIDVar Variable for partition id in the current OpenMP untied 1653 /// task region. 1654 /// \param TaskTVar Variable for task_t argument. 1655 /// \param InnermostKind Kind of innermost directive (for simple directives it 1656 /// is a directive itself, for combined - its innermost directive). 1657 /// \param CodeGen Code generation sequence for the \a D directive. 1658 /// \param Tied true if task is generated for tied task, false otherwise. 1659 /// \param NumberOfParts Number of parts in untied task. Ignored for tied 1660 /// tasks. 1661 /// 1662 llvm::Function *emitTaskOutlinedFunction( 1663 const OMPExecutableDirective &D, const VarDecl *ThreadIDVar, 1664 const VarDecl *PartIDVar, const VarDecl *TaskTVar, 1665 OpenMPDirectiveKind InnermostKind, const RegionCodeGenTy &CodeGen, 1666 bool Tied, unsigned &NumberOfParts) override; 1667 1668 /// Emits code for parallel or serial call of the \a OutlinedFn with 1669 /// variables captured in a record which address is stored in \a 1670 /// CapturedStruct. 1671 /// \param OutlinedFn Outlined function to be run in parallel threads. Type of 1672 /// this function is void(*)(kmp_int32 *, kmp_int32, struct context_vars*). 1673 /// \param CapturedVars A pointer to the record with the references to 1674 /// variables used in \a OutlinedFn function. 1675 /// \param IfCond Condition in the associated 'if' clause, if it was 1676 /// specified, nullptr otherwise. 1677 /// 1678 void emitParallelCall(CodeGenFunction &CGF, SourceLocation Loc, 1679 llvm::Function *OutlinedFn, 1680 ArrayRef<llvm::Value *> CapturedVars, 1681 const Expr *IfCond) override; 1682 1683 /// Emits a critical region. 1684 /// \param CriticalName Name of the critical region. 1685 /// \param CriticalOpGen Generator for the statement associated with the given 1686 /// critical region. 1687 /// \param Hint Value of the 'hint' clause (optional). 1688 void emitCriticalRegion(CodeGenFunction &CGF, StringRef CriticalName, 1689 const RegionCodeGenTy &CriticalOpGen, 1690 SourceLocation Loc, 1691 const Expr *Hint = nullptr) override; 1692 1693 /// Emits a master region. 1694 /// \param MasterOpGen Generator for the statement associated with the given 1695 /// master region. 1696 void emitMasterRegion(CodeGenFunction &CGF, 1697 const RegionCodeGenTy &MasterOpGen, 1698 SourceLocation Loc) override; 1699 1700 /// Emits code for a taskyield directive. 1701 void emitTaskyieldCall(CodeGenFunction &CGF, SourceLocation Loc) override; 1702 1703 /// Emit a taskgroup region. 1704 /// \param TaskgroupOpGen Generator for the statement associated with the 1705 /// given taskgroup region. 1706 void emitTaskgroupRegion(CodeGenFunction &CGF, 1707 const RegionCodeGenTy &TaskgroupOpGen, 1708 SourceLocation Loc) override; 1709 1710 /// Emits a single region. 1711 /// \param SingleOpGen Generator for the statement associated with the given 1712 /// single region. 1713 void emitSingleRegion(CodeGenFunction &CGF, 1714 const RegionCodeGenTy &SingleOpGen, SourceLocation Loc, 1715 ArrayRef<const Expr *> CopyprivateVars, 1716 ArrayRef<const Expr *> DestExprs, 1717 ArrayRef<const Expr *> SrcExprs, 1718 ArrayRef<const Expr *> AssignmentOps) override; 1719 1720 /// Emit an ordered region. 1721 /// \param OrderedOpGen Generator for the statement associated with the given 1722 /// ordered region. 1723 void emitOrderedRegion(CodeGenFunction &CGF, 1724 const RegionCodeGenTy &OrderedOpGen, 1725 SourceLocation Loc, bool IsThreads) override; 1726 1727 /// Emit an implicit/explicit barrier for OpenMP threads. 1728 /// \param Kind Directive for which this implicit barrier call must be 1729 /// generated. Must be OMPD_barrier for explicit barrier generation. 1730 /// \param EmitChecks true if need to emit checks for cancellation barriers. 