CGOpenMPRuntime.cpp

Go to the documentation of this file.

//===----- CGOpenMPRuntime.cpp - Interface to OpenMP Runtimes -------------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
//
// This provides a class for OpenMP runtime code generation.
//
//===----------------------------------------------------------------------===//

#include "CGCXXABI.h"
#include "CGCleanup.h"
#include "CGOpenMPRuntime.h"
#include "CGRecordLayout.h"
#include "CodeGenFunction.h"
#include "clang/CodeGen/ConstantInitBuilder.h"
#include "clang/AST/Decl.h"
#include "clang/AST/StmtOpenMP.h"
#include "clang/Basic/BitmaskEnum.h"
#include "llvm/ADT/ArrayRef.h"
#include "llvm/Bitcode/BitcodeReader.h"
#include "llvm/IR/DerivedTypes.h"
#include "llvm/IR/GlobalValue.h"
#include "llvm/IR/Value.h"
#include "llvm/Support/Format.h"
#include "llvm/Support/raw_ostream.h"
#include <cassert>

using namespace clang;
using namespace CodeGen;

namespace {
/// Base class for handling code generation inside OpenMP regions.
class CGOpenMPRegionInfo : public CodeGenFunction::CGCapturedStmtInfo {
public:
  /// Kinds of OpenMP regions used in codegen.
  enum CGOpenMPRegionKind {
    /// Region with outlined function for standalone 'parallel'
    /// directive.
    ParallelOutlinedRegion,
    /// Region with outlined function for standalone 'task' directive.
    TaskOutlinedRegion,
    /// Region for constructs that do not require function outlining,
    /// like 'for', 'sections', 'atomic' etc. directives.
    InlinedRegion,
    /// Region with outlined function for standalone 'target' directive.
    TargetRegion,
  };

  CGOpenMPRegionInfo(const CapturedStmt &CS,
                     const CGOpenMPRegionKind RegionKind,
                     const RegionCodeGenTy &CodeGen, OpenMPDirectiveKind Kind,
                     bool HasCancel)
      : CGCapturedStmtInfo(CS, CR_OpenMP), RegionKind(RegionKind),
        CodeGen(CodeGen), Kind(Kind), HasCancel(HasCancel) {}

  CGOpenMPRegionInfo(const CGOpenMPRegionKind RegionKind,
                     const RegionCodeGenTy &CodeGen, OpenMPDirectiveKind Kind,
                     bool HasCancel)
      : CGCapturedStmtInfo(CR_OpenMP), RegionKind(RegionKind), CodeGen(CodeGen),
        Kind(Kind), HasCancel(HasCancel) {}

  /// Get a variable or parameter for storing global thread id
  /// inside OpenMP construct.
  virtual const VarDecl *getThreadIDVariable() const = 0;

  /// Emit the captured statement body.
  void EmitBody(CodeGenFunction &CGF, const Stmt *S) override;

  /// Get an LValue for the current ThreadID variable.
  /// \return LValue for thread id variable. This LValue always has type int32*.
  virtual LValue getThreadIDVariableLValue(CodeGenFunction &CGF);

  virtual void emitUntiedSwitch(CodeGenFunction & /*CGF*/) {}

  CGOpenMPRegionKind getRegionKind() const { return RegionKind; }

  OpenMPDirectiveKind getDirectiveKind() const { return Kind; }

  bool hasCancel() const { return HasCancel; }

  static bool classof(const CGCapturedStmtInfo *Info) {
    return Info->getKind() == CR_OpenMP;
  }

  ~CGOpenMPRegionInfo() override = default;

protected:
  CGOpenMPRegionKind RegionKind;
  RegionCodeGenTy CodeGen;
  OpenMPDirectiveKind Kind;
  bool HasCancel;
};

/// API for captured statement code generation in OpenMP constructs.
class CGOpenMPOutlinedRegionInfo final : public CGOpenMPRegionInfo {
public:
  CGOpenMPOutlinedRegionInfo(const CapturedStmt &CS, const VarDecl *ThreadIDVar,
                             const RegionCodeGenTy &CodeGen,
                             OpenMPDirectiveKind Kind, bool HasCancel,
                             StringRef HelperName)
      : CGOpenMPRegionInfo(CS, ParallelOutlinedRegion, CodeGen, Kind,
                           HasCancel),
        ThreadIDVar(ThreadIDVar), HelperName(HelperName) {
    assert(ThreadIDVar != nullptr && "No ThreadID in OpenMP region.");
  }

  /// Get a variable or parameter for storing global thread id
  /// inside OpenMP construct.
  const VarDecl *getThreadIDVariable() const override { return ThreadIDVar; }

  /// Get the name of the capture helper.
  StringRef getHelperName() const override { return HelperName; }

  static bool classof(const CGCapturedStmtInfo *Info) {
    return CGOpenMPRegionInfo::classof(Info) &&
           cast<CGOpenMPRegionInfo>(Info)->getRegionKind() ==
               ParallelOutlinedRegion;
  }

private:
  /// A variable or parameter storing global thread id for OpenMP
  /// constructs.
  const VarDecl *ThreadIDVar;
  StringRef HelperName;
};

/// API for captured statement code generation in OpenMP constructs.
class CGOpenMPTaskOutlinedRegionInfo final : public CGOpenMPRegionInfo {
public:
  class UntiedTaskActionTy final : public PrePostActionTy {
    bool Untied;
    const VarDecl *PartIDVar;
    const RegionCodeGenTy UntiedCodeGen;
    llvm::SwitchInst *UntiedSwitch = nullptr;

  public:
    UntiedTaskActionTy(bool Tied, const VarDecl *PartIDVar,
                       const RegionCodeGenTy &UntiedCodeGen)
        : Untied(!Tied), PartIDVar(PartIDVar), UntiedCodeGen(UntiedCodeGen) {}
    void Enter(CodeGenFunction &CGF) override {
      if (Untied) {
        // Emit task switching point.
        LValue PartIdLVal = CGF.EmitLoadOfPointerLValue(
            CGF.GetAddrOfLocalVar(PartIDVar),
            PartIDVar->getType()->castAs<PointerType>());
        llvm::Value *Res =
            CGF.EmitLoadOfScalar(PartIdLVal, PartIDVar->getLocation());
        llvm::BasicBlock *DoneBB = CGF.createBasicBlock(".untied.done.");
        UntiedSwitch = CGF.Builder.CreateSwitch(Res, DoneBB);
        CGF.EmitBlock(DoneBB);
        CGF.EmitBranchThroughCleanup(CGF.ReturnBlock);
        CGF.EmitBlock(CGF.createBasicBlock(".untied.jmp."));
        UntiedSwitch->addCase(CGF.Builder.getInt32(0),
                              CGF.Builder.GetInsertBlock());
        emitUntiedSwitch(CGF);
      }
    }
    void emitUntiedSwitch(CodeGenFunction &CGF) const {
      if (Untied) {
        LValue PartIdLVal = CGF.EmitLoadOfPointerLValue(
            CGF.GetAddrOfLocalVar(PartIDVar),
            PartIDVar->getType()->castAs<PointerType>());
        CGF.EmitStoreOfScalar(CGF.Builder.getInt32(UntiedSwitch->getNumCases()),
                              PartIdLVal);
        UntiedCodeGen(CGF);
        CodeGenFunction::JumpDest CurPoint =
            CGF.getJumpDestInCurrentScope(".untied.next.");
        CGF.EmitBranchThroughCleanup(CGF.ReturnBlock);
        CGF.EmitBlock(CGF.createBasicBlock(".untied.jmp."));
        UntiedSwitch->addCase(CGF.Builder.getInt32(UntiedSwitch->getNumCases()),
                              CGF.Builder.GetInsertBlock());
        CGF.EmitBranchThroughCleanup(CurPoint);
        CGF.EmitBlock(CurPoint.getBlock());
      }
    }
    unsigned getNumberOfParts() const { return UntiedSwitch->getNumCases(); }
  };
  CGOpenMPTaskOutlinedRegionInfo(const CapturedStmt &CS,
                                 const VarDecl *ThreadIDVar,
                                 const RegionCodeGenTy &CodeGen,
                                 OpenMPDirectiveKind Kind, bool HasCancel,
                                 const UntiedTaskActionTy &Action)
      : CGOpenMPRegionInfo(CS, TaskOutlinedRegion, CodeGen, Kind, HasCancel),
        ThreadIDVar(ThreadIDVar), Action(Action) {
    assert(ThreadIDVar != nullptr && "No ThreadID in OpenMP region.");
  }

  /// Get a variable or parameter for storing global thread id
  /// inside OpenMP construct.
  const VarDecl *getThreadIDVariable() const override { return ThreadIDVar; }

  /// Get an LValue for the current ThreadID variable.
  LValue getThreadIDVariableLValue(CodeGenFunction &CGF) override;

  /// Get the name of the capture helper.
  StringRef getHelperName() const override { return ".omp_outlined."; }

  void emitUntiedSwitch(CodeGenFunction &CGF) override {
    Action.emitUntiedSwitch(CGF);
  }

  static bool classof(const CGCapturedStmtInfo *Info) {
    return CGOpenMPRegionInfo::classof(Info) &&
           cast<CGOpenMPRegionInfo>(Info)->getRegionKind() ==
               TaskOutlinedRegion;
  }

private:
  /// A variable or parameter storing global thread id for OpenMP
  /// constructs.
  const VarDecl *ThreadIDVar;
  /// Action for emitting code for untied tasks.
  const UntiedTaskActionTy &Action;
};

/// API for inlined captured statement code generation in OpenMP
/// constructs.
class CGOpenMPInlinedRegionInfo : public CGOpenMPRegionInfo {
public:
  CGOpenMPInlinedRegionInfo(CodeGenFunction::CGCapturedStmtInfo *OldCSI,
                            const RegionCodeGenTy &CodeGen,
                            OpenMPDirectiveKind Kind, bool HasCancel)
      : CGOpenMPRegionInfo(InlinedRegion, CodeGen, Kind, HasCancel),
        OldCSI(OldCSI),
        OuterRegionInfo(dyn_cast_or_null<CGOpenMPRegionInfo>(OldCSI)) {}

  // Retrieve the value of the context parameter.
  llvm::Value *getContextValue() const override {
    if (OuterRegionInfo)
      return OuterRegionInfo->getContextValue();
    llvm_unreachable("No context value for inlined OpenMP region");
  }

  void setContextValue(llvm::Value *V) override {
    if (OuterRegionInfo) {
      OuterRegionInfo->setContextValue(V);
      return;
    }
    llvm_unreachable("No context value for inlined OpenMP region");
  }

  /// Lookup the captured field decl for a variable.
  const FieldDecl *lookup(const VarDecl *VD) const override {
    if (OuterRegionInfo)
      return OuterRegionInfo->lookup(VD);
    // If there is no outer outlined region,no need to lookup in a list of
    // captured variables, we can use the original one.
    return nullptr;
  }

  FieldDecl *getThisFieldDecl() const override {
    if (OuterRegionInfo)
      return OuterRegionInfo->getThisFieldDecl();
    return nullptr;
  }

  /// Get a variable or parameter for storing global thread id
  /// inside OpenMP construct.
  const VarDecl *getThreadIDVariable() const override {
    if (OuterRegionInfo)
      return OuterRegionInfo->getThreadIDVariable();
    return nullptr;
  }

  /// Get an LValue for the current ThreadID variable.
  LValue getThreadIDVariableLValue(CodeGenFunction &CGF) override {
    if (OuterRegionInfo)
      return OuterRegionInfo->getThreadIDVariableLValue(CGF);
    llvm_unreachable("No LValue for inlined OpenMP construct");
  }

  /// Get the name of the capture helper.
  StringRef getHelperName() const override {
    if (auto *OuterRegionInfo = getOldCSI())
      return OuterRegionInfo->getHelperName();
    llvm_unreachable("No helper name for inlined OpenMP construct");
  }

  void emitUntiedSwitch(CodeGenFunction &CGF) override {
    if (OuterRegionInfo)
      OuterRegionInfo->emitUntiedSwitch(CGF);
  }

  CodeGenFunction::CGCapturedStmtInfo *getOldCSI() const { return OldCSI; }

  static bool classof(const CGCapturedStmtInfo *Info) {
    return CGOpenMPRegionInfo::classof(Info) &&
           cast<CGOpenMPRegionInfo>(Info)->getRegionKind() == InlinedRegion;
  }

  ~CGOpenMPInlinedRegionInfo() override = default;

private:
  /// CodeGen info about outer OpenMP region.
  CodeGenFunction::CGCapturedStmtInfo *OldCSI;
  CGOpenMPRegionInfo *OuterRegionInfo;
};

