1 //===----- CGOpenMPRuntime.h - Interface to OpenMP Runtimes -----*- C++ -*-===// 2 // 3 // The LLVM Compiler Infrastructure 4 // 5 // This file is distributed under the University of Illinois Open Source 6 // License. See LICENSE.TXT for details. 7 // 8 //===----------------------------------------------------------------------===// 9 // 10 // This provides a class for OpenMP runtime code generation. 11 // 12 //===----------------------------------------------------------------------===// 13 14 #ifndef LLVM_CLANG_LIB_CODEGEN_CGOPENMPRUNTIME_H 15 #define LLVM_CLANG_LIB_CODEGEN_CGOPENMPRUNTIME_H 16 17 #include "clang/AST/Type.h" 18 #include "clang/Basic/OpenMPKinds.h" 19 #include "clang/Basic/SourceLocation.h" 20 #include "llvm/ADT/DenseMap.h" 21 #include "llvm/ADT/DenseSet.h" 22 #include "llvm/ADT/StringMap.h" 23 #include "llvm/IR/ValueHandle.h" 24 25 namespace llvm { 26 class ArrayType; 27 class Constant; 28 class Function; 29 class FunctionType; 30 class GlobalVariable; 31 class StructType; 32 class Type; 33 class Value; 34 } // namespace llvm 35 36 namespace clang { 37 class Expr; 38 class OMPExecutableDirective; 39 class VarDecl; 40 41 namespace CodeGen { 42 class Address; 43 class CodeGenFunction; 44 class CodeGenModule; 45 46 typedef llvm::function_ref<void(CodeGenFunction &)> RegionCodeGenTy; 47 48 class CGOpenMPRuntime { 49 private: 50 enum OpenMPRTLFunction { 51 /// \brief Call to void __kmpc_fork_call(ident_t *loc, kmp_int32 argc, 52 /// kmpc_micro microtask, ...); 53 OMPRTL__kmpc_fork_call, 54 /// \brief Call to void *__kmpc_threadprivate_cached(ident_t *loc, 55 /// kmp_int32 global_tid, void *data, size_t size, void ***cache); 56 OMPRTL__kmpc_threadprivate_cached, 57 /// \brief Call to void __kmpc_threadprivate_register( ident_t *, 58 /// void *data, kmpc_ctor ctor, kmpc_cctor cctor, kmpc_dtor dtor); 59 OMPRTL__kmpc_threadprivate_register, 60 // Call to __kmpc_int32 kmpc_global_thread_num(ident_t *loc); 61 OMPRTL__kmpc_global_thread_num, 62 // Call to void __kmpc_critical(ident_t *loc, kmp_int32 global_tid, 63 // kmp_critical_name *crit); 64 OMPRTL__kmpc_critical, 65 // Call to void __kmpc_critical_with_hint(ident_t *loc, kmp_int32 66 // global_tid, kmp_critical_name *crit, uintptr_t hint); 67 OMPRTL__kmpc_critical_with_hint, 68 // Call to void __kmpc_end_critical(ident_t *loc, kmp_int32 global_tid, 69 // kmp_critical_name *crit); 70 OMPRTL__kmpc_end_critical, 71 // Call to kmp_int32 __kmpc_cancel_barrier(ident_t *loc, kmp_int32 72 // global_tid); 73 OMPRTL__kmpc_cancel_barrier, 74 // Call to void __kmpc_barrier(ident_t *loc, kmp_int32 global_tid); 75 OMPRTL__kmpc_barrier, 76 // Call to void __kmpc_for_static_fini(ident_t *loc, kmp_int32 global_tid); 77 OMPRTL__kmpc_for_static_fini, 78 // Call to void __kmpc_serialized_parallel(ident_t *loc, kmp_int32 79 // global_tid); 80 OMPRTL__kmpc_serialized_parallel, 81 // Call to void __kmpc_end_serialized_parallel(ident_t *loc, kmp_int32 82 // global_tid); 83 OMPRTL__kmpc_end_serialized_parallel, 84 // Call to void __kmpc_push_num_threads(ident_t *loc, kmp_int32 global_tid, 85 // kmp_int32 num_threads); 86 OMPRTL__kmpc_push_num_threads, 87 // Call to void __kmpc_flush(ident_t *loc); 88 OMPRTL__kmpc_flush, 89 // Call to kmp_int32 __kmpc_master(ident_t *, kmp_int32 global_tid); 90 OMPRTL__kmpc_master, 91 // Call to void __kmpc_end_master(ident_t *, kmp_int32 global_tid); 92 OMPRTL__kmpc_end_master, 93 // Call to kmp_int32 __kmpc_omp_taskyield(ident_t *, kmp_int32 global_tid, 94 // int end_part); 95 OMPRTL__kmpc_omp_taskyield, 96 // Call to kmp_int32 __kmpc_single(ident_t *, kmp_int32 global_tid); 97 OMPRTL__kmpc_single, 98 // Call to void __kmpc_end_single(ident_t *, kmp_int32 global_tid); 99 OMPRTL__kmpc_end_single, 100 // Call to kmp_task_t * __kmpc_omp_task_alloc(ident_t *, kmp_int32 gtid, 101 // kmp_int32 flags, size_t sizeof_kmp_task_t, size_t sizeof_shareds, 102 // kmp_routine_entry_t *task_entry); 103 OMPRTL__kmpc_omp_task_alloc, 104 // Call to kmp_int32 __kmpc_omp_task(ident_t *, kmp_int32 gtid, kmp_task_t * 105 // new_task); 106 OMPRTL__kmpc_omp_task, 107 // Call to void __kmpc_copyprivate(ident_t *loc, kmp_int32 global_tid, 108 // size_t cpy_size, void *cpy_data, void(*cpy_func)(void *, void *), 109 // kmp_int32 didit); 110 OMPRTL__kmpc_copyprivate, 111 // Call to kmp_int32 __kmpc_reduce(ident_t *loc, kmp_int32 global_tid, 112 // kmp_int32 