1731 /// \param ForceSimpleCall true simple barrier call must be emitted, false if 1732 /// runtime class decides which one to emit (simple or with cancellation 1733 /// checks). 1734 /// 1735 void emitBarrierCall(CodeGenFunction &CGF, SourceLocation Loc, 1736 OpenMPDirectiveKind Kind, bool EmitChecks = true, 1737 bool ForceSimpleCall = false) override; 1738 1739 /// This is used for non static scheduled types and when the ordered 1740 /// clause is present on the loop construct. 1741 /// Depending on the loop schedule, it is necessary to call some runtime 1742 /// routine before start of the OpenMP loop to get the loop upper / lower 1743 /// bounds \a LB and \a UB and stride \a ST. 1744 /// 1745 /// \param CGF Reference to current CodeGenFunction. 1746 /// \param Loc Clang source location. 1747 /// \param ScheduleKind Schedule kind, specified by the 'schedule' clause. 1748 /// \param IVSize Size of the iteration variable in bits. 1749 /// \param IVSigned Sign of the iteration variable. 1750 /// \param Ordered true if loop is ordered, false otherwise. 1751 /// \param DispatchValues struct containing llvm values for lower bound, upper 1752 /// bound, and chunk expression. 1753 /// For the default (nullptr) value, the chunk 1 will be used. 1754 /// 1755 void emitForDispatchInit(CodeGenFunction &CGF, SourceLocation Loc, 1756 const OpenMPScheduleTy &ScheduleKind, 1757 unsigned IVSize, bool IVSigned, bool Ordered, 1758 const DispatchRTInput &DispatchValues) override; 1759 1760 /// Call the appropriate runtime routine to initialize it before start 1761 /// of loop. 1762 /// 1763 /// This is used only in case of static schedule, when the user did not 1764 /// specify a ordered clause on the loop construct. 1765 /// Depending on the loop schedule, it is necessary to call some runtime 1766 /// routine before start of the OpenMP loop to get the loop upper / lower 1767 /// bounds LB and UB and stride ST. 1768 /// 1769 /// \param CGF Reference to current CodeGenFunction. 1770 /// \param Loc Clang source location. 1771 /// \param DKind Kind of the directive. 1772 /// \param ScheduleKind Schedule kind, specified by the 'schedule' clause. 1773 /// \param Values Input arguments for the construct. 1774 /// 1775 void emitForStaticInit(CodeGenFunction &CGF, SourceLocation Loc, 1776 OpenMPDirectiveKind DKind, 1777 const OpenMPScheduleTy &ScheduleKind, 1778 const StaticRTInput &Values) override; 1779 1780 /// 1781 /// \param CGF Reference to current CodeGenFunction. 1782 /// \param Loc Clang source location. 1783 /// \param SchedKind Schedule kind, specified by the 'dist_schedule' clause. 1784 /// \param Values Input arguments for the construct. 1785 /// 1786 void emitDistributeStaticInit(CodeGenFunction &CGF, SourceLocation Loc, 1787 OpenMPDistScheduleClauseKind SchedKind, 1788 const StaticRTInput &Values) override; 1789 1790 /// Call the appropriate runtime routine to notify that we finished 1791 /// iteration of the ordered loop with the dynamic scheduling. 1792 /// 1793 /// \param CGF Reference to current CodeGenFunction. 1794 /// \param Loc Clang source location. 1795 /// \param IVSize Size of the iteration variable in bits. 1796 /// \param IVSigned Sign of the iteration variable. 1797 /// 1798 void emitForOrderedIterationEnd(CodeGenFunction &CGF, SourceLocation Loc, 1799 unsigned IVSize, bool IVSigned) override; 1800 1801 /// Call the appropriate runtime routine to notify that we finished 1802 /// all the work with current loop. 1803 /// 1804 /// \param CGF Reference to current CodeGenFunction. 1805 /// \param Loc Clang source location. 1806 /// \param DKind Kind of the directive for which the static finish is emitted. 