/// API for captured statement code generation in OpenMP target
/// constructs. For this captures, implicit parameters are used instead of the
/// captured fields. The name of the target region has to be unique in a given
/// application so it is provided by the client, because only the client has
/// the information to generate that.
class CGOpenMPTargetRegionInfo final : public CGOpenMPRegionInfo {
public:
  CGOpenMPTargetRegionInfo(const CapturedStmt &CS,
                           const RegionCodeGenTy &CodeGen, StringRef HelperName)
      : CGOpenMPRegionInfo(CS, TargetRegion, CodeGen, OMPD_target,
                           /*HasCancel=*/false),
        HelperName(HelperName) {}

  /// This is unused for target regions because each starts executing
  /// with a single thread.
  const VarDecl *getThreadIDVariable() const override { return nullptr; }

  /// Get the name of the capture helper.
  StringRef getHelperName() const override { return HelperName; }

  static bool classof(const CGCapturedStmtInfo *Info) {
    return CGOpenMPRegionInfo::classof(Info) &&
           cast<CGOpenMPRegionInfo>(Info)->getRegionKind() == TargetRegion;
  }

private:
  StringRef HelperName;
};

static void EmptyCodeGen(CodeGenFunction &, PrePostActionTy &) {
  llvm_unreachable("No codegen for expressions");
}
/// API for generation of expressions captured in a innermost OpenMP
/// region.
class CGOpenMPInnerExprInfo final : public CGOpenMPInlinedRegionInfo {
public:
  CGOpenMPInnerExprInfo(CodeGenFunction &CGF, const CapturedStmt &CS)
      : CGOpenMPInlinedRegionInfo(CGF.CapturedStmtInfo, EmptyCodeGen,
                                  OMPD_unknown,
                                  /*HasCancel=*/false),
        PrivScope(CGF) {
    // Make sure the globals captured in the provided statement are local by
    // using the privatization logic. We assume the same variable is not
    // captured more than once.
    for (const auto &C : CS.captures()) {
      if (!C.capturesVariable() && !C.capturesVariableByCopy())
        continue;

      const VarDecl *VD = C.getCapturedVar();
      if (VD->isLocalVarDeclOrParm())
        continue;

      DeclRefExpr DRE(CGF.getContext(), const_cast<VarDecl *>(VD),
                      /*RefersToEnclosingVariableOrCapture=*/false,
                      VD->getType().getNonReferenceType(), VK_LValue,
                      C.getLocation());
      PrivScope.addPrivate(
          VD, [&CGF, &DRE]() { return CGF.EmitLValue(&DRE).getAddress(); });
    }
    (void)PrivScope.Privatize();
  }

  /// Lookup the captured field decl for a variable.
  const FieldDecl *lookup(const VarDecl *VD) const override {
    if (const FieldDecl *FD = CGOpenMPInlinedRegionInfo::lookup(VD))
      return FD;
    return nullptr;
  }

  /// Emit the captured statement body.
  void EmitBody(CodeGenFunction &CGF, const Stmt *S) override {
    llvm_unreachable("No body for expressions");
  }

  /// Get a variable or parameter for storing global thread id
  /// inside OpenMP construct.
  const VarDecl *getThreadIDVariable() const override {
    llvm_unreachable("No thread id for expressions");
  }

  /// Get the name of the capture helper.
  StringRef getHelperName() const override {
    llvm_unreachable("No helper name for expressions");
  }

  static bool classof(const CGCapturedStmtInfo *Info) { return false; }

private:
  /// Private scope to capture global variables.
  CodeGenFunction::OMPPrivateScope PrivScope;
};

/// RAII for emitting code of OpenMP constructs.
class InlinedOpenMPRegionRAII {
  CodeGenFunction &CGF;
  llvm::DenseMap<const VarDecl *, FieldDecl *> LambdaCaptureFields;
  FieldDecl *LambdaThisCaptureField = nullptr;
  const CodeGen::CGBlockInfo *BlockInfo = nullptr;

public:
  /// Constructs region for combined constructs.
  /// \param CodeGen Code generation sequence for combined directives. Includes
  /// a list of functions used for code generation of implicitly inlined
  /// regions.
  InlinedOpenMPRegionRAII(CodeGenFunction &CGF, const RegionCodeGenTy &CodeGen,
                          OpenMPDirectiveKind Kind, bool HasCancel)
      : CGF(CGF) {
    // Start emission for the construct.
    CGF.CapturedStmtInfo = new CGOpenMPInlinedRegionInfo(
        CGF.CapturedStmtInfo, CodeGen, Kind, HasCancel);
    std::swap(CGF.LambdaCaptureFields, LambdaCaptureFields);
    LambdaThisCaptureField = CGF.LambdaThisCaptureField;
    CGF.LambdaThisCaptureField = nullptr;
    BlockInfo = CGF.BlockInfo;
    CGF.BlockInfo = nullptr;
  }

  ~InlinedOpenMPRegionRAII() {
    // Restore original CapturedStmtInfo only if we're done with code emission.
    auto *OldCSI =
        cast<CGOpenMPInlinedRegionInfo>(CGF.CapturedStmtInfo)->getOldCSI();
    delete CGF.CapturedStmtInfo;
    CGF.CapturedStmtInfo = OldCSI;
    std::swap(CGF.LambdaCaptureFields, LambdaCaptureFields);
    CGF.LambdaThisCaptureField = LambdaThisCaptureField;
    CGF.BlockInfo = BlockInfo;
  }
};

/// Values for bit flags used in the ident_t to describe the fields.
/// All enumeric elements are named and described in accordance with the code
/// from https://github.com/llvm/llvm-project/blob/master/openmp/runtime/src/kmp.h
enum OpenMPLocationFlags : unsigned {
  /// Use trampoline for internal microtask.
  OMP_IDENT_IMD = 0x01,
  /// Use c-style ident structure.
  OMP_IDENT_KMPC = 0x02,
  /// Atomic reduction option for kmpc_reduce.
  OMP_ATOMIC_REDUCE = 0x10,
  /// Explicit 'barrier' directive.
  OMP_IDENT_BARRIER_EXPL = 0x20,
  /// Implicit barrier in code.
  OMP_IDENT_BARRIER_IMPL = 0x40,
  /// Implicit barrier in 'for' directive.
  OMP_IDENT_BARRIER_IMPL_FOR = 0x40,
  /// Implicit barrier in 'sections' directive.
  OMP_IDENT_BARRIER_IMPL_SECTIONS = 0xC0,
  /// Implicit barrier in 'single' directive.
  OMP_IDENT_BARRIER_IMPL_SINGLE = 0x140,
  /// Call of __kmp_for_static_init for static loop.
  OMP_IDENT_WORK_LOOP = 0x200,
  /// Call of __kmp_for_static_init for sections.
  OMP_IDENT_WORK_SECTIONS = 0x400,
  /// Call of __kmp_for_static_init for distribute.
  OMP_IDENT_WORK_DISTRIBUTE = 0x800,
  LLVM_MARK_AS_BITMASK_ENUM(/*LargestValue=*/OMP_IDENT_WORK_DISTRIBUTE)
};

/// Describes ident structure that describes a source location.
/// All descriptions are taken from
/// https://github.com/llvm/llvm-project/blob/master/openmp/runtime/src/kmp.h
/// Original structure:
/// typedef struct ident {
///    kmp_int32 reserved_1;   /**<  might be used in Fortran;
///                                  see above  */
///    kmp_int32 flags;        /**<  also f.flags; KMP_IDENT_xxx flags;
///                                  KMP_IDENT_KMPC identifies this union
///                                  member  */
///    kmp_int32 reserved_2;   /**<  not really used in Fortran any more;
///                                  see above */
///#if USE_ITT_BUILD
///                            /*  but currently used for storing
///                                region-specific ITT */
///                            /*  contextual information. */
///#endif /* USE_ITT_BUILD */
///    kmp_int32 reserved_3;   /**< source[4] in Fortran, do not use for
///                                 C++  */
///    char const *psource;    /**< String describing the source location.
///                            The string is composed of semi-colon separated
//                             fields which describe the source file,
///                            the function and a pair of line numbers that
///                            delimit the construct.
///                             */
/// } ident_t;
enum IdentFieldIndex {
  /// might be used in Fortran
  IdentField_Reserved_1,
  /// OMP_IDENT_xxx flags; OMP_IDENT_KMPC identifies this union member.
  IdentField_Flags,
  /// Not really used in Fortran any more
  IdentField_Reserved_2,
  /// Source[4] in Fortran, do not use for C++
  IdentField_Reserved_3,
  /// String describing the source location. The string is composed of
  /// semi-colon separated fields which describe the source file, the function
  /// and a pair of line numbers that delimit the construct.
  IdentField_PSource
};

/// Schedule types for 'omp for' loops (these enumerators are taken from
/// the enum sched_type in kmp.h).
enum OpenMPSchedType {
  /// Lower bound for default (unordered) versions.
  OMP_sch_lower = 32,
  OMP_sch_static_chunked = 33,
  OMP_sch_static = 34,
  OMP_sch_dynamic_chunked = 35,
  OMP_sch_guided_chunked = 36,
  OMP_sch_runtime = 37,
  OMP_sch_auto = 38,
  /// static with chunk adjustment (e.g., simd)
  OMP_sch_static_balanced_chunked = 45,
  /// Lower bound for 'ordered' versions.
  OMP_ord_lower = 64,
  OMP_ord_static_chunked = 65,
  OMP_ord_static = 66,
  OMP_ord_dynamic_chunked = 67,
  OMP_ord_guided_chunked = 68,
  OMP_ord_runtime = 69,
  OMP_ord_auto = 70,
  OMP_sch_default = OMP_sch_static,
  /// dist_schedule types
  OMP_dist_sch_static_chunked = 91,
  OMP_dist_sch_static = 92,
  /// Support for OpenMP 4.5 monotonic and nonmonotonic schedule modifiers.
  /// Set if the monotonic schedule modifier was present.
  OMP_sch_modifier_monotonic = (1 << 29),
  /// Set if the nonmonotonic schedule modifier was present.
  OMP_sch_modifier_nonmonotonic = (1 << 30),
};