num_vars, size_t reduce_size, void *reduce_data, void 113 // (*reduce_func)(void *lhs_data, void *rhs_data), kmp_critical_name *lck); 114 OMPRTL__kmpc_reduce, 115 // Call to kmp_int32 __kmpc_reduce_nowait(ident_t *loc, kmp_int32 116 // global_tid, kmp_int32 num_vars, size_t reduce_size, void *reduce_data, 117 // void (*reduce_func)(void *lhs_data, void *rhs_data), kmp_critical_name 118 // *lck); 119 OMPRTL__kmpc_reduce_nowait, 120 // Call to void __kmpc_end_reduce(ident_t *loc, kmp_int32 global_tid, 121 // kmp_critical_name *lck); 122 OMPRTL__kmpc_end_reduce, 123 // Call to void __kmpc_end_reduce_nowait(ident_t *loc, kmp_int32 global_tid, 124 // kmp_critical_name *lck); 125 OMPRTL__kmpc_end_reduce_nowait, 126 // Call to void __kmpc_omp_task_begin_if0(ident_t *, kmp_int32 gtid, 127 // kmp_task_t * new_task); 128 OMPRTL__kmpc_omp_task_begin_if0, 129 // Call to void __kmpc_omp_task_complete_if0(ident_t *, kmp_int32 gtid, 130 // kmp_task_t * new_task); 131 OMPRTL__kmpc_omp_task_complete_if0, 132 // Call to void __kmpc_ordered(ident_t *loc, kmp_int32 global_tid); 133 OMPRTL__kmpc_ordered, 134 // Call to void __kmpc_end_ordered(ident_t *loc, kmp_int32 global_tid); 135 OMPRTL__kmpc_end_ordered, 136 // Call to kmp_int32 __kmpc_omp_taskwait(ident_t *loc, kmp_int32 137 // global_tid); 138 OMPRTL__kmpc_omp_taskwait, 139 // Call to void __kmpc_taskgroup(ident_t *loc, kmp_int32 global_tid); 140 OMPRTL__kmpc_taskgroup, 141 // Call to void __kmpc_end_taskgroup(ident_t *loc, kmp_int32 global_tid); 142 OMPRTL__kmpc_end_taskgroup, 143 // Call to void __kmpc_push_proc_bind(ident_t *loc, kmp_int32 global_tid, 144 // int proc_bind); 145 OMPRTL__kmpc_push_proc_bind, 146 // Call to kmp_int32 __kmpc_omp_task_with_deps(ident_t *loc_ref, kmp_int32 147 // gtid, kmp_task_t * new_task, kmp_int32 ndeps, kmp_depend_info_t 148 // *dep_list, kmp_int32 ndeps_noalias, kmp_depend_info_t *noalias_dep_list); 149 OMPRTL__kmpc_omp_task_with_deps, 150 // Call to void __kmpc_omp_wait_deps(ident_t *loc_ref, kmp_int32 151 // gtid, kmp_int32 ndeps, kmp_depend_info_t *dep_list, kmp_int32 152 // ndeps_noalias, kmp_depend_info_t *noalias_dep_list); 153 OMPRTL__kmpc_omp_wait_deps, 154 // Call to kmp_int32 __kmpc_cancellationpoint(ident_t *loc, kmp_int32 155 // global_tid, kmp_int32 cncl_kind); 156 OMPRTL__kmpc_cancellationpoint, 157 // Call to kmp_int32 __kmpc_cancel(ident_t *loc, kmp_int32 global_tid, 158 // kmp_int32 cncl_kind); 159 OMPRTL__kmpc_cancel, 160 161 // 162 // Offloading related calls 163 // 164 // Call to int32_t __tgt_target(int32_t device_id, void *host_ptr, int32_t 165 // arg_num, void** args_base, void **args, size_t *arg_sizes, int32_t 166 // *arg_types); 167 OMPRTL__tgt_target, 168 }; 169 170 /// \brief Values for bit flags used in the ident_t to describe the fields. 171 /// All enumeric elements are named and described in accordance with the code 172 /// from http://llvm.org/svn/llvm-project/openmp/trunk/runtime/src/kmp.h 173 enum OpenMPLocationFlags { 174 /// \brief Use trampoline for internal microtask. 175 OMP_IDENT_IMD = 0x01, 176 /// \brief Use c-style ident structure. 177 OMP_IDENT_KMPC = 0x02, 178 /// \brief Atomic reduction option for kmpc_reduce. 179 OMP_ATOMIC_REDUCE = 0x10, 180 /// \brief Explicit 'barrier' directive. 181 OMP_IDENT_BARRIER_EXPL = 0x20, 182 /// \brief Implicit barrier in code. 183 OMP_IDENT_BARRIER_IMPL = 0x40, 184 /// \brief Implicit barrier in 'for' directive. 185 OMP_IDENT_BARRIER_IMPL_FOR = 0x40, 186 /// \brief Implicit barrier in 'sections' directive. 187 OMP_IDENT_BARRIER_IMPL_SECTIONS = 0xC0, 188 /// \brief Implicit barrier in 'single' directive. 189 OMP_IDENT_BARRIER_IMPL_SINGLE = 0x140 190 }; 191 CodeGenModule &CGM; 192 /// \brief Default const ident_t object used for initialization of all other 193 /// ident_t objects. 194 llvm::Constant *DefaultOpenMPPSource; 195 /// \brief Map of flags and corresponding default locations. 196 typedef llvm::DenseMap<unsigned, llvm::Value *> OpenMPDefaultLocMapTy; 197 OpenMPDefaultLocMapTy OpenMPDefaultLocMap; 198 Address getOrCreateDefaultLocation(OpenMPLocationFlags Flags); 199 200 public: 201 /// \brief Describes ident structure that describes a source location. 