1807 /// 1808 void emitForStaticFinish(CodeGenFunction &CGF, SourceLocation Loc, 1809 OpenMPDirectiveKind DKind) override; 1810 1811 /// Call __kmpc_dispatch_next( 1812 /// ident_t *loc, kmp_int32 tid, kmp_int32 *p_lastiter, 1813 /// kmp_int[32|64] *p_lower, kmp_int[32|64] *p_upper, 1814 /// kmp_int[32|64] *p_stride); 1815 /// \param IVSize Size of the iteration variable in bits. 1816 /// \param IVSigned Sign of the iteration variable. 1817 /// \param IL Address of the output variable in which the flag of the 1818 /// last iteration is returned. 1819 /// \param LB Address of the output variable in which the lower iteration 1820 /// number is returned. 1821 /// \param UB Address of the output variable in which the upper iteration 1822 /// number is returned. 1823 /// \param ST Address of the output variable in which the stride value is 1824 /// returned. 1825 llvm::Value *emitForNext(CodeGenFunction &CGF, SourceLocation Loc, 1826 unsigned IVSize, bool IVSigned, Address IL, 1827 Address LB, Address UB, Address ST) override; 1828 1829 /// Emits call to void __kmpc_push_num_threads(ident_t *loc, kmp_int32 1830 /// global_tid, kmp_int32 num_threads) to generate code for 'num_threads' 1831 /// clause. 1832 /// \param NumThreads An integer value of threads. 1833 void emitNumThreadsClause(CodeGenFunction &CGF, llvm::Value *NumThreads, 1834 SourceLocation Loc) override; 1835 1836 /// Emit call to void __kmpc_push_proc_bind(ident_t *loc, kmp_int32 1837 /// global_tid, int proc_bind) to generate code for 'proc_bind' clause. 1838 void emitProcBindClause(CodeGenFunction &CGF, 1839 OpenMPProcBindClauseKind ProcBind, 1840 SourceLocation Loc) override; 1841 1842 /// Returns address of the threadprivate variable for the current 1843 /// thread. 1844 /// \param VD Threadprivate variable. 1845 /// \param VDAddr Address of the global variable \a VD. 1846 /// \param Loc Location of the reference to threadprivate var. 1847 /// \return Address of the threadprivate variable for the current thread. 1848 Address getAddrOfThreadPrivate(CodeGenFunction &CGF, const VarDecl *VD, 1849 Address VDAddr, SourceLocation Loc) override; 1850 1851 /// Emit a code for initialization of threadprivate variable. It emits 1852 /// a call to runtime library which adds initial value to the newly created 1853 /// threadprivate variable (if it is not constant) and registers destructor 1854 /// for the variable (if any). 1855 /// \param VD Threadprivate variable. 1856 /// \param VDAddr Address of the global variable \a VD. 1857 /// \param Loc Location of threadprivate declaration. 1858 /// \param PerformInit true if initialization expression is not constant. 1859 llvm::Function * 1860 emitThreadPrivateVarDefinition(const VarDecl *VD, Address VDAddr, 1861 SourceLocation Loc, bool PerformInit, 1862 CodeGenFunction *CGF = nullptr) override; 1863 1864 /// Creates artificial threadprivate variable with name \p Name and type \p 1865 /// VarType. 1866 /// \param VarType Type of the artificial threadprivate variable. 1867 /// \param Name Name of the artificial threadprivate variable. 1868 Address getAddrOfArtificialThreadPrivate(CodeGenFunction &CGF, 1869 QualType VarType, 1870 StringRef Name) override; 1871 1872 /// Emit flush of the variables specified in 'omp flush' directive. 1873 /// \param Vars List of variables to flush. 1874 void emitFlush(CodeGenFunction &CGF, ArrayRef<const Expr *> Vars, 1875 SourceLocation Loc) override; 1876 1877 /// Emit task region for the task directive. The task region is 1878 /// emitted in several steps: 1879 /// 1. Emit a call to kmp_task_t *__kmpc_omp_task_alloc(ident_t *, kmp_int32 1880 /// gtid, kmp_int32 flags, size_t sizeof_kmp_task_t, size_t sizeof_shareds, 1881 /// kmp_routine_entry_t *task_entry). Here task_entry is a pointer to the 1882 /// function: 1883 /// kmp_int32 .omp_task_entry.