enum OpenMPRTLFunction {
  /// Call to void __kmpc_fork_call(ident_t *loc, kmp_int32 argc,
  /// kmpc_micro microtask, ...);
  OMPRTL__kmpc_fork_call,
  /// Call to void *__kmpc_threadprivate_cached(ident_t *loc,
  /// kmp_int32 global_tid, void *data, size_t size, void ***cache);
  OMPRTL__kmpc_threadprivate_cached,
  /// Call to void __kmpc_threadprivate_register( ident_t *,
  /// void *data, kmpc_ctor ctor, kmpc_cctor cctor, kmpc_dtor dtor);
  OMPRTL__kmpc_threadprivate_register,
  // Call to __kmpc_int32 kmpc_global_thread_num(ident_t *loc);
  OMPRTL__kmpc_global_thread_num,
  // Call to void __kmpc_critical(ident_t *loc, kmp_int32 global_tid,
  // kmp_critical_name *crit);
  OMPRTL__kmpc_critical,
  // Call to void __kmpc_critical_with_hint(ident_t *loc, kmp_int32
  // global_tid, kmp_critical_name *crit, uintptr_t hint);
  OMPRTL__kmpc_critical_with_hint,
  // Call to void __kmpc_end_critical(ident_t *loc, kmp_int32 global_tid,
  // kmp_critical_name *crit);
  OMPRTL__kmpc_end_critical,
  // Call to kmp_int32 __kmpc_cancel_barrier(ident_t *loc, kmp_int32
  // global_tid);
  OMPRTL__kmpc_cancel_barrier,
  // Call to void __kmpc_barrier(ident_t *loc, kmp_int32 global_tid);
  OMPRTL__kmpc_barrier,
  // Call to void __kmpc_for_static_fini(ident_t *loc, kmp_int32 global_tid);
  OMPRTL__kmpc_for_static_fini,
  // Call to void __kmpc_serialized_parallel(ident_t *loc, kmp_int32
  // global_tid);
  OMPRTL__kmpc_serialized_parallel,
  // Call to void __kmpc_end_serialized_parallel(ident_t *loc, kmp_int32
  // global_tid);
  OMPRTL__kmpc_end_serialized_parallel,
  // Call to void __kmpc_push_num_threads(ident_t *loc, kmp_int32 global_tid,
  // kmp_int32 num_threads);
  OMPRTL__kmpc_push_num_threads,
  // Call to void __kmpc_flush(ident_t *loc);
  OMPRTL__kmpc_flush,
  // Call to kmp_int32 __kmpc_master(ident_t *, kmp_int32 global_tid);
  OMPRTL__kmpc_master,
  // Call to void __kmpc_end_master(ident_t *, kmp_int32 global_tid);
  OMPRTL__kmpc_end_master,
  // Call to kmp_int32 __kmpc_omp_taskyield(ident_t *, kmp_int32 global_tid,
  // int end_part);
  OMPRTL__kmpc_omp_taskyield,
  // Call to kmp_int32 __kmpc_single(ident_t *, kmp_int32 global_tid);
  OMPRTL__kmpc_single,
  // Call to void __kmpc_end_single(ident_t *, kmp_int32 global_tid);
  OMPRTL__kmpc_end_single,
  // Call to kmp_task_t * __kmpc_omp_task_alloc(ident_t *, kmp_int32 gtid,
  // kmp_int32 flags, size_t sizeof_kmp_task_t, size_t sizeof_shareds,
  // kmp_routine_entry_t *task_entry);
  OMPRTL__kmpc_omp_task_alloc,
  // Call to kmp_int32 __kmpc_omp_task(ident_t *, kmp_int32 gtid, kmp_task_t *
  // new_task);
  OMPRTL__kmpc_omp_task,
  // Call to void __kmpc_copyprivate(ident_t *loc, kmp_int32 global_tid,
  // size_t cpy_size, void *cpy_data, void(*cpy_func)(void *, void *),
  // kmp_int32 didit);
  OMPRTL__kmpc_copyprivate,
  // Call to kmp_int32 __kmpc_reduce(ident_t *loc, kmp_int32 global_tid,
  // kmp_int32 num_vars, size_t reduce_size, void *reduce_data, void
  // (*reduce_func)(void *lhs_data, void *rhs_data), kmp_critical_name *lck);
  OMPRTL__kmpc_reduce,
  // Call to kmp_int32 __kmpc_reduce_nowait(ident_t *loc, kmp_int32
  // global_tid, kmp_int32 num_vars, size_t reduce_size, void *reduce_data,
  // void (*reduce_func)(void *lhs_data, void *rhs_data), kmp_critical_name
  // *lck);
  OMPRTL__kmpc_reduce_nowait,
  // Call to void __kmpc_end_reduce(ident_t *loc, kmp_int32 global_tid,
  // kmp_critical_name *lck);
  OMPRTL__kmpc_end_reduce,
  // Call to void __kmpc_end_reduce_nowait(ident_t *loc, kmp_int32 global_tid,
  // kmp_critical_name *lck);
  OMPRTL__kmpc_end_reduce_nowait,
  // Call to void __kmpc_omp_task_begin_if0(ident_t *, kmp_int32 gtid,
  // kmp_task_t * new_task);
  OMPRTL__kmpc_omp_task_begin_if0,
  // Call to void __kmpc_omp_task_complete_if0(ident_t *, kmp_int32 gtid,
  // kmp_task_t * new_task);
  OMPRTL__kmpc_omp_task_complete_if0,
  // Call to void __kmpc_ordered(ident_t *loc, kmp_int32 global_tid);
  OMPRTL__kmpc_ordered,
  // Call to void __kmpc_end_ordered(ident_t *loc, kmp_int32 global_tid);
  OMPRTL__kmpc_end_ordered,
  // Call to kmp_int32 __kmpc_omp_taskwait(ident_t *loc, kmp_int32
  // global_tid);
  OMPRTL__kmpc_omp_taskwait,
  // Call to void __kmpc_taskgroup(ident_t *loc, kmp_int32 global_tid);
  OMPRTL__kmpc_taskgroup,
  // Call to void __kmpc_end_taskgroup(ident_t *loc, kmp_int32 global_tid);
  OMPRTL__kmpc_end_taskgroup,
  // Call to void __kmpc_push_proc_bind(ident_t *loc, kmp_int32 global_tid,
  // int proc_bind);
  OMPRTL__kmpc_push_proc_bind,
  // Call to kmp_int32 __kmpc_omp_task_with_deps(ident_t *loc_ref, kmp_int32
  // gtid, kmp_task_t * new_task, kmp_int32 ndeps, kmp_depend_info_t
  // *dep_list, kmp_int32 ndeps_noalias, kmp_depend_info_t *noalias_dep_list);
  OMPRTL__kmpc_omp_task_with_deps,
  // Call to void __kmpc_omp_wait_deps(ident_t *loc_ref, kmp_int32
  // gtid, kmp_int32 ndeps, kmp_depend_info_t *dep_list, kmp_int32
  // ndeps_noalias, kmp_depend_info_t *noalias_dep_list);
  OMPRTL__kmpc_omp_wait_deps,
  // Call to kmp_int32 __kmpc_cancellationpoint(ident_t *loc, kmp_int32
  // global_tid, kmp_int32 cncl_kind);
  OMPRTL__kmpc_cancellationpoint,
  // Call to kmp_int32 __kmpc_cancel(ident_t *loc, kmp_int32 global_tid,
  // kmp_int32 cncl_kind);
  OMPRTL__kmpc_cancel,
  // Call to void __kmpc_push_num_teams(ident_t *loc, kmp_int32 global_tid,
  // kmp_int32 num_teams, kmp_int32 thread_limit);
  OMPRTL__kmpc_push_num_teams,
  // Call to void __kmpc_fork_teams(ident_t *loc, kmp_int32 argc, kmpc_micro
  // microtask, ...);
  OMPRTL__kmpc_fork_teams,
  // Call to void __kmpc_taskloop(ident_t *loc, int gtid, kmp_task_t *task, int
  // if_val, kmp_uint64 *lb, kmp_uint64 *ub, kmp_int64 st, int nogroup, int
  // sched, kmp_uint64 grainsize, void *task_dup);
  OMPRTL__kmpc_taskloop,
  // Call to void __kmpc_doacross_init(ident_t *loc, kmp_int32 gtid, kmp_int32
  // num_dims, struct kmp_dim *dims);
  OMPRTL__kmpc_doacross_init,
  // Call to void __kmpc_doacross_fini(ident_t *loc, kmp_int32 gtid);
  OMPRTL__kmpc_doacross_fini,
  // Call to void __kmpc_doacross_post(ident_t *loc, kmp_int32 gtid, kmp_int64
  // *vec);
  OMPRTL__kmpc_doacross_post,
  // Call to void __kmpc_doacross_wait(ident_t *loc, kmp_int32 gtid, kmp_int64
  // *vec);
  OMPRTL__kmpc_doacross_wait,
  // Call to void *__kmpc_task_reduction_init(int gtid, int num_data, void
  // *data);
  OMPRTL__kmpc_task_reduction_init,
  // Call to void *__kmpc_task_reduction_get_th_data(int gtid, void *tg, void
  // *d);
  OMPRTL__kmpc_task_reduction_get_th_data,

  //
  // Offloading related calls
  //
  // Call to void __kmpc_push_target_tripcount(int64_t device_id, kmp_uint64
  // size);
  OMPRTL__kmpc_push_target_tripcount,
  // Call to int32_t __tgt_target(int64_t device_id, void *host_ptr, int32_t
  // arg_num, void** args_base, void **args, size_t *arg_sizes, int64_t
  // *arg_types);
  OMPRTL__tgt_target,
  // Call to int32_t __tgt_target_nowait(int64_t device_id, void *host_ptr,
  // int32_t arg_num, void** args_base, void **args, size_t *arg_sizes, int64_t
  // *arg_types);
  OMPRTL__tgt_target_nowait,
  // Call to int32_t __tgt_target_teams(int64_t device_id, void *host_ptr,
  // int32_t arg_num, void** args_base, void **args, size_t *arg_sizes, int64_t
  // *arg_types, int32_t num_teams, int32_t thread_limit);
  OMPRTL__tgt_target_teams,
  // Call to int32_t __tgt_target_teams_nowait(int64_t device_id, void
  // *host_ptr, int32_t arg_num, void** args_base, void **args, size_t
  // *arg_sizes, int64_t *arg_types, int32_t num_teams, int32_t thread_limit);
  OMPRTL__tgt_target_teams_nowait,
  // Call to void __tgt_register_lib(__tgt_bin_desc *desc);
  OMPRTL__tgt_register_lib,
  // Call to void __tgt_unregister_lib(__tgt_bin_desc *desc);
  OMPRTL__tgt_unregister_lib,
  // Call to void __tgt_target_data_begin(int64_t device_id, int32_t arg_num,
  // void** args_base, void **args, size_t *arg_sizes, int64_t *arg_types);
  OMPRTL__tgt_target_data_begin,
  // Call to void __tgt_target_data_begin_nowait(int64_t device_id, int32_t
  // arg_num, void** args_base, void **args, size_t *arg_sizes, int64_t
  // *arg_types);
  OMPRTL__tgt_target_data_begin_nowait,
  // Call to void __tgt_target_data_end(int64_t device_id, int32_t arg_num,
  // void** args_base, void **args, size_t *arg_sizes, int64_t *arg_types);
  OMPRTL__tgt_target_data_end,
  // Call to void __tgt_target_data_end_nowait(int64_t device_id, int32_t
  // arg_num, void** args_base, void **args, size_t *arg_sizes, int64_t
  // *arg_types);
  OMPRTL__tgt_target_data_end_nowait,
  // Call to void __tgt_target_data_update(int64_t device_id, int32_t arg_num,
  // void** args_base, void **args, size_t *arg_sizes, int64_t *arg_types);
  OMPRTL__tgt_target_data_update,
  // Call to void __tgt_target_data_update_nowait(int64_t device_id, int32_t
  // arg_num, void** args_base, void **args, size_t *arg_sizes, int64_t
  // *arg_types);
  OMPRTL__tgt_target_data_update_nowait,
};

/// A basic class for pre|post-action for advanced codegen sequence for OpenMP
/// region.
class CleanupTy final : public EHScopeStack::Cleanup {
  PrePostActionTy *Action;

public:
  explicit CleanupTy(PrePostActionTy *Action) : Action(Action) {}
  void Emit(CodeGenFunction &CGF, Flags /*flags*/) override {
    if (!CGF.HaveInsertPoint())
      return;
    Action->Exit(CGF);
  }
};

} // anonymous namespace

void RegionCodeGenTy::operator()(CodeGenFunction &CGF) const {
  CodeGenFunction::RunCleanupsScope Scope(CGF);
  if (PrePostAction) {
    CGF.EHStack.pushCleanup<CleanupTy>(NormalAndEHCleanup, PrePostAction);
    Callback(CodeGen, CGF, *PrePostAction);
  } else {
    PrePostActionTy Action;
    Callback(CodeGen, CGF, Action);
  }
}

/// Check if the combiner is a call to UDR combiner and if it is so return the
/// UDR decl used for reduction.
static const OMPDeclareReductionDecl *
getReductionInit(const Expr *ReductionOp) {
  if (const auto *CE = dyn_cast<CallExpr>(ReductionOp))
    if (const auto *OVE = dyn_cast<OpaqueValueExpr>(CE->getCallee()))
      if (const auto *DRE =
              dyn_cast<DeclRefExpr>(OVE->getSourceExpr()->IgnoreImpCasts()))
        if (const auto *DRD = dyn_cast<OMPDeclareReductionDecl>(DRE->getDecl()))
          return DRD;
  return nullptr;
}

static void emitInitWithReductionInitializer(CodeGenFunction &CGF,
                                             const OMPDeclareReductionDecl *DRD,
                                             const Expr *InitOp,
                                             Address Private, Address Original,
                                             QualType Ty) {
  if (DRD->getInitializer()) {
    std::pair<llvm::Function *, llvm::Function *> Reduction =
        CGF.CGM.getOpenMPRuntime().getUserDefinedReduction(DRD);
    const auto *CE = cast<CallExpr>(InitOp);
    const auto *OVE = cast<OpaqueValueExpr>(CE->getCallee());
    const Expr *LHS = CE->getArg(/*Arg=*/0)->IgnoreParenImpCasts();
    const Expr *RHS = CE->getArg(/*Arg=*/1)->IgnoreParenImpCasts();
    const auto *LHSDRE =
        cast<DeclRefExpr>(cast<UnaryOperator>(LHS)->getSubExpr());
    const auto *RHSDRE =
        cast<DeclRefExpr>(cast<UnaryOperator>(RHS)->getSubExpr());
    CodeGenFunction::OMPPrivateScope PrivateScope(CGF);
    PrivateScope.addPrivate(cast<VarDecl>(LHSDRE->getDecl()),
                            [=]() { return Private; });
    PrivateScope.addPrivate(cast<VarDecl>(RHSDRE->getDecl()),
                            [=]() { return Original; });
    (void)PrivateScope.Privatize();
    RValue Func = RValue::get(Reduction.second);
    CodeGenFunction::OpaqueValueMapping Map(CGF, OVE, Func);
    CGF.EmitIgnoredExpr(InitOp);
  } else {
    llvm::Constant *Init = CGF.CGM.EmitNullConstant(Ty);
    std::string Name = CGF.CGM.getOpenMPRuntime().getName({"init"});
    auto *GV = new llvm::GlobalVariable(
        CGF.CGM.getModule(), Init->getType(), /*isConstant=*/true,
        llvm::GlobalValue::PrivateLinkage, Init, Name);
    LValue LV = CGF.MakeNaturalAlignAddrLValue(GV, Ty);
    RValue InitRVal;
    switch (CGF.getEvaluationKind(Ty)) {
    case TEK_Scalar:
      InitRVal = CGF.EmitLoadOfLValue(LV, DRD->getLocation());
      break;
    case TEK_Complex:
      InitRVal =
          RValue::getComplex(CGF.EmitLoadOfComplex(LV, DRD->getLocation()));
      break;
    case TEK_Aggregate:
      InitRVal = RValue::getAggregate(LV.getAddress());
      break;
    }
    OpaqueValueExpr OVE(DRD->getLocation(), Ty, VK_RValue);
    CodeGenFunction::OpaqueValueMapping OpaqueMap(CGF, &OVE, InitRVal);
    CGF.EmitAnyExprToMem(&OVE, Private, Ty.getQualifiers(),
                         /*IsInitializer=*/false);
  }
}