202 /// All descriptions are taken from 203 /// http://llvm.org/svn/llvm-project/openmp/trunk/runtime/src/kmp.h 204 /// Original structure: 205 /// typedef struct ident { 206 /// kmp_int32 reserved_1; /**< might be used in Fortran; 207 /// see above */ 208 /// kmp_int32 flags; /**< also f.flags; KMP_IDENT_xxx flags; 209 /// KMP_IDENT_KMPC identifies this union 210 /// member */ 211 /// kmp_int32 reserved_2; /**< not really used in Fortran any more; 212 /// see above */ 213 ///#if USE_ITT_BUILD 214 /// /* but currently used for storing 215 /// region-specific ITT */ 216 /// /* contextual information. */ 217 ///#endif /* USE_ITT_BUILD */ 218 /// kmp_int32 reserved_3; /**< source[4] in Fortran, do not use for 219 /// C++ */ 220 /// char const *psource; /**< String describing the source location. 221 /// The string is composed of semi-colon separated 222 // fields which describe the source file, 223 /// the function and a pair of line numbers that 224 /// delimit the construct. 225 /// */ 226 /// } ident_t; 227 enum IdentFieldIndex { 228 /// \brief might be used in Fortran 229 IdentField_Reserved_1, 230 /// \brief OMP_IDENT_xxx flags; OMP_IDENT_KMPC identifies this union member. 231 IdentField_Flags, 232 /// \brief Not really used in Fortran any more 233 IdentField_Reserved_2, 234 /// \brief Source[4] in Fortran, do not use for C++ 235 IdentField_Reserved_3, 236 /// \brief String describing the source location. The string is composed of 237 /// semi-colon separated fields which describe the source file, the function 238 /// and a pair of line numbers that delimit the construct. 239 IdentField_PSource 240 }; 241 private: 242 llvm::StructType *IdentTy; 243 /// \brief Map for SourceLocation and OpenMP runtime library debug locations. 244 typedef llvm::DenseMap<unsigned, llvm::Value *> OpenMPDebugLocMapTy; 245 OpenMPDebugLocMapTy OpenMPDebugLocMap; 246 /// \brief The type for a microtask which gets passed to __kmpc_fork_call(). 247 /// Original representation is: 248 /// typedef void (kmpc_micro)(kmp_int32 global_tid, kmp_int32 bound_tid,...); 249 llvm::FunctionType *Kmpc_MicroTy; 250 /// \brief Stores debug location and ThreadID for the function. 251 struct DebugLocThreadIdTy { 252 llvm::Value *DebugLoc; 253 llvm::Value *ThreadID; 254 }; 255 /// \brief Map of local debug location, ThreadId and functions. 256 typedef llvm::DenseMap<llvm::Function *, DebugLocThreadIdTy> 257 OpenMPLocThreadIDMapTy; 258 OpenMPLocThreadIDMapTy OpenMPLocThreadIDMap; 259 /// \brief Type kmp_critical_name, originally defined as typedef kmp_int32 260 /// kmp_critical_name[8]; 261 llvm::ArrayType *KmpCriticalNameTy; 262 /// \brief An ordered map of auto-generated variables to their unique names. 263 /// It stores variables with the following names: 1) ".gomp_critical_user_" + 264 /// <critical_section_name> + ".var" for "omp critical" directives; 2) 265 /// <mangled_name_for_global_var> + ".cache." for cache for threadprivate 266 /// variables. 267 llvm::StringMap<llvm::AssertingVH<llvm::Constant>, llvm::BumpPtrAllocator> 268 InternalVars; 269 /// \brief Type typedef kmp_int32 (* kmp_routine_entry_t)(kmp_int32, void *); 270 llvm::Type *KmpRoutineEntryPtrTy; 271 QualType KmpRoutineEntryPtrQTy; 272 /// \brief Type typedef struct kmp_task { 273 /// void * shareds; /**< pointer to block of pointers to 274 /// shared vars */ 275 /// kmp_routine_entry_t routine; /**< pointer to routine to call for 276 /// executing task */ 277 /// kmp_int32 part_id; /**< part id for the task */ 278 /// kmp_routine_entry_t destructors; /* pointer to function to invoke 279 /// deconstructors of firstprivate C++ objects */ 280 /// } kmp_task_t; 281 QualType KmpTaskTQTy; 282 /// \brief Type typedef struct kmp_depend_info { 283 /// kmp_intptr_t base_addr; 284 /// size_t len; 285 /// struct { 286 /// bool in:1; 287 /// bool out:1; 288 /// } flags; 289 /// } kmp_depend_info_t; 290 QualType KmpDependInfoTy; 291 292 293 /// \brief Build type kmp_routine_entry_t (if not built yet). 294 void emitKmpRoutineEntryT(QualType KmpInt32Ty); 295 296 /// \brief Emits object of ident_t type with info for source location. 297 /// \param Flags Flags for OpenMP location. 298 /// 299 llvm::Value *emitUpdateLocation(CodeGenFunction &CGF, SourceLocation Loc, 300 OpenMPLocationFlags Flags = OMP_IDENT_KMPC); 301 302 /// \brief Returns pointer to ident_t type. 303 llvm::Type *getIdentTyPointerTy(); 304 305 /// \brief Returns pointer to kmpc_micro type. 306 llvm::Type *getKmpc_MicroPointerTy(); 307 308 /// \brief Returns specified OpenMP runtime function. 309 /// \param Function OpenMP runtime function. 