(kmp_int32 gtid, kmp_task_t *tt) { 1884 /// TaskFunction(gtid, tt->part_id, tt->shareds); 1885 /// return 0; 1886 /// } 1887 /// 2. Copy a list of shared variables to field shareds of the resulting 1888 /// structure kmp_task_t returned by the previous call (if any). 1889 /// 3. Copy a pointer to destructions function to field destructions of the 1890 /// resulting structure kmp_task_t. 1891 /// 4. Emit a call to kmp_int32 __kmpc_omp_task(ident_t *, kmp_int32 gtid, 1892 /// kmp_task_t *new_task), where new_task is a resulting structure from 1893 /// previous items. 1894 /// \param D Current task directive. 1895 /// \param TaskFunction An LLVM function with type void (*)(i32 /*gtid*/, i32 1896 /// /*part_id*/, captured_struct */*__context*/); 1897 /// \param SharedsTy A type which contains references the shared variables. 1898 /// \param Shareds Context with the list of shared variables from the \p 1899 /// TaskFunction. 1900 /// \param IfCond Not a nullptr if 'if' clause was specified, nullptr 1901 /// otherwise. 1902 /// \param Data Additional data for task generation like tiednsee, final 1903 /// state, list of privates etc. 1904 void emitTaskCall(CodeGenFunction &CGF, SourceLocation Loc, 1905 const OMPExecutableDirective &D, 1906 llvm::Function *TaskFunction, QualType SharedsTy, 1907 Address Shareds, const Expr *IfCond, 1908 const OMPTaskDataTy &Data) override; 1909 1910 /// Emit task region for the taskloop directive. The taskloop region is 1911 /// emitted in several steps: 1912 /// 1. Emit a call to kmp_task_t *__kmpc_omp_task_alloc(ident_t *, kmp_int32 1913 /// gtid, kmp_int32 flags, size_t sizeof_kmp_task_t, size_t sizeof_shareds, 1914 /// kmp_routine_entry_t *task_entry). Here task_entry is a pointer to the 1915 /// function: 1916 /// kmp_int32 .omp_task_entry.(kmp_int32 gtid, kmp_task_t *tt) { 1917 /// TaskFunction(gtid, tt->part_id, tt->shareds); 1918 /// return 0; 1919 /// } 1920 /// 2. Copy a list of shared variables to field shareds of the resulting 1921 /// structure kmp_task_t returned by the previous call (if any). 1922 /// 3. Copy a pointer to destructions function to field destructions of the 1923 /// resulting structure kmp_task_t. 1924 /// 4. Emit a call to void __kmpc_taskloop(ident_t *loc, int gtid, kmp_task_t 1925 /// *task, int if_val, kmp_uint64 *lb, kmp_uint64 *ub, kmp_int64 st, int 1926 /// nogroup, int sched, kmp_uint64 grainsize, void *task_dup ), where new_task 1927 /// is a resulting structure from 1928 /// previous items. 1929 /// \param D Current task directive. 1930 /// \param TaskFunction An LLVM function with type void (*)(i32 /*gtid*/, i32 1931 /// /*part_id*/, captured_struct */*__context*/); 1932 /// \param SharedsTy A type which contains references the shared variables. 1933 /// \param Shareds Context with the list of shared variables from the \p 1934 /// TaskFunction. 1935 /// \param IfCond Not a nullptr if 'if' clause was specified, nullptr 1936 /// otherwise. 1937 /// \param Data Additional data for task generation like tiednsee, final 1938 /// state, list of privates etc. 1939 void emitTaskLoopCall(CodeGenFunction &CGF, SourceLocation Loc, 1940 const OMPLoopDirective &D, llvm::Function *TaskFunction, 1941 QualType SharedsTy, Address Shareds, const Expr *IfCond, 1942 const OMPTaskDataTy &Data) override; 1943 1944 /// Emit a code for reduction clause. Next code should be emitted for 1945 /// reduction: 1946 /// \code 1947 /// 1948 /// static kmp_critical_name lock = { 0 }; 1949 /// 1950 /// void reduce_func(void *lhs[<n>], void *rhs[<n>]) { 1951 /// ... 1952 /// *(Type<i>*)lhs[i] = RedOp<i>(*(Type<i>*)lhs[i], *(Type<i>*)rhs[i]); 1953 /// ... 