/// Emit initialization of arrays of complex types.
/// \param DestAddr Address of the array.
/// \param Type Type of array.
/// \param Init Initial expression of array.
/// \param SrcAddr Address of the original array.
static void EmitOMPAggregateInit(CodeGenFunction &CGF, Address DestAddr,
                                 QualType Type, bool EmitDeclareReductionInit,
                                 const Expr *Init,
                                 const OMPDeclareReductionDecl *DRD,
                                 Address SrcAddr = Address::invalid()) {
  // Perform element-by-element initialization.
  QualType ElementTy;

  // Drill down to the base element type on both arrays.
  const ArrayType *ArrayTy = Type->getAsArrayTypeUnsafe();
  llvm::Value *NumElements = CGF.emitArrayLength(ArrayTy, ElementTy, DestAddr);
  DestAddr =
      CGF.Builder.CreateElementBitCast(DestAddr, DestAddr.getElementType());
  if (DRD)
    SrcAddr =
        CGF.Builder.CreateElementBitCast(SrcAddr, DestAddr.getElementType());

  llvm::Value *SrcBegin = nullptr;
  if (DRD)
    SrcBegin = SrcAddr.getPointer();
  llvm::Value *DestBegin = DestAddr.getPointer();
  // Cast from pointer to array type to pointer to single element.
  llvm::Value *DestEnd = CGF.Builder.CreateGEP(DestBegin, NumElements);
  // The basic structure here is a while-do loop.
  llvm::BasicBlock *BodyBB = CGF.createBasicBlock("omp.arrayinit.body");
  llvm::BasicBlock *DoneBB = CGF.createBasicBlock("omp.arrayinit.done");
  llvm::Value *IsEmpty =
      CGF.Builder.CreateICmpEQ(DestBegin, DestEnd, "omp.arrayinit.isempty");
  CGF.Builder.CreateCondBr(IsEmpty, DoneBB, BodyBB);

  // Enter the loop body, making that address the current address.
  llvm::BasicBlock *EntryBB = CGF.Builder.GetInsertBlock();
  CGF.EmitBlock(BodyBB);

  CharUnits ElementSize = CGF.getContext().getTypeSizeInChars(ElementTy);

  llvm::PHINode *SrcElementPHI = nullptr;
  Address SrcElementCurrent = Address::invalid();
  if (DRD) {
    SrcElementPHI = CGF.Builder.CreatePHI(SrcBegin->getType(), 2,
                                          "omp.arraycpy.srcElementPast");
    SrcElementPHI->addIncoming(SrcBegin, EntryBB);
    SrcElementCurrent =
        Address(SrcElementPHI,
                SrcAddr.getAlignment().alignmentOfArrayElement(ElementSize));
  }
  llvm::PHINode *DestElementPHI = CGF.Builder.CreatePHI(
      DestBegin->getType(), 2, "omp.arraycpy.destElementPast");
  DestElementPHI->addIncoming(DestBegin, EntryBB);
  Address DestElementCurrent =
      Address(DestElementPHI,
              DestAddr.getAlignment().alignmentOfArrayElement(ElementSize));

  // Emit copy.
  {
    CodeGenFunction::RunCleanupsScope InitScope(CGF);
    if (EmitDeclareReductionInit) {
      emitInitWithReductionInitializer(CGF, DRD, Init, DestElementCurrent,
                                       SrcElementCurrent, ElementTy);
    } else
      CGF.EmitAnyExprToMem(Init, DestElementCurrent, ElementTy.getQualifiers(),
                           /*IsInitializer=*/false);
  }

  if (DRD) {
    // Shift the address forward by one element.
    llvm::Value *SrcElementNext = CGF.Builder.CreateConstGEP1_32(
        SrcElementPHI, /*Idx0=*/1, "omp.arraycpy.dest.element");
    SrcElementPHI->addIncoming(SrcElementNext, CGF.Builder.GetInsertBlock());
  }

  // Shift the address forward by one element.
  llvm::Value *DestElementNext = CGF.Builder.CreateConstGEP1_32(
      DestElementPHI, /*Idx0=*/1, "omp.arraycpy.dest.element");
  // Check whether we've reached the end.
  llvm::Value *Done =
      CGF.Builder.CreateICmpEQ(DestElementNext, DestEnd, "omp.arraycpy.done");
  CGF.Builder.CreateCondBr(Done, DoneBB, BodyBB);
  DestElementPHI->addIncoming(DestElementNext, CGF.Builder.GetInsertBlock());

  // Done.
  CGF.EmitBlock(DoneBB, /*IsFinished=*/true);
}

LValue ReductionCodeGen::emitSharedLValue(CodeGenFunction &CGF, const Expr *E) {
  return CGF.EmitOMPSharedLValue(E);
}

LValue ReductionCodeGen::emitSharedLValueUB(CodeGenFunction &CGF,
                                            const Expr *E) {
  if (const auto *OASE = dyn_cast<OMPArraySectionExpr>(E))
    return CGF.EmitOMPArraySectionExpr(OASE, /*IsLowerBound=*/false);
  return LValue();
}

void ReductionCodeGen::emitAggregateInitialization(
    CodeGenFunction &CGF, unsigned N, Address PrivateAddr, LValue SharedLVal,
    const OMPDeclareReductionDecl *DRD) {
  // Emit VarDecl with copy init for arrays.
  // Get the address of the original variable captured in current
  // captured region.
  const auto *PrivateVD =
      cast<VarDecl>(cast<DeclRefExpr>(ClausesData[N].Private)->getDecl());
  bool EmitDeclareReductionInit =
      DRD && (DRD->getInitializer() || !PrivateVD->hasInit());
  EmitOMPAggregateInit(CGF, PrivateAddr, PrivateVD->getType(),
                       EmitDeclareReductionInit,
                       EmitDeclareReductionInit ? ClausesData[N].ReductionOp
                                                : PrivateVD->getInit(),
                       DRD, SharedLVal.getAddress());
}

ReductionCodeGen::ReductionCodeGen(ArrayRef<const Expr *> Shareds,
                                   ArrayRef<const Expr *> Privates,
                                   ArrayRef<const Expr *> ReductionOps) {
  ClausesData.reserve(Shareds.size());
  SharedAddresses.reserve(Shareds.size());
  Sizes.reserve(Shareds.size());
  BaseDecls.reserve(Shareds.size());
  auto IPriv = Privates.begin();
  auto IRed = ReductionOps.begin();
  for (const Expr *Ref : Shareds) {
    ClausesData.emplace_back(Ref, *IPriv, *IRed);
    std::advance(IPriv, 1);
    std::advance(IRed, 1);
  }
}

void ReductionCodeGen::emitSharedLValue(CodeGenFunction &CGF, unsigned N) {
  assert(SharedAddresses.size() == N &&
         "Number of generated lvalues must be exactly N.");
  LValue First = emitSharedLValue(CGF, ClausesData[N].Ref);
  LValue Second = emitSharedLValueUB(CGF, ClausesData[N].Ref);
  SharedAddresses.emplace_back(First, Second);
}

void ReductionCodeGen::emitAggregateType(CodeGenFunction &CGF, unsigned N) {
  const auto *PrivateVD =
      cast<VarDecl>(cast<DeclRefExpr>(ClausesData[N].Private)->getDecl());
  QualType PrivateType = PrivateVD->getType();
  bool AsArraySection = isa<OMPArraySectionExpr>(ClausesData[N].Ref);
  if (!PrivateType->isVariablyModifiedType()) {
    Sizes.emplace_back(
        CGF.getTypeSize(
            SharedAddresses[N].first.getType().getNonReferenceType()),
        nullptr);
    return;
  }
  llvm::Value *Size;
  llvm::Value *SizeInChars;
  auto *ElemType =
      cast<llvm::PointerType>(SharedAddresses[N].first.getPointer()->getType())
          ->getElementType();
  auto *ElemSizeOf = llvm::ConstantExpr::getSizeOf(ElemType);
  if (AsArraySection) {
    Size = CGF.Builder.CreatePtrDiff(SharedAddresses[N].second.getPointer(),
                                     SharedAddresses[N].first.getPointer());
    Size = CGF.Builder.CreateNUWAdd(
        Size, llvm::ConstantInt::get(Size->getType(), /*V=*/1));
    SizeInChars = CGF.Builder.CreateNUWMul(Size, ElemSizeOf);
  } else {
    SizeInChars = CGF.getTypeSize(
        SharedAddresses[N].first.getType().getNonReferenceType());
    Size = CGF.Builder.CreateExactUDiv(SizeInChars, ElemSizeOf);
  }
  Sizes.emplace_back(SizeInChars, Size);
  CodeGenFunction::OpaqueValueMapping OpaqueMap(
      CGF,
      cast<OpaqueValueExpr>(
          CGF.getContext().getAsVariableArrayType(PrivateType)->getSizeExpr()),
      RValue::get(Size));
  CGF.EmitVariablyModifiedType(PrivateType);
}

void ReductionCodeGen::emitAggregateType(CodeGenFunction &CGF, unsigned N,
                                         llvm::Value *Size) {
  const auto *PrivateVD =
      cast<VarDecl>(cast<DeclRefExpr>(ClausesData[N].Private)->getDecl());
  QualType PrivateType = PrivateVD->getType();
  if (!PrivateType->isVariablyModifiedType()) {
    assert(!Size && !Sizes[N].second &&
           "Size should be nullptr for non-variably modified reduction "
           "items.");
    return;
  }
  CodeGenFunction::OpaqueValueMapping OpaqueMap(
      CGF,
      cast<OpaqueValueExpr>(
          CGF.getContext().getAsVariableArrayType(PrivateType)->getSizeExpr()),
      RValue::get(Size));
  CGF.EmitVariablyModifiedType(PrivateType);
}

void ReductionCodeGen::emitInitialization(
    CodeGenFunction &CGF, unsigned N, Address PrivateAddr, LValue SharedLVal,
    llvm::function_ref<bool(CodeGenFunction &)> DefaultInit) {
  assert(SharedAddresses.size() > N && "No variable was generated");
  const auto *PrivateVD =
      cast<VarDecl>(cast<DeclRefExpr>(ClausesData[N].Private)->getDecl());
  const OMPDeclareReductionDecl *DRD =
      getReductionInit(ClausesData[N].ReductionOp);
  QualType PrivateType = PrivateVD->getType();
  PrivateAddr = CGF.Builder.CreateElementBitCast(
      PrivateAddr, CGF.ConvertTypeForMem(PrivateType));
  QualType SharedType = SharedAddresses[N].first.getType();
  SharedLVal = CGF.MakeAddrLValue(
      CGF.Builder.CreateElementBitCast(SharedLVal.getAddress(),
                                       CGF.ConvertTypeForMem(SharedType)),
      SharedType, SharedAddresses[N].first.getBaseInfo(),
      CGF.CGM.getTBAAInfoForSubobject(SharedAddresses[N].first, SharedType));
  if (CGF.getContext().getAsArrayType(PrivateVD->getType())) {
    emitAggregateInitialization(CGF, N, PrivateAddr, SharedLVal, DRD);
  } else if (DRD && (DRD->getInitializer() || !PrivateVD->hasInit())) {
    emitInitWithReductionInitializer(CGF, DRD, ClausesData[N].ReductionOp,
                                     PrivateAddr, SharedLVal.getAddress(),
                                     SharedLVal.getType());
  } else if (!DefaultInit(CGF) && PrivateVD->hasInit() &&
             !CGF.isTrivialInitializer(PrivateVD->getInit())) {
    CGF.EmitAnyExprToMem(PrivateVD->getInit(), PrivateAddr,
                         PrivateVD->getType().getQualifiers(),
                         /*IsInitializer=*/false);
  }
}