310 /// \return Specified function. 311 llvm::Constant *createRuntimeFunction(OpenMPRTLFunction Function); 312 313 /// \brief Returns __kmpc_for_static_init_* runtime function for the specified 314 /// size \a IVSize and sign \a IVSigned. 315 llvm::Constant *createForStaticInitFunction(unsigned IVSize, bool IVSigned); 316 317 /// \brief Returns __kmpc_dispatch_init_* runtime function for the specified 318 /// size \a IVSize and sign \a IVSigned. 319 llvm::Constant *createDispatchInitFunction(unsigned IVSize, bool IVSigned); 320 321 /// \brief Returns __kmpc_dispatch_next_* runtime function for the specified 322 /// size \a IVSize and sign \a IVSigned. 323 llvm::Constant *createDispatchNextFunction(unsigned IVSize, bool IVSigned); 324 325 /// \brief Returns __kmpc_dispatch_fini_* runtime function for the specified 326 /// size \a IVSize and sign \a IVSigned. 327 llvm::Constant *createDispatchFiniFunction(unsigned IVSize, bool IVSigned); 328 329 /// \brief If the specified mangled name is not in the module, create and 330 /// return threadprivate cache object. This object is a pointer's worth of 331 /// storage that's reserved for use by the OpenMP runtime. 332 /// \param VD Threadprivate variable. 333 /// \return Cache variable for the specified threadprivate. 334 llvm::Constant *getOrCreateThreadPrivateCache(const VarDecl *VD); 335 336 /// \brief Emits address of the word in a memory where current thread id is 337 /// stored. 338 virtual Address emitThreadIDAddress(CodeGenFunction &CGF, SourceLocation Loc); 339 340 /// \brief Gets thread id value for the current thread. 341 /// 342 llvm::Value *getThreadID(CodeGenFunction &CGF, SourceLocation Loc); 343 344 /// \brief Gets (if variable with the given name already exist) or creates 345 /// internal global variable with the specified Name. The created variable has 346 /// linkage CommonLinkage by default and is initialized by null value. 347 /// \param Ty Type of the global variable. If it is exist already the type 348 /// must be the same. 349 /// \param Name Name of the variable. 350 llvm::Constant *getOrCreateInternalVariable(llvm::Type *Ty, 351 const llvm::Twine &Name); 352 353 /// \brief Set of threadprivate variables with the generated initializer. 354 llvm::DenseSet<const VarDecl *> ThreadPrivateWithDefinition; 355 356 /// \brief Emits initialization code for the threadprivate variables. 357 /// \param VDAddr Address of the global variable \a VD. 358 /// \param Ctor Pointer to a global init function for \a VD. 359 /// \param CopyCtor Pointer to a global copy function for \a VD. 360 /// \param Dtor Pointer to a global destructor function for \a VD. 361 /// \param Loc Location of threadprivate declaration. 362 void emitThreadPrivateVarInit(CodeGenFunction &CGF, Address VDAddr, 363 llvm::Value *Ctor, llvm::Value *CopyCtor, 364 llvm::Value *Dtor, SourceLocation Loc); 365 366 /// \brief Returns corresponding lock object for the specified critical region 367 /// name. If the lock object does not exist it is created, otherwise the 368 /// reference to the existing copy is returned. 369 /// \param CriticalName Name of the critical region. 370 /// 371 llvm::Value *getCriticalRegionLock(StringRef CriticalName); 372 373 public: 374 explicit CGOpenMPRuntime(CodeGenModule &CGM); 375 virtual ~CGOpenMPRuntime() {} 376 virtual void clear(); 377 378 /// \brief Emits outlined function for the specified OpenMP parallel directive 379 /// \a D. This outlined function has type void(*)(kmp_int32 *ThreadID, 380 /// kmp_int32 BoundID, struct context_vars*). 381 /// \param D OpenMP directive. 382 /// \param ThreadIDVar Variable for thread id in the current OpenMP region. 383 /// \param InnermostKind Kind of innermost directive (for simple directives it 384 /// is a directive itself, for combined - its innermost directive). 385 /// \param CodeGen Code generation sequence for the \a D directive. 386 virtual llvm::Value *emitParallelOutlinedFunction( 387 const OMPExecutableDirective &D, const VarDecl *ThreadIDVar, 388 OpenMPDirectiveKind InnermostKind, const RegionCodeGenTy &CodeGen); 389 390 /// \brief Emits outlined function for the OpenMP task directive \a D. This 391 /// outlined function has type void(*)(kmp_int32 ThreadID, kmp_int32 392 /// PartID, struct context_vars*). 393 /// \param D OpenMP directive. 394 /// \param ThreadIDVar Variable for thread id in the current OpenMP region. 