1954 /// } 1955 /// 1956 /// ... 1957 /// void *RedList[<n>] = {&<RHSExprs>[0], ..., &<RHSExprs>[<n>-1]}; 1958 /// switch (__kmpc_reduce{_nowait}(<loc>, <gtid>, <n>, sizeof(RedList), 1959 /// RedList, reduce_func, &<lock>)) { 1960 /// case 1: 1961 /// ... 1962 /// <LHSExprs>[i] = RedOp<i>(*<LHSExprs>[i], *<RHSExprs>[i]); 1963 /// ... 1964 /// __kmpc_end_reduce{_nowait}(<loc>, <gtid>, &<lock>); 1965 /// break; 1966 /// case 2: 1967 /// ... 1968 /// Atomic(<LHSExprs>[i] = RedOp<i>(*<LHSExprs>[i], *<RHSExprs>[i])); 1969 /// ... 1970 /// break; 1971 /// default:; 1972 /// } 1973 /// \endcode 1974 /// 1975 /// \param Privates List of private copies for original reduction arguments. 1976 /// \param LHSExprs List of LHS in \a ReductionOps reduction operations. 1977 /// \param RHSExprs List of RHS in \a ReductionOps reduction operations. 1978 /// \param ReductionOps List of reduction operations in form 'LHS binop RHS' 1979 /// or 'operator binop(LHS, RHS)'. 1980 /// \param Options List of options for reduction codegen: 1981 /// WithNowait true if parent directive has also nowait clause, false 1982 /// otherwise. 1983 /// SimpleReduction Emit reduction operation only. Used for omp simd 1984 /// directive on the host. 1985 /// ReductionKind The kind of reduction to perform. 1986 void emitReduction(CodeGenFunction &CGF, SourceLocation Loc, 1987 ArrayRef<const Expr *> Privates, 1988 ArrayRef<const Expr *> LHSExprs, 1989 ArrayRef<const Expr *> RHSExprs, 1990 ArrayRef<const Expr *> ReductionOps, 1991 ReductionOptionsTy Options) override; 1992 1993 /// Emit a code for initialization of task reduction clause. Next code 1994 /// should be emitted for reduction: 1995 /// \code 1996 /// 1997 /// _task_red_item_t red_data[n]; 1998 /// ... 1999 /// red_data[i].shar = &origs[i]; 2000 /// red_data[i].size = sizeof(origs[i]); 2001 /// red_data[i].f_init = (void*)RedInit<i>; 2002 /// red_data[i].f_fini = (void*)RedDest<i>; 2003 /// red_data[i].f_comb = (void*)RedOp<i>; 2004 /// red_data[i].flags = <Flag_i>; 2005 /// ... 2006 /// void* tg1 = __kmpc_task_reduction_init(gtid, n, red_data); 2007 /// \endcode 2008 /// 2009 /// \param LHSExprs List of LHS in \a Data.ReductionOps reduction operations. 2010 /// \param RHSExprs List of RHS in \a Data.ReductionOps reduction operations. 2011 /// \param Data Additional data for task generation like tiedness, final 2012 /// state, list of privates, reductions etc. 2013 llvm::Value *emitTaskReductionInit(CodeGenFunction &CGF, SourceLocation Loc, 2014 ArrayRef<const Expr *> LHSExprs, 2015 ArrayRef<const Expr *> RHSExprs, 2016 const OMPTaskDataTy &Data) override; 2017 2018 /// Required to resolve existing problems in the runtime. Emits threadprivate 2019 /// variables to store the size of the VLAs/array sections for 2020 /// initializer/combiner/finalizer functions + emits threadprivate variable to 2021 /// store the pointer to the original reduction item for the custom 2022 /// initializer defined by declare reduction construct. 2023 /// \param RCG Allows to reuse an existing data for the reductions. 2024 /// \param N Reduction item for which fixups must be emitted. 2025 void emitTaskReductionFixups(CodeGenFunction &CGF, SourceLocation Loc, 2026 ReductionCodeGen &RCG, unsigned N) override; 2027 2028 /// Get the address of `void *` type of the privatue copy of the reduction 2029 /// item specified by the \p SharedLVal. 2030 /// \param ReductionsPtr Pointer to the reduction data returned by the 2031 /// emitTaskReductionInit function. 2032 /// \param SharedLVal Address of the original reduction item. 2033 Address getTaskReductionItem(CodeGenFunction &CGF, SourceLocation Loc, 2034 llvm::Value *ReductionsPtr, 2035 LValue SharedLVal) override; 2036 2037 /// Emit code for 'taskwait' directive. 