bool ReductionCodeGen::needCleanups(unsigned N) {
  const auto *PrivateVD =
      cast<VarDecl>(cast<DeclRefExpr>(ClausesData[N].Private)->getDecl());
  QualType PrivateType = PrivateVD->getType();
  QualType::DestructionKind DTorKind = PrivateType.isDestructedType();
  return DTorKind != QualType::DK_none;
}

void ReductionCodeGen::emitCleanups(CodeGenFunction &CGF, unsigned N,
                                    Address PrivateAddr) {
  const auto *PrivateVD =
      cast<VarDecl>(cast<DeclRefExpr>(ClausesData[N].Private)->getDecl());
  QualType PrivateType = PrivateVD->getType();
  QualType::DestructionKind DTorKind = PrivateType.isDestructedType();
  if (needCleanups(N)) {
    PrivateAddr = CGF.Builder.CreateElementBitCast(
        PrivateAddr, CGF.ConvertTypeForMem(PrivateType));
    CGF.pushDestroy(DTorKind, PrivateAddr, PrivateType);
  }
}

static LValue loadToBegin(CodeGenFunction &CGF, QualType BaseTy, QualType ElTy,
                          LValue BaseLV) {
  BaseTy = BaseTy.getNonReferenceType();
  while ((BaseTy->isPointerType() || BaseTy->isReferenceType()) &&
         !CGF.getContext().hasSameType(BaseTy, ElTy)) {
    if (const auto *PtrTy = BaseTy->getAs<PointerType>()) {
      BaseLV = CGF.EmitLoadOfPointerLValue(BaseLV.getAddress(), PtrTy);
    } else {
      LValue RefLVal = CGF.MakeAddrLValue(BaseLV.getAddress(), BaseTy);
      BaseLV = CGF.EmitLoadOfReferenceLValue(RefLVal);
    }
    BaseTy = BaseTy->getPointeeType();
  }
  return CGF.MakeAddrLValue(
      CGF.Builder.CreateElementBitCast(BaseLV.getAddress(),
                                       CGF.ConvertTypeForMem(ElTy)),
      BaseLV.getType(), BaseLV.getBaseInfo(),
      CGF.CGM.getTBAAInfoForSubobject(BaseLV, BaseLV.getType()));
}

static Address castToBase(CodeGenFunction &CGF, QualType BaseTy, QualType ElTy,
                          llvm::Type *BaseLVType, CharUnits BaseLVAlignment,
                          llvm::Value *Addr) {
  Address Tmp = Address::invalid();
  Address TopTmp = Address::invalid();
  Address MostTopTmp = Address::invalid();
  BaseTy = BaseTy.getNonReferenceType();
  while ((BaseTy->isPointerType() || BaseTy->isReferenceType()) &&
         !CGF.getContext().hasSameType(BaseTy, ElTy)) {
    Tmp = CGF.CreateMemTemp(BaseTy);
    if (TopTmp.isValid())
      CGF.Builder.CreateStore(Tmp.getPointer(), TopTmp);
    else
      MostTopTmp = Tmp;
    TopTmp = Tmp;
    BaseTy = BaseTy->getPointeeType();
  }
  llvm::Type *Ty = BaseLVType;
  if (Tmp.isValid())
    Ty = Tmp.getElementType();
  Addr = CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(Addr, Ty);
  if (Tmp.isValid()) {
    CGF.Builder.CreateStore(Addr, Tmp);
    return MostTopTmp;
  }
  return Address(Addr, BaseLVAlignment);
}

static const VarDecl *getBaseDecl(const Expr *Ref, const DeclRefExpr *&DE) {
  const VarDecl *OrigVD = nullptr;
  if (const auto *OASE = dyn_cast<OMPArraySectionExpr>(Ref)) {
    const Expr *Base = OASE->getBase()->IgnoreParenImpCasts();
    while (const auto *TempOASE = dyn_cast<OMPArraySectionExpr>(Base))
      Base = TempOASE->getBase()->IgnoreParenImpCasts();
    while (const auto *TempASE = dyn_cast<ArraySubscriptExpr>(Base))
      Base = TempASE->getBase()->IgnoreParenImpCasts();
    DE = cast<DeclRefExpr>(Base);
    OrigVD = cast<VarDecl>(DE->getDecl());
  } else if (const auto *ASE = dyn_cast<ArraySubscriptExpr>(Ref)) {
    const Expr *Base = ASE->getBase()->IgnoreParenImpCasts();
    while (const auto *TempASE = dyn_cast<ArraySubscriptExpr>(Base))
      Base = TempASE->getBase()->IgnoreParenImpCasts();
    DE = cast<DeclRefExpr>(Base);
    OrigVD = cast<VarDecl>(DE->getDecl());
  }
  return OrigVD;
}

Address ReductionCodeGen::adjustPrivateAddress(CodeGenFunction &CGF, unsigned N,
                                               Address PrivateAddr) {
  const DeclRefExpr *DE;
  if (const VarDecl *OrigVD = ::getBaseDecl(ClausesData[N].Ref, DE)) {
    BaseDecls.emplace_back(OrigVD);
    LValue OriginalBaseLValue = CGF.EmitLValue(DE);
    LValue BaseLValue =
        loadToBegin(CGF, OrigVD->getType(), SharedAddresses[N].first.getType(),
                    OriginalBaseLValue);
    llvm::Value *Adjustment = CGF.Builder.CreatePtrDiff(
        BaseLValue.getPointer(), SharedAddresses[N].first.getPointer());
    llvm::Value *PrivatePointer =
        CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
            PrivateAddr.getPointer(),
            SharedAddresses[N].first.getAddress().getType());
    llvm::Value *Ptr = CGF.Builder.CreateGEP(PrivatePointer, Adjustment);
    return castToBase(CGF, OrigVD->getType(),
                      SharedAddresses[N].first.getType(),
                      OriginalBaseLValue.getAddress().getType(),
                      OriginalBaseLValue.getAlignment(), Ptr);
  }
  BaseDecls.emplace_back(
      cast<VarDecl>(cast<DeclRefExpr>(ClausesData[N].Ref)->getDecl()));
  return PrivateAddr;
}

bool ReductionCodeGen::usesReductionInitializer(unsigned N) const {
  const OMPDeclareReductionDecl *DRD =
      getReductionInit(ClausesData[N].ReductionOp);
  return DRD && DRD->getInitializer();
}

LValue CGOpenMPRegionInfo::getThreadIDVariableLValue(CodeGenFunction &CGF) {
  return CGF.EmitLoadOfPointerLValue(
      CGF.GetAddrOfLocalVar(getThreadIDVariable()),
      getThreadIDVariable()->getType()->castAs<PointerType>());
}

void CGOpenMPRegionInfo::EmitBody(CodeGenFunction &CGF, const Stmt * /*S*/) {
  if (!CGF.HaveInsertPoint())
    return;
  // 1.2.2 OpenMP Language Terminology
  // Structured block - An executable statement with a single entry at the
  // top and a single exit at the bottom.
  // The point of exit cannot be a branch out of the structured block.
  // longjmp() and throw() must not violate the entry/exit criteria.
  CGF.EHStack.pushTerminate();
  CodeGen(CGF);
  CGF.EHStack.popTerminate();
}

LValue CGOpenMPTaskOutlinedRegionInfo::getThreadIDVariableLValue(
    CodeGenFunction &CGF) {
  return CGF.MakeAddrLValue(CGF.GetAddrOfLocalVar(getThreadIDVariable()),
                            getThreadIDVariable()->getType(),
                            AlignmentSource::Decl);
}

static FieldDecl *addFieldToRecordDecl(ASTContext &C, DeclContext *DC,
                                       QualType FieldTy) {
  auto *Field = FieldDecl::Create(
      C, DC, SourceLocation(), SourceLocation(), /*Id=*/nullptr, FieldTy,
      C.getTrivialTypeSourceInfo(FieldTy, SourceLocation()),
      /*BW=*/nullptr, /*Mutable=*/false, /*InitStyle=*/ICIS_NoInit);
  Field->setAccess(AS_public);
  DC->addDecl(Field);
  return Field;
}

CGOpenMPRuntime::CGOpenMPRuntime(CodeGenModule &CGM, StringRef FirstSeparator,
                                 StringRef Separator)
    : CGM(CGM), FirstSeparator(FirstSeparator), Separator(Separator),
      OffloadEntriesInfoManager(CGM) {
  ASTContext &C = CGM.getContext();
  RecordDecl *RD = C.buildImplicitRecord("ident_t");
  QualType KmpInt32Ty = C.getIntTypeForBitwidth(/*DestWidth=*/32, /*Signed=*/1);
  RD->startDefinition();
  // reserved_1
  addFieldToRecordDecl(C, RD, KmpInt32Ty);
  // flags
  addFieldToRecordDecl(C, RD, KmpInt32Ty);
  // reserved_2
  addFieldToRecordDecl(C, RD, KmpInt32Ty);
  // reserved_3
  addFieldToRecordDecl(C, RD, KmpInt32Ty);
  // psource
  addFieldToRecordDecl(C, RD, C.VoidPtrTy);
  RD->completeDefinition();
  IdentQTy = C.getRecordType(RD);
  IdentTy = CGM.getTypes().ConvertRecordDeclType(RD);
  KmpCriticalNameTy = llvm::ArrayType::get(CGM.Int32Ty, /*NumElements*/ 8);

  loadOffloadInfoMetadata();
}

void CGOpenMPRuntime::clear() {
  InternalVars.clear();
  // Clean non-target variable declarations possibly used only in debug info.
  for (const auto &Data : EmittedNonTargetVariables) {
    if (!Data.getValue().pointsToAliveValue())
      continue;
    auto *GV = dyn_cast<llvm::GlobalVariable>(Data.getValue());
    if (!GV)
      continue;
    if (!GV->isDeclaration() || GV->getNumUses() > 0)
      continue;
    GV->eraseFromParent();
  }
}

std::string CGOpenMPRuntime::getName(ArrayRef<StringRef> Parts) const {
  SmallString<128> Buffer;
  llvm::raw_svector_ostream OS(Buffer);
  StringRef Sep = FirstSeparator;
  for (StringRef Part : Parts) {
    OS << Sep << Part;
    Sep = Separator;
  }
  return OS.str();
}

static llvm::Function *
emitCombinerOrInitializer(CodeGenModule &CGM, QualType Ty,
                          const Expr *CombinerInitializer, const VarDecl *In,
                          const VarDecl *Out, bool IsCombiner) {
  // void .omp_combiner.(Ty *in, Ty *out);
  ASTContext &C = CGM.getContext();
  QualType PtrTy = C.getPointerType(Ty).withRestrict();
  FunctionArgList Args;
  ImplicitParamDecl OmpOutParm(C, /*DC=*/nullptr, Out->getLocation(),
                               /*Id=*/nullptr, PtrTy, ImplicitParamDecl::Other);
  ImplicitParamDecl OmpInParm(C, /*DC=*/nullptr, In->getLocation(),
                              /*Id=*/nullptr, PtrTy, ImplicitParamDecl::Other);
  Args.push_back(&OmpOutParm);
  Args.push_back(&OmpInParm);
  const CGFunctionInfo &FnInfo =
      CGM.getTypes().arrangeBuiltinFunctionDeclaration(C.VoidTy, Args);
  llvm::FunctionType *FnTy = CGM.getTypes().GetFunctionType(FnInfo);
  std::string Name = CGM.getOpenMPRuntime().getName(
      {IsCombiner ? "omp_combiner" : "omp_initializer", ""});
  auto *Fn = llvm::Function::Create(FnTy, llvm::GlobalValue::InternalLinkage,
                                    Name, &CGM.getModule());
  CGM.SetInternalFunctionAttributes(GlobalDecl(), Fn, FnInfo);
  Fn->removeFnAttr(llvm::Attribute::NoInline);
  Fn->removeFnAttr(llvm::Attribute::OptimizeNone);
  Fn->addFnAttr(llvm::Attribute::AlwaysInline);
  CodeGenFunction CGF(CGM);
  // Map "T omp_in;" variable to "*omp_in_parm" value in all expressions.
  // Map "T omp_out;" variable to "*omp_out_parm" value in all expressions.
  CGF.StartFunction(GlobalDecl(), C.VoidTy, Fn, FnInfo, Args, In->getLocation(),
                    Out->getLocation());
  CodeGenFunction::OMPPrivateScope Scope(CGF);
  Address AddrIn = CGF.GetAddrOfLocalVar(&OmpInParm);
  Scope.addPrivate(In, [&CGF, AddrIn, PtrTy]() {
    return CGF.EmitLoadOfPointerLValue(AddrIn, PtrTy->castAs<PointerType>())
        .getAddress();
  });
  Address AddrOut = CGF.GetAddrOfLocalVar(&OmpOutParm);
  Scope.addPrivate(Out, [&CGF, AddrOut, PtrTy]() {
    return CGF.EmitLoadOfPointerLValue(AddrOut, PtrTy->castAs<PointerType>())
        .getAddress();
  });
  (void)Scope.Privatize();
  if (!IsCombiner && Out->hasInit() &&
      !CGF.isTrivialInitializer(Out->getInit())) {
    CGF.EmitAnyExprToMem(Out->getInit(), CGF.GetAddrOfLocalVar(Out),
                         Out->getType().getQualifiers(),
                         /*IsInitializer=*/true);
  }
  if (CombinerInitializer)
    CGF.EmitIgnoredExpr(CombinerInitializer);
  Scope.ForceCleanup();
  CGF.FinishFunction();
  return Fn;
}