395 /// \param InnermostKind Kind of innermost directive (for simple directives it 396 /// is a directive itself, for combined - its innermost directive). 397 /// \param CodeGen Code generation sequence for the \a D directive. 398 /// 399 virtual llvm::Value *emitTaskOutlinedFunction( 400 const OMPExecutableDirective &D, const VarDecl *ThreadIDVar, 401 OpenMPDirectiveKind InnermostKind, const RegionCodeGenTy &CodeGen); 402 403 /// \brief Cleans up references to the objects in finished function. 404 /// 405 void functionFinished(CodeGenFunction &CGF); 406 407 /// \brief Emits code for parallel or serial call of the \a OutlinedFn with 408 /// variables captured in a record which address is stored in \a 409 /// CapturedStruct. 410 /// \param OutlinedFn Outlined function to be run in parallel threads. Type of 411 /// this function is void(*)(kmp_int32 *, kmp_int32, struct context_vars*). 412 /// \param CapturedVars A pointer to the record with the references to 413 /// variables used in \a OutlinedFn function. 414 /// \param IfCond Condition in the associated 'if' clause, if it was 415 /// specified, nullptr otherwise. 416 /// 417 virtual void emitParallelCall(CodeGenFunction &CGF, SourceLocation Loc, 418 llvm::Value *OutlinedFn, 419 ArrayRef<llvm::Value *> CapturedVars, 420 const Expr *IfCond); 421 422 /// \brief Emits a critical region. 423 /// \param CriticalName Name of the critical region. 424 /// \param CriticalOpGen Generator for the statement associated with the given 425 /// critical region. 426 /// \param Hint Value of the 'hint' clause (optional). 427 virtual void emitCriticalRegion(CodeGenFunction &CGF, StringRef CriticalName, 428 const RegionCodeGenTy &CriticalOpGen, 429 SourceLocation Loc, 430 const Expr *Hint = nullptr); 431 432 /// \brief Emits a master region. 433 /// \param MasterOpGen Generator for the statement associated with the given 434 /// master region. 435 virtual void emitMasterRegion(CodeGenFunction &CGF, 436 const RegionCodeGenTy &MasterOpGen, 437 SourceLocation Loc); 438 439 /// \brief Emits code for a taskyield directive. 440 virtual void emitTaskyieldCall(CodeGenFunction &CGF, SourceLocation Loc); 441 442 /// \brief Emit a taskgroup region. 443 /// \param TaskgroupOpGen Generator for the statement associated with the 444 /// given taskgroup region. 445 virtual void emitTaskgroupRegion(CodeGenFunction &CGF, 446 const RegionCodeGenTy &TaskgroupOpGen, 447 SourceLocation Loc); 448 449 /// \brief Emits a single region. 450 /// \param SingleOpGen Generator for the statement associated with the given 451 /// single region. 452 virtual void emitSingleRegion(CodeGenFunction &CGF, 453 const RegionCodeGenTy &SingleOpGen, 454 SourceLocation Loc, 455 ArrayRef<const Expr *> CopyprivateVars, 456 ArrayRef<const Expr *> DestExprs, 457 ArrayRef<const Expr *> SrcExprs, 458 ArrayRef<const Expr *> AssignmentOps); 459 460 /// \brief Emit an ordered region. 461 /// \param OrderedOpGen Generator for the statement associated with the given 462 /// ordered region. 463 virtual void emitOrderedRegion(CodeGenFunction &CGF, 464 const RegionCodeGenTy &OrderedOpGen, 465 SourceLocation Loc, bool IsThreads); 466 467 /// \brief Emit an implicit/explicit barrier for OpenMP threads. 468 /// \param Kind Directive for which this implicit barrier call must be 469 /// generated. Must be OMPD_barrier for explicit barrier generation. 470 /// \param EmitChecks true if need to emit checks for cancellation barriers. 471 /// \param ForceSimpleCall true simple barrier call must be emitted, false if 472 /// runtime class decides which one to emit (simple or with cancellation 473 /// checks). 474 /// 475 virtual void emitBarrierCall(CodeGenFunction &CGF, SourceLocation Loc, 476 OpenMPDirectiveKind Kind, 477 bool EmitChecks = true, 478 bool ForceSimpleCall = false); 479 480 /// \brief Check if the specified \a ScheduleKind is static non-chunked. 481 /// This kind of worksharing directive is emitted without outer loop. 482 /// \param ScheduleKind Schedule kind specified in the 'schedule' clause. 483 /// \param Chunked True if chunk is specified in the clause. 484 /// 485 virtual bool isStaticNonchunked(OpenMPScheduleClauseKind ScheduleKind, 486 bool Chunked) const; 487 488 /// \brief Check if the specified \a ScheduleKind is dynamic. 