2038 void emitTaskwaitCall(CodeGenFunction &CGF, SourceLocation Loc) override; 2039 2040 /// Emit code for 'cancellation point' construct. 2041 /// \param CancelRegion Region kind for which the cancellation point must be 2042 /// emitted. 2043 /// 2044 void emitCancellationPointCall(CodeGenFunction &CGF, SourceLocation Loc, 2045 OpenMPDirectiveKind CancelRegion) override; 2046 2047 /// Emit code for 'cancel' construct. 2048 /// \param IfCond Condition in the associated 'if' clause, if it was 2049 /// specified, nullptr otherwise. 2050 /// \param CancelRegion Region kind for which the cancel must be emitted. 2051 /// 2052 void emitCancelCall(CodeGenFunction &CGF, SourceLocation Loc, 2053 const Expr *IfCond, 2054 OpenMPDirectiveKind CancelRegion) override; 2055 2056 /// Emit outilined function for 'target' directive. 2057 /// \param D Directive to emit. 2058 /// \param ParentName Name of the function that encloses the target region. 2059 /// \param OutlinedFn Outlined function value to be defined by this call. 2060 /// \param OutlinedFnID Outlined function ID value to be defined by this call. 2061 /// \param IsOffloadEntry True if the outlined function is an offload entry. 2062 /// \param CodeGen Code generation sequence for the \a D directive. 2063 /// An outlined function may not be an entry if, e.g. the if clause always 2064 /// evaluates to false. 2065 void emitTargetOutlinedFunction(const OMPExecutableDirective &D, 2066 StringRef ParentName, 2067 llvm::Function *&OutlinedFn, 2068 llvm::Constant *&OutlinedFnID, 2069 bool IsOffloadEntry, 2070 const RegionCodeGenTy &CodeGen) override; 2071 2072 /// Emit the target offloading code associated with \a D. The emitted 2073 /// code attempts offloading the execution to the device, an the event of 2074 /// a failure it executes the host version outlined in \a OutlinedFn. 2075 /// \param D Directive to emit. 2076 /// \param OutlinedFn Host version of the code to be offloaded. 2077 /// \param OutlinedFnID ID of host version of the code to be offloaded. 2078 /// \param IfCond Expression evaluated in if clause associated with the target 2079 /// directive, or null if no if clause is used. 2080 /// \param Device Expression evaluated in device clause associated with the 2081 /// target directive, or null if no device clause is used. 2082 void emitTargetCall(CodeGenFunction &CGF, const OMPExecutableDirective &D, 2083 llvm::Function *OutlinedFn, llvm::Value *OutlinedFnID, 2084 const Expr *IfCond, const Expr *Device) override; 2085 2086 /// Emit the target regions enclosed in \a GD function definition or 2087 /// the function itself in case it is a valid device function. Returns true if 2088 /// \a GD was dealt with successfully. 2089 /// \param GD Function to scan. 2090 bool emitTargetFunctions(GlobalDecl GD) override; 2091 2092 /// Emit the global variable if it is a valid device global variable. 2093 /// Returns true if \a GD was dealt with successfully. 2094 /// \param GD Variable declaration to emit. 2095 bool emitTargetGlobalVariable(GlobalDecl GD) override; 2096 2097 /// Emit the global \a GD if it is meaningful for the target. Returns 2098 /// if it was emitted successfully. 2099 /// \param GD Global to scan. 2100 bool emitTargetGlobal(GlobalDecl GD) override; 2101 2102 /// Creates the offloading descriptor in the event any target region 2103 /// was emitted in the current module and return the function that registers 2104 /// it. 2105 llvm::Function *emitRegistrationFunction() override; 2106 2107 /// Emits code for teams call of the \a OutlinedFn with 2108 /// variables captured in a record which address is stored in \a 2109 /// CapturedStruct. 