void CGOpenMPRuntime::emitUserDefinedReduction(
    CodeGenFunction *CGF, const OMPDeclareReductionDecl *D) {
  if (UDRMap.count(D) > 0)
    return;
  llvm::Function *Combiner = emitCombinerOrInitializer(
      CGM, D->getType(), D->getCombiner(),
      cast<VarDecl>(cast<DeclRefExpr>(D->getCombinerIn())->getDecl()),
      cast<VarDecl>(cast<DeclRefExpr>(D->getCombinerOut())->getDecl()),
      /*IsCombiner=*/true);
  llvm::Function *Initializer = nullptr;
  if (const Expr *Init = D->getInitializer()) {
    Initializer = emitCombinerOrInitializer(
        CGM, D->getType(),
        D->getInitializerKind() == OMPDeclareReductionDecl::CallInit ? Init
                                                                     : nullptr,
        cast<VarDecl>(cast<DeclRefExpr>(D->getInitOrig())->getDecl()),
        cast<VarDecl>(cast<DeclRefExpr>(D->getInitPriv())->getDecl()),
        /*IsCombiner=*/false);
  }
  UDRMap.try_emplace(D, Combiner, Initializer);
  if (CGF) {
    auto &Decls = FunctionUDRMap.FindAndConstruct(CGF->CurFn);
    Decls.second.push_back(D);
  }
}

std::pair<llvm::Function *, llvm::Function *>
CGOpenMPRuntime::getUserDefinedReduction(const OMPDeclareReductionDecl *D) {
  auto I = UDRMap.find(D);
  if (I != UDRMap.end())
    return I->second;
  emitUserDefinedReduction(/*CGF=*/nullptr, D);
  return UDRMap.lookup(D);
}

static llvm::Function *emitParallelOrTeamsOutlinedFunction(
    CodeGenModule &CGM, const OMPExecutableDirective &D, const CapturedStmt *CS,
    const VarDecl *ThreadIDVar, OpenMPDirectiveKind InnermostKind,
    const StringRef OutlinedHelperName, const RegionCodeGenTy &CodeGen) {
  assert(ThreadIDVar->getType()->isPointerType() &&
         "thread id variable must be of type kmp_int32 *");
  CodeGenFunction CGF(CGM, true);
  bool HasCancel = false;
  if (const auto *OPD = dyn_cast<OMPParallelDirective>(&D))
    HasCancel = OPD->hasCancel();
  else if (const auto *OPSD = dyn_cast<OMPParallelSectionsDirective>(&D))
    HasCancel = OPSD->hasCancel();
  else if (const auto *OPFD = dyn_cast<OMPParallelForDirective>(&D))
    HasCancel = OPFD->hasCancel();
  else if (const auto *OPFD = dyn_cast<OMPTargetParallelForDirective>(&D))
    HasCancel = OPFD->hasCancel();
  else if (const auto *OPFD = dyn_cast<OMPDistributeParallelForDirective>(&D))
    HasCancel = OPFD->hasCancel();
  else if (const auto *OPFD =
               dyn_cast<OMPTeamsDistributeParallelForDirective>(&D))
    HasCancel = OPFD->hasCancel();
  else if (const auto *OPFD =
               dyn_cast<OMPTargetTeamsDistributeParallelForDirective>(&D))
    HasCancel = OPFD->hasCancel();
  CGOpenMPOutlinedRegionInfo CGInfo(*CS, ThreadIDVar, CodeGen, InnermostKind,
                                    HasCancel, OutlinedHelperName);
  CodeGenFunction::CGCapturedStmtRAII CapInfoRAII(CGF, &CGInfo);
  return CGF.GenerateOpenMPCapturedStmtFunction(*CS);
}

llvm::Function *CGOpenMPRuntime::emitParallelOutlinedFunction(
    const OMPExecutableDirective &D, const VarDecl *ThreadIDVar,
    OpenMPDirectiveKind InnermostKind, const RegionCodeGenTy &CodeGen) {
  const CapturedStmt *CS = D.getCapturedStmt(OMPD_parallel);
  return emitParallelOrTeamsOutlinedFunction(
      CGM, D, CS, ThreadIDVar, InnermostKind, getOutlinedHelperName(), CodeGen);
}

llvm::Function *CGOpenMPRuntime::emitTeamsOutlinedFunction(
    const OMPExecutableDirective &D, const VarDecl *ThreadIDVar,
    OpenMPDirectiveKind InnermostKind, const RegionCodeGenTy &CodeGen) {
  const CapturedStmt *CS = D.getCapturedStmt(OMPD_teams);
  return emitParallelOrTeamsOutlinedFunction(
      CGM, D, CS, ThreadIDVar, InnermostKind, getOutlinedHelperName(), CodeGen);
}

llvm::Function *CGOpenMPRuntime::emitTaskOutlinedFunction(
    const OMPExecutableDirective &D, const VarDecl *ThreadIDVar,
    const VarDecl *PartIDVar, const VarDecl *TaskTVar,
    OpenMPDirectiveKind InnermostKind, const RegionCodeGenTy &CodeGen,
    bool Tied, unsigned &NumberOfParts) {
  auto &&UntiedCodeGen = [this, &D, TaskTVar](CodeGenFunction &CGF,
                                              PrePostActionTy &) {
    llvm::Value *ThreadID = getThreadID(CGF, D.getBeginLoc());
    llvm::Value *UpLoc = emitUpdateLocation(CGF, D.getBeginLoc());
    llvm::Value *TaskArgs[] = {
        UpLoc, ThreadID,
        CGF.EmitLoadOfPointerLValue(CGF.GetAddrOfLocalVar(TaskTVar),
                                    TaskTVar->getType()->castAs<PointerType>())
            .getPointer()};
    CGF.EmitRuntimeCall(createRuntimeFunction(OMPRTL__kmpc_omp_task), TaskArgs);
  };
  CGOpenMPTaskOutlinedRegionInfo::UntiedTaskActionTy Action(Tied, PartIDVar,
                                                            UntiedCodeGen);
  CodeGen.setAction(Action);
  assert(!ThreadIDVar->getType()->isPointerType() &&
         "thread id variable must be of type kmp_int32 for tasks");
  const OpenMPDirectiveKind Region =
      isOpenMPTaskLoopDirective(D.getDirectiveKind()) ? OMPD_taskloop
                                                      : OMPD_task;
  const CapturedStmt *CS = D.getCapturedStmt(Region);
  const auto *TD = dyn_cast<OMPTaskDirective>(&D);
  CodeGenFunction CGF(CGM, true);
  CGOpenMPTaskOutlinedRegionInfo CGInfo(*CS, ThreadIDVar, CodeGen,
                                        InnermostKind,
                                        TD ? TD->hasCancel() : false, Action);
  CodeGenFunction::CGCapturedStmtRAII CapInfoRAII(CGF, &CGInfo);
  llvm::Function *Res = CGF.GenerateCapturedStmtFunction(*CS);
  if (!Tied)
    NumberOfParts = Action.getNumberOfParts();
  return Res;
}

static void buildStructValue(ConstantStructBuilder &Fields, CodeGenModule &CGM,
                             const RecordDecl *RD, const CGRecordLayout &RL,
                             ArrayRef<llvm::Constant *> Data) {
  llvm::StructType *StructTy = RL.getLLVMType();
  unsigned PrevIdx = 0;
  ConstantInitBuilder CIBuilder(CGM);
  auto DI = Data.begin();
  for (const FieldDecl *FD : RD->fields()) {
    unsigned Idx = RL.getLLVMFieldNo(FD);
    // Fill the alignment.
    for (unsigned I = PrevIdx; I < Idx; ++I)
      Fields.add(llvm::Constant::getNullValue(StructTy->getElementType(I)));
    PrevIdx = Idx + 1;
    Fields.add(*DI);
    ++DI;
  }
}

template <class... As>
static llvm::GlobalVariable *
createGlobalStruct(CodeGenModule &CGM, QualType Ty, bool IsConstant,
                   ArrayRef<llvm::Constant *> Data, const Twine &Name,
                   As &&... Args) {
  const auto *RD = cast<RecordDecl>(Ty->getAsTagDecl());
  const CGRecordLayout &RL = CGM.getTypes().getCGRecordLayout(RD);
  ConstantInitBuilder CIBuilder(CGM);
  ConstantStructBuilder Fields = CIBuilder.beginStruct(RL.getLLVMType());
  buildStructValue(Fields, CGM, RD, RL, Data);
  return Fields.finishAndCreateGlobal(
      Name, CGM.getContext().getAlignOfGlobalVarInChars(Ty), IsConstant,
      std::forward<As>(Args)...);
}

template <typename T>
static void
createConstantGlobalStructAndAddToParent(CodeGenModule &CGM, QualType Ty,
                                         ArrayRef<llvm::Constant *> Data,
                                         T &Parent) {
  const auto *RD = cast<RecordDecl>(Ty->getAsTagDecl());
  const CGRecordLayout &RL = CGM.getTypes().getCGRecordLayout(RD);
  ConstantStructBuilder Fields = Parent.beginStruct(RL.getLLVMType());
  buildStructValue(Fields, CGM, RD, RL, Data);
  Fields.finishAndAddTo(Parent);
}

Address CGOpenMPRuntime::getOrCreateDefaultLocation(unsigned Flags) {
  CharUnits Align = CGM.getContext().getTypeAlignInChars(IdentQTy);
  unsigned Reserved2Flags = getDefaultLocationReserved2Flags();
  FlagsTy FlagsKey(Flags, Reserved2Flags);
  llvm::Value *Entry = OpenMPDefaultLocMap.lookup(FlagsKey);
  if (!Entry) {
    if (!DefaultOpenMPPSource) {
      // Initialize default location for psource field of ident_t structure of
      // all ident_t objects. Format is ";file;function;line;column;;".
      // Taken from
      // https://github.com/llvm/llvm-project/blob/master/openmp/runtime/src/kmp_str.cpp
      DefaultOpenMPPSource =
          CGM.GetAddrOfConstantCString(";unknown;unknown;0;0;;").getPointer();
      DefaultOpenMPPSource =
          llvm::ConstantExpr::getBitCast(DefaultOpenMPPSource, CGM.Int8PtrTy);
    }

    llvm::Constant *Data[] = {
        llvm::ConstantInt::getNullValue(CGM.Int32Ty),
        llvm::ConstantInt::get(CGM.Int32Ty, Flags),
        llvm::ConstantInt::get(CGM.Int32Ty, Reserved2Flags),
        llvm::ConstantInt::getNullValue(CGM.Int32Ty), DefaultOpenMPPSource};
    llvm::GlobalValue *DefaultOpenMPLocation =
        createGlobalStruct(CGM, IdentQTy, isDefaultLocationConstant(), Data, "",
                           llvm::GlobalValue::PrivateLinkage);
    DefaultOpenMPLocation->setUnnamedAddr(
        llvm::GlobalValue::UnnamedAddr::Global);