489 /// This kind of worksharing directive is emitted without outer loop. 490 /// \param ScheduleKind Schedule Kind specified in the 'schedule' clause. 491 /// 492 virtual bool isDynamic(OpenMPScheduleClauseKind ScheduleKind) const; 493 494 virtual void emitForDispatchInit(CodeGenFunction &CGF, SourceLocation Loc, 495 OpenMPScheduleClauseKind SchedKind, 496 unsigned IVSize, bool IVSigned, 497 bool Ordered, llvm::Value *UB, 498 llvm::Value *Chunk = nullptr); 499 500 /// \brief Call the appropriate runtime routine to initialize it before start 501 /// of loop. 502 /// 503 /// Depending on the loop schedule, it is nesessary to call some runtime 504 /// routine before start of the OpenMP loop to get the loop upper / lower 505 /// bounds \a LB and \a UB and stride \a ST. 506 /// 507 /// \param CGF Reference to current CodeGenFunction. 508 /// \param Loc Clang source location. 509 /// \param SchedKind Schedule kind, specified by the 'schedule' clause. 510 /// \param IVSize Size of the iteration variable in bits. 511 /// \param IVSigned Sign of the interation variable. 512 /// \param Ordered true if loop is ordered, false otherwise. 513 /// \param IL Address of the output variable in which the flag of the 514 /// last iteration is returned. 515 /// \param LB Address of the output variable in which the lower iteration 516 /// number is returned. 517 /// \param UB Address of the output variable in which the upper iteration 518 /// number is returned. 519 /// \param ST Address of the output variable in which the stride value is 520 /// returned nesessary to generated the static_chunked scheduled loop. 521 /// \param Chunk Value of the chunk for the static_chunked scheduled loop. 522 /// For the default (nullptr) value, the chunk 1 will be used. 523 /// 524 virtual void emitForStaticInit(CodeGenFunction &CGF, SourceLocation Loc, 525 OpenMPScheduleClauseKind SchedKind, 526 unsigned IVSize, bool IVSigned, bool Ordered, 527 Address IL, Address LB, 528 Address UB, Address ST, 529 llvm::Value *Chunk = nullptr); 530 531 /// \brief Call the appropriate runtime routine to notify that we finished 532 /// iteration of the ordered loop with the dynamic scheduling. 533 /// 534 /// \param CGF Reference to current CodeGenFunction. 535 /// \param Loc Clang source location. 536 /// \param IVSize Size of the iteration variable in bits. 537 /// \param IVSigned Sign of the interation variable. 538 /// 539 virtual void emitForOrderedIterationEnd(CodeGenFunction &CGF, 540 SourceLocation Loc, unsigned IVSize, 541 bool IVSigned); 542 543 /// \brief Call the appropriate runtime routine to notify that we finished 544 /// all the work with current loop. 545 /// 546 /// \param CGF Reference to current CodeGenFunction. 547 /// \param Loc Clang source location. 548 /// 549 virtual void emitForStaticFinish(CodeGenFunction &CGF, SourceLocation Loc); 550 551 /// Call __kmpc_dispatch_next( 552 /// ident_t *loc, kmp_int32 tid, kmp_int32 *p_lastiter, 553 /// kmp_int[32|64] *p_lower, kmp_int[32|64] *p_upper, 554 /// kmp_int[32|64] *p_stride); 555 /// \param IVSize Size of the iteration variable in bits. 556 /// \param IVSigned Sign of the interation variable. 557 /// \param IL Address of the output variable in which the flag of the 558 /// last iteration is returned. 559 /// \param LB Address of the output variable in which the lower iteration 560 /// number is returned. 561 /// \param UB Address of the output variable in which the upper iteration 562 /// number is returned. 563 /// \param ST Address of the output variable in which the stride value is 564 /// returned. 565 virtual llvm::Value *emitForNext(CodeGenFunction &CGF, SourceLocation Loc, 566 unsigned IVSize, bool IVSigned, 567 Address IL, Address LB, 568 Address UB, Address ST); 569 570 /// \brief Emits call to void __kmpc_push_num_threads(ident_t *loc, kmp_int32 571 /// global_tid, kmp_int32 num_threads) to generate code for 'num_threads' 572 /// clause. 573 /// \param NumThreads An integer value of threads. 574 virtual void emitNumThreadsClause(CodeGenFunction &CGF, 575 llvm::Value *NumThreads, 576 SourceLocation Loc); 577 578 /// \brief Emit call to void __kmpc_push_proc_bind(ident_t *loc, kmp_int32 579 /// global_tid, int proc_bind) to generate code for 'proc_bind' clause. 580 virtual void emitProcBindClause(CodeGenFunction &CGF, 581 OpenMPProcBindClauseKind ProcBind, 582 SourceLocation Loc); 583 584 /// \brief Returns address of the threadprivate variable for the current 585 /// thread. 