2110 /// \param OutlinedFn Outlined function to be run by team masters. Type of 2111 /// this function is void(*)(kmp_int32 *, kmp_int32, struct context_vars*). 2112 /// \param CapturedVars A pointer to the record with the references to 2113 /// variables used in \a OutlinedFn function. 2114 /// 2115 void emitTeamsCall(CodeGenFunction &CGF, const OMPExecutableDirective &D, 2116 SourceLocation Loc, llvm::Function *OutlinedFn, 2117 ArrayRef<llvm::Value *> CapturedVars) override; 2118 2119 /// Emits call to void __kmpc_push_num_teams(ident_t *loc, kmp_int32 2120 /// global_tid, kmp_int32 num_teams, kmp_int32 thread_limit) to generate code 2121 /// for num_teams clause. 2122 /// \param NumTeams An integer expression of teams. 2123 /// \param ThreadLimit An integer expression of threads. 2124 void emitNumTeamsClause(CodeGenFunction &CGF, const Expr *NumTeams, 2125 const Expr *ThreadLimit, SourceLocation Loc) override; 2126 2127 /// Emit the target data mapping code associated with \a D. 2128 /// \param D Directive to emit. 2129 /// \param IfCond Expression evaluated in if clause associated with the 2130 /// target directive, or null if no device clause is used. 2131 /// \param Device Expression evaluated in device clause associated with the 2132 /// target directive, or null if no device clause is used. 2133 /// \param Info A record used to store information that needs to be preserved 2134 /// until the region is closed. 2135 void emitTargetDataCalls(CodeGenFunction &CGF, 2136 const OMPExecutableDirective &D, const Expr *IfCond, 2137 const Expr *Device, const RegionCodeGenTy &CodeGen, 2138 TargetDataInfo &Info) override; 2139 2140 /// Emit the data mapping/movement code associated with the directive 2141 /// \a D that should be of the form 'target [{enter|exit} data | update]'. 2142 /// \param D Directive to emit. 2143 /// \param IfCond Expression evaluated in if clause associated with the target 2144 /// directive, or null if no if clause is used. 2145 /// \param Device Expression evaluated in device clause associated with the 2146 /// target directive, or null if no device clause is used. 2147 void emitTargetDataStandAloneCall(CodeGenFunction &CGF, 2148 const OMPExecutableDirective &D, 2149 const Expr *IfCond, 2150 const Expr *Device) override; 2151 2152 /// Emit initialization for doacross loop nesting support. 2153 /// \param D Loop-based construct used in doacross nesting construct. 2154 void emitDoacrossInit(CodeGenFunction &CGF, const OMPLoopDirective &D, 2155 ArrayRef<Expr *> NumIterations) override; 2156 2157 /// Emit code for doacross ordered directive with 'depend' clause. 2158 /// \param C 'depend' clause with 'sink|source' dependency kind. 2159 void emitDoacrossOrdered(CodeGenFunction &CGF, 2160 const OMPDependClause *C) override; 2161 2162 /// Translates the native parameter of outlined function if this is required 2163 /// for target. 2164 /// \param FD Field decl from captured record for the parameter. 2165 /// \param NativeParam Parameter itself. 2166 const VarDecl *translateParameter(const FieldDecl *FD, 2167 const VarDecl *NativeParam) const override; 2168 2169 /// Gets the address of the native argument basing on the address of the 2170 /// target-specific parameter. 2171 /// \param NativeParam Parameter itself. 2172 /// \param TargetParam Corresponding target-specific parameter. 2173 Address getParameterAddress(CodeGenFunction &CGF, const VarDecl *NativeParam, 2174 const VarDecl *TargetParam) const override; 2175 2176 /// Gets the OpenMP-specific address of the local variable. 2177 Address getAddressOfLocalVariable(CodeGenFunction &CGF, 2178 const VarDecl *VD) override { 2179 return Address::invalid(); 2180 } 2181 }; 2182 2183 } // namespace CodeGen 2184 } // namespace clang 2185 2186 #endif 2187