    OpenMPDefaultLocMap[FlagsKey] = Entry = DefaultOpenMPLocation;
  }
  return Address(Entry, Align);
}

void CGOpenMPRuntime::setLocThreadIdInsertPt(CodeGenFunction &CGF,
                                             bool AtCurrentPoint) {
  auto &Elem = OpenMPLocThreadIDMap.FindAndConstruct(CGF.CurFn);
  assert(!Elem.second.ServiceInsertPt && "Insert point is set already.");

  llvm::Value *Undef = llvm::UndefValue::get(CGF.Int32Ty);
  if (AtCurrentPoint) {
    Elem.second.ServiceInsertPt = new llvm::BitCastInst(
        Undef, CGF.Int32Ty, "svcpt", CGF.Builder.GetInsertBlock());
  } else {
    Elem.second.ServiceInsertPt =
        new llvm::BitCastInst(Undef, CGF.Int32Ty, "svcpt");
    Elem.second.ServiceInsertPt->insertAfter(CGF.AllocaInsertPt);
  }
}

void CGOpenMPRuntime::clearLocThreadIdInsertPt(CodeGenFunction &CGF) {
  auto &Elem = OpenMPLocThreadIDMap.FindAndConstruct(CGF.CurFn);
  if (Elem.second.ServiceInsertPt) {
    llvm::Instruction *Ptr = Elem.second.ServiceInsertPt;
    Elem.second.ServiceInsertPt = nullptr;
    Ptr->eraseFromParent();
  }
}

llvm::Value *CGOpenMPRuntime::emitUpdateLocation(CodeGenFunction &CGF,
                                                 SourceLocation Loc,
                                                 unsigned Flags) {
  Flags |= OMP_IDENT_KMPC;
  // If no debug info is generated - return global default location.
  if (CGM.getCodeGenOpts().getDebugInfo() == codegenoptions::NoDebugInfo ||
      Loc.isInvalid())
    return getOrCreateDefaultLocation(Flags).getPointer();

  assert(CGF.CurFn && "No function in current CodeGenFunction.");

  CharUnits Align = CGM.getContext().getTypeAlignInChars(IdentQTy);
  Address LocValue = Address::invalid();
  auto I = OpenMPLocThreadIDMap.find(CGF.CurFn);
  if (I != OpenMPLocThreadIDMap.end())
    LocValue = Address(I->second.DebugLoc, Align);

  // OpenMPLocThreadIDMap may have null DebugLoc and non-null ThreadID, if
  // GetOpenMPThreadID was called before this routine.
  if (!LocValue.isValid()) {
    // Generate "ident_t .kmpc_loc.addr;"
    Address AI = CGF.CreateMemTemp(IdentQTy, ".kmpc_loc.addr");
    auto &Elem = OpenMPLocThreadIDMap.FindAndConstruct(CGF.CurFn);
    Elem.second.DebugLoc = AI.getPointer();
    LocValue = AI;

    if (!Elem.second.ServiceInsertPt)
      setLocThreadIdInsertPt(CGF);
    CGBuilderTy::InsertPointGuard IPG(CGF.Builder);
    CGF.Builder.SetInsertPoint(Elem.second.ServiceInsertPt);
    CGF.Builder.CreateMemCpy(LocValue, getOrCreateDefaultLocation(Flags),
                             CGF.getTypeSize(IdentQTy));
  }

  // char **psource = &.kmpc_loc_<flags>.addr.psource;
  LValue Base = CGF.MakeAddrLValue(LocValue, IdentQTy);
  auto Fields = cast<RecordDecl>(IdentQTy->getAsTagDecl())->field_begin();
  LValue PSource =
      CGF.EmitLValueForField(Base, *std::next(Fields, IdentField_PSource));

  llvm::Value *OMPDebugLoc = OpenMPDebugLocMap.lookup(Loc.getRawEncoding());
  if (OMPDebugLoc == nullptr) {
    SmallString<128> Buffer2;
    llvm::raw_svector_ostream OS2(Buffer2);
    // Build debug location
    PresumedLoc PLoc = CGF.getContext().getSourceManager().getPresumedLoc(Loc);
    OS2 << ";" << PLoc.getFilename() << ";";
    if (const auto *FD = dyn_cast_or_null<FunctionDecl>(CGF.CurFuncDecl))
      OS2 << FD->getQualifiedNameAsString();
    OS2 << ";" << PLoc.getLine() << ";" << PLoc.getColumn() << ";;";
    OMPDebugLoc = CGF.Builder.CreateGlobalStringPtr(OS2.str());
    OpenMPDebugLocMap[Loc.getRawEncoding()] = OMPDebugLoc;
  }
  // *psource = ";<File>;<Function>;<Line>;<Column>;;";
  CGF.EmitStoreOfScalar(OMPDebugLoc, PSource);

  // Our callers always pass this to a runtime function, so for
  // convenience, go ahead and return a naked pointer.
  return LocValue.getPointer();
}

llvm::Value *CGOpenMPRuntime::getThreadID(CodeGenFunction &CGF,
                                          SourceLocation Loc) {
  assert(CGF.CurFn && "No function in current CodeGenFunction.");

  llvm::Value *ThreadID = nullptr;
  // Check whether we've already cached a load of the thread id in this
  // function.
  auto I = OpenMPLocThreadIDMap.find(CGF.CurFn);
  if (I != OpenMPLocThreadIDMap.end()) {
    ThreadID = I->second.ThreadID;
    if (ThreadID != nullptr)
      return ThreadID;
  }
  // If exceptions are enabled, do not use parameter to avoid possible crash.
  if (!CGF.EHStack.requiresLandingPad() || !CGF.getLangOpts().Exceptions ||
      !CGF.getLangOpts().CXXExceptions ||
      CGF.Builder.GetInsertBlock() == CGF.AllocaInsertPt->getParent()) {
    if (auto *OMPRegionInfo =
            dyn_cast_or_null<CGOpenMPRegionInfo>(CGF.CapturedStmtInfo)) {
      if (OMPRegionInfo->getThreadIDVariable()) {
        // Check if this an outlined function with thread id passed as argument.
        LValue LVal = OMPRegionInfo->getThreadIDVariableLValue(CGF);
        ThreadID = CGF.EmitLoadOfScalar(LVal, Loc);
        // If value loaded in entry block, cache it and use it everywhere in
        // function.
        if (CGF.Builder.GetInsertBlock() == CGF.AllocaInsertPt->getParent()) {
          auto &Elem = OpenMPLocThreadIDMap.FindAndConstruct(CGF.CurFn);
          Elem.second.ThreadID = ThreadID;
        }
        return ThreadID;
      }
    }
  }

  // This is not an outlined function region - need to call __kmpc_int32
  // kmpc_global_thread_num(ident_t *loc).
  // Generate thread id value and cache this value for use across the
  // function.
  auto &Elem = OpenMPLocThreadIDMap.FindAndConstruct(CGF.CurFn);
  if (!Elem.second.ServiceInsertPt)
    setLocThreadIdInsertPt(CGF);
  CGBuilderTy::InsertPointGuard IPG(CGF.Builder);
  CGF.Builder.SetInsertPoint(Elem.second.ServiceInsertPt);
  llvm::CallInst *Call = CGF.Builder.CreateCall(
      createRuntimeFunction(OMPRTL__kmpc_global_thread_num),
      emitUpdateLocation(CGF, Loc));
  Call->setCallingConv(CGF.getRuntimeCC());
  Elem.second.ThreadID = Call;
  return Call;
}

void CGOpenMPRuntime::functionFinished(CodeGenFunction &CGF) {
  assert(CGF.CurFn && "No function in current CodeGenFunction.");
  if (OpenMPLocThreadIDMap.count(CGF.CurFn)) {
    clearLocThreadIdInsertPt(CGF);
    OpenMPLocThreadIDMap.erase(CGF.CurFn);
  }
  if (FunctionUDRMap.count(CGF.CurFn) > 0) {
    for(auto *D : FunctionUDRMap[CGF.CurFn])
      UDRMap.erase(D);
    FunctionUDRMap.erase(CGF.CurFn);
  }
}

llvm::Type *CGOpenMPRuntime::getIdentTyPointerTy() {
  return IdentTy->getPointerTo();
}

llvm::Type *CGOpenMPRuntime::getKmpc_MicroPointerTy() {
  if (!Kmpc_MicroTy) {
    // Build void (*kmpc_micro)(kmp_int32 *global_tid, kmp_int32 *bound_tid,...)
    llvm::Type *MicroParams[] = {llvm::PointerType::getUnqual(CGM.Int32Ty),
                                 llvm::PointerType::getUnqual(CGM.Int32Ty)};
    Kmpc_MicroTy = llvm::FunctionType::get(CGM.VoidTy, MicroParams, true);
  }
  return llvm::PointerType::getUnqual(Kmpc_MicroTy);
}