586 /// \param VD Threadprivate variable. 587 /// \param VDAddr Address of the global variable \a VD. 588 /// \param Loc Location of the reference to threadprivate var. 589 /// \return Address of the threadprivate variable for the current thread. 590 virtual Address getAddrOfThreadPrivate(CodeGenFunction &CGF, 591 const VarDecl *VD, 592 Address VDAddr, 593 SourceLocation Loc); 594 595 /// \brief Emit a code for initialization of threadprivate variable. It emits 596 /// a call to runtime library which adds initial value to the newly created 597 /// threadprivate variable (if it is not constant) and registers destructor 598 /// for the variable (if any). 599 /// \param VD Threadprivate variable. 600 /// \param VDAddr Address of the global variable \a VD. 601 /// \param Loc Location of threadprivate declaration. 602 /// \param PerformInit true if initialization expression is not constant. 603 virtual llvm::Function * 604 emitThreadPrivateVarDefinition(const VarDecl *VD, Address VDAddr, 605 SourceLocation Loc, bool PerformInit, 606 CodeGenFunction *CGF = nullptr); 607 608 /// \brief Emit flush of the variables specified in 'omp flush' directive. 609 /// \param Vars List of variables to flush. 610 virtual void emitFlush(CodeGenFunction &CGF, ArrayRef<const Expr *> Vars, 611 SourceLocation Loc); 612 613 /// \brief Emit task region for the task directive. The task region is 614 /// emitted in several steps: 615 /// 1. Emit a call to kmp_task_t *__kmpc_omp_task_alloc(ident_t *, kmp_int32 616 /// gtid, kmp_int32 flags, size_t sizeof_kmp_task_t, size_t sizeof_shareds, 617 /// kmp_routine_entry_t *task_entry). Here task_entry is a pointer to the 618 /// function: 619 /// kmp_int32 .omp_task_entry.(kmp_int32 gtid, kmp_task_t *tt) { 620 /// TaskFunction(gtid, tt->part_id, tt->shareds); 621 /// return 0; 622 /// } 623 /// 2. Copy a list of shared variables to field shareds of the resulting 624 /// structure kmp_task_t returned by the previous call (if any). 625 /// 3. Copy a pointer to destructions function to field destructions of the 626 /// resulting structure kmp_task_t. 627 /// 4. Emit a call to kmp_int32 __kmpc_omp_task(ident_t *, kmp_int32 gtid, 628 /// kmp_task_t *new_task), where new_task is a resulting structure from 629 /// previous items. 630 /// \param D Current task directive. 631 /// \param Tied true if the task is tied (the task is tied to the thread that 632 /// can suspend its task region), false - untied (the task is not tied to any 633 /// thread). 634 /// \param Final Contains either constant bool value, or llvm::Value * of i1 635 /// type for final clause. If the value is true, the task forces all of its 636 /// child tasks to become final and included tasks. 637 /// \param TaskFunction An LLVM function with type void (*)(i32 /*gtid*/, i32 638 /// /*part_id*/, captured_struct */*__context*/); 639 /// \param SharedsTy A type which contains references the shared variables. 640 /// \param Shareds Context with the list of shared variables from the \p 641 /// TaskFunction. 642 /// \param IfCond Not a nullptr if 'if' clause was specified, nullptr 643 /// otherwise. 644 /// \param PrivateVars List of references to private variables for the task 645 /// directive. 646 /// \param PrivateCopies List of private copies for each private variable in 647 /// \p PrivateVars. 648 /// \param FirstprivateVars List of references to private variables for the 649 /// task directive. 650 /// \param FirstprivateCopies List of private copies for each private variable 651 /// in \p FirstprivateVars. 652 /// \param FirstprivateInits List of references to auto generated variables 653 /// used for initialization of a single array element. Used if firstprivate 654 /// variable is of array type. 655 /// \param Dependences List of dependences for the 'task' construct, including 656 /// original expression and dependency type. 657 virtual void emitTaskCall( 658 CodeGenFunction &CGF, SourceLocation Loc, const OMPExecutableDirective &D, 659 bool Tied, llvm::PointerIntPair<llvm::Value *, 1, bool> Final, 660 llvm::Value *TaskFunction, QualType SharedsTy, Address Shareds, 661 const Expr *IfCond, ArrayRef<const Expr *> PrivateVars, 662 ArrayRef<const Expr *> PrivateCopies, 663 ArrayRef<const Expr *> FirstprivateVars, 664 ArrayRef<const Expr *> FirstprivateCopies, 665 ArrayRef<const Expr *> FirstprivateInits, 666 ArrayRef<std::pair<OpenMPDependClauseKind, const Expr *>> Dependences); 667 668 /// \brief Emit code for the directive that does not require outlining. 669 /// 670 /// \param InnermostKind Kind of innermost directive (for simple directives it 671 /// is a directive itself, for combined - its innermost directive). 672 /// \param CodeGen Code generation sequence for the \a D directive. 673 /// \param HasCancel true if region has inner cancel directive, false 674 /// otherwise. 675 virtual void emitInlinedDirective(CodeGenFunction &CGF, 676 OpenMPDirectiveKind InnermostKind, 677 const RegionCodeGenTy &CodeGen, 678 bool HasCancel = false); 679 /// \brief Emit a code for reduction clause. Next code should be emitted for 680 /// reduction: 681 /// \code 682 /// 683 /// static kmp_critical_name lock = { 0 }; 684 /// 685 /// void reduce_func(void *lhs[<n>], void *rhs[<n>]) { 686 /// ... 687 /// *(Type<i>*)lhs[i] = RedOp<i>(*(Type<i>*)lhs[i], *(Type<i>*)rhs[i]); 688 /// ... 689 /// } 690 /// 691 /// ... 692 /// void *RedList[<n>] = {&<RHSExprs>[0], ..., &<RHSExprs>[<n>-1]}; 693 /// switch (__kmpc_reduce{_nowait}(<loc>, <gtid>, <n>, sizeof(RedList), 694 /// RedList, reduce_func, &<lock>)) { 695 /// case 1: 696 /// ... 697 /// <LHSExprs>[i] = RedOp<i>(*<LHSExprs>[i], *<RHSExprs>[i]); 698 /// ... 699 /// __kmpc_end_reduce{_nowait}(<loc>, <gtid>, &<lock>); 700 /// break; 701 /// case 2: 702 /// ... 703 /// Atomic(<LHSExprs>[i] = RedOp<i>(*<LHSExprs>[i], *<RHSExprs>[i])); 704 /// ... 705 /// break; 706 /// default:; 707 /// } 708 /// \endcode 709 /// 710 /// \param Privates List of private copies for original reduction arguments. 711 /// \param LHSExprs List of LHS in \a ReductionOps reduction operations. 712 /// \param RHSExprs List of RHS in \a ReductionOps reduction operations. 713 /// \param ReductionOps List of reduction operations in form 'LHS binop RHS' 714 /// or 'operator binop(LHS, RHS)'. 715 /// \param WithNowait true if parent directive has also nowait clause, false 716 /// otherwise. 717 virtual void emitReduction(CodeGenFunction &CGF, SourceLocation Loc, 718 ArrayRef<const Expr *> Privates, 719 ArrayRef<const Expr *> LHSExprs, 720 ArrayRef<const Expr *> RHSExprs, 721 ArrayRef<const Expr *> ReductionOps, 722 bool WithNowait, bool SimpleReduction); 723 724 /// \brief Emit code for 'taskwait' directive. 725 virtual void emitTaskwaitCall(CodeGenFunction &CGF, SourceLocation Loc); 726 727 /// \brief Emit code for 'cancellation point' construct. 728 /// \param CancelRegion Region kind for which the cancellation point must be 729 /// emitted. 730 /// 731 virtual void emitCancellationPointCall(CodeGenFunction &CGF, 732 SourceLocation Loc, 733 OpenMPDirectiveKind CancelRegion); 734 735 /// \brief Emit code for 'cancel' construct. 736 /// \param IfCond Condition in the associated 'if' clause, if it was 737 /// specified, nullptr otherwise. 738 /// \param CancelRegion Region kind for which the cancel must be emitted. 739 /// 740 virtual void emitCancelCall(CodeGenFunction &CGF, SourceLocation Loc, 741 const Expr *IfCond, 742 OpenMPDirectiveKind CancelRegion); 743 744 /// \brief Emit outilined function for 'target' directive. 745 /// \param D Directive to emit. 746 /// \param CodeGen Code generation sequence for the \a D directive. 747 virtual llvm::Value * 748 emitTargetOutlinedFunction(const OMPExecutableDirective &D, 749 const RegionCodeGenTy &CodeGen); 750 751 /// \brief Emit the target offloading code associated with \a D. The emitted 752 /// code attempts offloading the execution to the device, an the event of 753 /// a failure it executes the host version outlined in \a OutlinedFn. 754 /// \param D Directive to emit. 755 /// \param OutlinedFn Host version of the code to be offloaded. 756 /// \param IfCond Expression evaluated in if clause associated with the target 757 /// directive, or null if no if clause is used. 758 /// \param Device Expression evaluated in device clause associated with the 759 /// target directive, or null if no device clause is used. 760 /// \param CapturedVars Values captured in the current region. 761 virtual void emitTargetCall(CodeGenFunction &CGF, 762 const OMPExecutableDirective &D, 763 llvm::Value *OutlinedFn, const Expr *IfCond, 764 const Expr *Device, 765 ArrayRef<llvm::Value *> CapturedVars); 766 }; 767 768 } // namespace CodeGen 769 } // namespace clang 770 771 #endif 772