llvm::FunctionCallee CGOpenMPRuntime::createRuntimeFunction(unsigned Function) {
  llvm::FunctionCallee RTLFn = nullptr;
  switch (static_cast<OpenMPRTLFunction>(Function)) {
  case OMPRTL__kmpc_fork_call: {
    // Build void __kmpc_fork_call(ident_t *loc, kmp_int32 argc, kmpc_micro
    // microtask, ...);
    llvm::Type *TypeParams[] = {getIdentTyPointerTy(), CGM.Int32Ty,
                                getKmpc_MicroPointerTy()};
    auto *FnTy =
        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg*/ true);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, "__kmpc_fork_call");
    if (auto *F = dyn_cast<llvm::Function>(RTLFn.getCallee())) {
      if (!F->hasMetadata(llvm::LLVMContext::MD_callback)) {
        llvm::LLVMContext &Ctx = F->getContext();
        llvm::MDBuilder MDB(Ctx);
        // Annotate the callback behavior of the __kmpc_fork_call:
        //  - The callback callee is argument number 2 (microtask).
        //  - The first two arguments of the callback callee are unknown (-1).
        //  - All variadic arguments to the __kmpc_fork_call are passed to the
        //    callback callee.
        F->addMetadata(
            llvm::LLVMContext::MD_callback,
            *llvm::MDNode::get(Ctx, {MDB.createCallbackEncoding(
                                        2, {-1, -1},
                                        /* VarArgsArePassed */ true)}));
      }
    }
    break;
  }
  case OMPRTL__kmpc_global_thread_num: {
    // Build kmp_int32 __kmpc_global_thread_num(ident_t *loc);
    llvm::Type *TypeParams[] = {getIdentTyPointerTy()};
    auto *FnTy =
        llvm::FunctionType::get(CGM.Int32Ty, TypeParams, /*isVarArg*/ false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, "__kmpc_global_thread_num");
    break;
  }
  case OMPRTL__kmpc_threadprivate_cached: {
    // Build void *__kmpc_threadprivate_cached(ident_t *loc,
    // kmp_int32 global_tid, void *data, size_t size, void ***cache);
    llvm::Type *TypeParams[] = {getIdentTyPointerTy(), CGM.Int32Ty,
                                CGM.VoidPtrTy, CGM.SizeTy,
                                CGM.VoidPtrTy->getPointerTo()->getPointerTo()};
    auto *FnTy =
        llvm::FunctionType::get(CGM.VoidPtrTy, TypeParams, /*isVarArg*/ false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, "__kmpc_threadprivate_cached");
    break;
  }
  case OMPRTL__kmpc_critical: {
    // Build void __kmpc_critical(ident_t *loc, kmp_int32 global_tid,
    // kmp_critical_name *crit);
    llvm::Type *TypeParams[] = {
        getIdentTyPointerTy(), CGM.Int32Ty,
        llvm::PointerType::getUnqual(KmpCriticalNameTy)};
    auto *FnTy =
        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg*/ false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, "__kmpc_critical");
    break;
  }
  case OMPRTL__kmpc_critical_with_hint: {
    // Build void __kmpc_critical_with_hint(ident_t *loc, kmp_int32 global_tid,
    // kmp_critical_name *crit, uintptr_t hint);
    llvm::Type *TypeParams[] = {getIdentTyPointerTy(), CGM.Int32Ty,
                                llvm::PointerType::getUnqual(KmpCriticalNameTy),
                                CGM.IntPtrTy};
    auto *FnTy =
        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg*/ false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, "__kmpc_critical_with_hint");
    break;
  }
  case OMPRTL__kmpc_threadprivate_register: {
    // Build void __kmpc_threadprivate_register(ident_t *, void *data,
    // kmpc_ctor ctor, kmpc_cctor cctor, kmpc_dtor dtor);
    // typedef void *(*kmpc_ctor)(void *);
    auto *KmpcCtorTy =
        llvm::FunctionType::get(CGM.VoidPtrTy, CGM.VoidPtrTy,
                                /*isVarArg*/ false)->getPointerTo();
    // typedef void *(*kmpc_cctor)(void *, void *);
    llvm::Type *KmpcCopyCtorTyArgs[] = {CGM.VoidPtrTy, CGM.VoidPtrTy};
    auto *KmpcCopyCtorTy =
        llvm::FunctionType::get(CGM.VoidPtrTy, KmpcCopyCtorTyArgs,
                                /*isVarArg*/ false)
            ->getPointerTo();
    // typedef void (*kmpc_dtor)(void *);
    auto *KmpcDtorTy =
        llvm::FunctionType::get(CGM.VoidTy, CGM.VoidPtrTy, /*isVarArg*/ false)
            ->getPointerTo();
    llvm::Type *FnTyArgs[] = {getIdentTyPointerTy(), CGM.VoidPtrTy, KmpcCtorTy,
                              KmpcCopyCtorTy, KmpcDtorTy};
    auto *FnTy = llvm::FunctionType::get(CGM.VoidTy, FnTyArgs,
                                        /*isVarArg*/ false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, "__kmpc_threadprivate_register");
    break;
  }
  case OMPRTL__kmpc_end_critical: {
    // Build void __kmpc_end_critical(ident_t *loc, kmp_int32 global_tid,
    // kmp_critical_name *crit);
    llvm::Type *TypeParams[] = {
        getIdentTyPointerTy(), CGM.Int32Ty,
        llvm::PointerType::getUnqual(KmpCriticalNameTy)};
    auto *FnTy =
        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg*/ false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, "__kmpc_end_critical");
    break;
  }
  case OMPRTL__kmpc_cancel_barrier: {
    // Build kmp_int32 __kmpc_cancel_barrier(ident_t *loc, kmp_int32
    // global_tid);
    llvm::Type *TypeParams[] = {getIdentTyPointerTy(), CGM.Int32Ty};
    auto *FnTy =
        llvm::FunctionType::get(CGM.Int32Ty, TypeParams, /*isVarArg*/ false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, /*Name*/ "__kmpc_cancel_barrier");
    break;
  }
  case OMPRTL__kmpc_barrier: {
    // Build void __kmpc_barrier(ident_t *loc, kmp_int32 global_tid);
    llvm::Type *TypeParams[] = {getIdentTyPointerTy(), CGM.Int32Ty};
    auto *FnTy =
        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg*/ false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, /*Name*/ "__kmpc_barrier");
    break;
  }
  case OMPRTL__kmpc_for_static_fini: {
    // Build void __kmpc_for_static_fini(ident_t *loc, kmp_int32 global_tid);
    llvm::Type *TypeParams[] = {getIdentTyPointerTy(), CGM.Int32Ty};
    auto *FnTy =
        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg*/ false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, "__kmpc_for_static_fini");
    break;
  }
  case OMPRTL__kmpc_push_num_threads: {
    // Build void __kmpc_push_num_threads(ident_t *loc, kmp_int32 global_tid,
    // kmp_int32 num_threads)
    llvm::Type *TypeParams[] = {getIdentTyPointerTy(), CGM.Int32Ty,
                                CGM.Int32Ty};
    auto *FnTy =
        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg*/ false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, "__kmpc_push_num_threads");
    break;
  }
  case OMPRTL__kmpc_serialized_parallel: {
    // Build void __kmpc_serialized_parallel(ident_t *loc, kmp_int32
    // global_tid);
    llvm::Type *TypeParams[] = {getIdentTyPointerTy(), CGM.Int32Ty};
    auto *FnTy =
        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg*/ false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, "__kmpc_serialized_parallel");
    break;
  }
  case OMPRTL__kmpc_end_serialized_parallel: {
    // Build void __kmpc_end_serialized_parallel(ident_t *loc, kmp_int32
    // global_tid);
    llvm::Type *TypeParams[] = {getIdentTyPointerTy(), CGM.Int32Ty};
    auto *FnTy =
        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg*/ false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, "__kmpc_end_serialized_parallel");
    break;
  }
  case OMPRTL__kmpc_flush: {
    // Build void __kmpc_flush(ident_t *loc);
    llvm::Type *TypeParams[] = {getIdentTyPointerTy()};
    auto *FnTy =
        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg*/ false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, "__kmpc_flush");
    break;
  }
  case OMPRTL__kmpc_master: {
    // Build kmp_int32 __kmpc_master(ident_t *loc, kmp_int32 global_tid);
    llvm::Type *TypeParams[] = {getIdentTyPointerTy(), CGM.Int32Ty};
    auto *FnTy =
        llvm::FunctionType::get(CGM.Int32Ty, TypeParams, /*isVarArg=*/false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, /*Name=*/"__kmpc_master");
    break;
  }
  case OMPRTL__kmpc_end_master: {
    // Build void __kmpc_end_master(ident_t *loc, kmp_int32 global_tid);
    llvm::Type *TypeParams[] = {getIdentTyPointerTy(), CGM.Int32Ty};
    auto *FnTy =
        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg=*/false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, /*Name=*/"__kmpc_end_master");
    break;
  }
  case OMPRTL__kmpc_omp_taskyield: {
    // Build kmp_int32 __kmpc_omp_taskyield(ident_t *, kmp_int32 global_tid,
    // int end_part);
    llvm::Type *TypeParams[] = {getIdentTyPointerTy(), CGM.Int32Ty, CGM.IntTy};
    auto *FnTy =
        llvm::FunctionType::get(CGM.Int32Ty, TypeParams, /*isVarArg=*/false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, /*Name=*/"__kmpc_omp_taskyield");
    break;
  }
  case OMPRTL__kmpc_single: {
    // Build kmp_int32 __kmpc_single(ident_t *loc, kmp_int32 global_tid);
    llvm::Type *TypeParams[] = {getIdentTyPointerTy(), CGM.Int32Ty};
    auto *FnTy =
        llvm::FunctionType::get(CGM.Int32Ty, TypeParams, /*isVarArg=*/false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, /*Name=*/"__kmpc_single");
    break;
  }
  case OMPRTL__kmpc_end_single: {
    // Build void __kmpc_end_single(ident_t *loc, kmp_int32 global_tid);
    llvm::Type *TypeParams[] = {getIdentTyPointerTy(), CGM.Int32Ty};
    auto *FnTy =
        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg=*/false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, /*Name=*/"__kmpc_end_single");
    break;
  }
  case OMPRTL__kmpc_omp_task_alloc: {
    // Build kmp_task_t *__kmpc_omp_task_alloc(ident_t *, kmp_int32 gtid,
    // kmp_int32 flags, size_t sizeof_kmp_task_t, size_t sizeof_shareds,
    // kmp_routine_entry_t *task_entry);
    assert(KmpRoutineEntryPtrTy != nullptr &&
           "Type kmp_routine_entry_t must be created.");
    llvm::Type *TypeParams[] = {getIdentTyPointerTy(), CGM.Int32Ty, CGM.Int32Ty,
                                CGM.SizeTy, CGM.SizeTy, KmpRoutineEntryPtrTy};
    // Return void * and then cast to particular kmp_task_t type.
    auto *FnTy =
        llvm::FunctionType::get(CGM.VoidPtrTy, TypeParams, /*isVarArg=*/false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, /*Name=*/"__kmpc_omp_task_alloc");
    break;
  }
  case OMPRTL__kmpc_omp_task: {
    // Build kmp_int32 __kmpc_omp_task(ident_t *, kmp_int32 gtid, kmp_task_t
    // *new_task);
    llvm::Type *TypeParams[] = {getIdentTyPointerTy(), CGM.Int32Ty,
                                CGM.VoidPtrTy};
    auto *FnTy =
        llvm::FunctionType::get(CGM.Int32Ty, TypeParams, /*isVarArg=*/false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, /*Name=*/"__kmpc_omp_task");
    break;
  }
  case OMPRTL__kmpc_copyprivate: {
    // Build void __kmpc_copyprivate(ident_t *loc, kmp_int32 global_tid,
    // size_t cpy_size, void *cpy_data, void(*cpy_func)(void *, void *),
    // kmp_int32 didit);
    llvm::Type *CpyTypeParams[] = {CGM.VoidPtrTy, CGM.VoidPtrTy};
    auto *CpyFnTy =
        llvm::FunctionType::get(CGM.VoidTy, CpyTypeParams, /*isVarArg=*/false);
    llvm::Type *TypeParams[] = {getIdentTyPointerTy(), CGM.Int32Ty, CGM.SizeTy,
                                CGM.VoidPtrTy, CpyFnTy->getPointerTo(),
                                CGM.Int32Ty};
    auto *FnTy =
        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg=*/false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, /*Name=*/"__kmpc_copyprivate");
    break;
  }
  case OMPRTL__kmpc_reduce: {
    // Build kmp_int32 __kmpc_reduce(ident_t *loc, kmp_int32 global_tid,
    // kmp_int32 num_vars, size_t reduce_size, void *reduce_data, void
    // (*reduce_func)(void *lhs_data, void *rhs_data), kmp_critical_name *lck);
    llvm::Type *ReduceTypeParams[] = {CGM.VoidPtrTy, CGM.VoidPtrTy};
    auto *ReduceFnTy = llvm::FunctionType::get(CGM.VoidTy, ReduceTypeParams,
                                               /*isVarArg=*/false);
    llvm::Type *TypeParams[] = {
        getIdentTyPointerTy(), CGM.Int32Ty, CGM.Int32Ty, CGM.SizeTy,
        CGM.VoidPtrTy, ReduceFnTy->getPointerTo(),
        llvm::PointerType::getUnqual(KmpCriticalNameTy)};
    auto *FnTy =
        llvm::FunctionType::get(CGM.Int32Ty, TypeParams, /*isVarArg=*/false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, /*Name=*/"__kmpc_reduce");
    break;
  }
  case OMPRTL__kmpc_reduce_nowait: {
    // Build kmp_int32 __kmpc_reduce_nowait(ident_t *loc, kmp_int32
    // global_tid, kmp_int32 num_vars, size_t reduce_size, void *reduce_data,
    // void (*reduce_func)(void *lhs_data, void *rhs_data), kmp_critical_name
    // *lck);
    llvm::Type *ReduceTypeParams[] = {CGM.VoidPtrTy, CGM.VoidPtrTy};
    auto *ReduceFnTy = llvm::FunctionType::get(CGM.VoidTy, ReduceTypeParams,
                                               /*isVarArg=*/false);
    llvm::Type *TypeParams[] = {
        getIdentTyPointerTy(), CGM.Int32Ty, CGM.Int32Ty, CGM.SizeTy,
        CGM.VoidPtrTy, ReduceFnTy->getPointerTo(),
        llvm::PointerType::getUnqual(KmpCriticalNameTy)};
    auto *FnTy =
        llvm::FunctionType::get(CGM.Int32Ty, TypeParams, /*isVarArg=*/false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, /*Name=*/"__kmpc_reduce_nowait");
    break;
  }
  case OMPRTL__kmpc_end_reduce: {
    // Build void __kmpc_end_reduce(ident_t *loc, kmp_int32 global_tid,
    // kmp_critical_name *lck);
    llvm::Type *TypeParams[] = {
        getIdentTyPointerTy(), CGM.Int32Ty,
        llvm::PointerType::getUnqual(KmpCriticalNameTy)};
    auto *FnTy =
        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg=*/false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, /*Name=*/"__kmpc_end_reduce");
    break;
  }
  case OMPRTL__kmpc_end_reduce_nowait: {
    // Build __kmpc_end_reduce_nowait(ident_t *loc, kmp_int32 global_tid,
    // kmp_critical_name *lck);
    llvm::Type *TypeParams[] = {
        getIdentTyPointerTy(), CGM.Int32Ty,
        llvm::PointerType::getUnqual(KmpCriticalNameTy)};
    auto *FnTy =
        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg=*/false);
    RTLFn =
        CGM.CreateRuntimeFunction(FnTy, /*Name=*/"__kmpc_end_reduce_nowait");
    break;
  }
  case OMPRTL__kmpc_omp_task_begin_if0: {
    // Build void __kmpc_omp_task(ident_t *, kmp_int32 gtid, kmp_task_t
    // *new_task);
    llvm::Type *TypeParams[] = {getIdentTyPointerTy(), CGM.Int32Ty,
                                CGM.VoidPtrTy};
    auto *FnTy =
        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg=*/false);
    RTLFn =
        CGM.CreateRuntimeFunction(FnTy, /*Name=*/"__kmpc_omp_task_begin_if0");
    break;
  }
  case OMPRTL__kmpc_omp_task_complete_if0: {
    // Build void __kmpc_omp_task(ident_t *, kmp_int32 gtid, kmp_task_t
    // *new_task);
    llvm::Type *TypeParams[] = {getIdentTyPointerTy(), CGM.Int32Ty,
                                CGM.VoidPtrTy};
    auto *FnTy =
        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg=*/false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy,
                                      /*Name=*/"__kmpc_omp_task_complete_if0");
    break;
  }
  case OMPRTL__kmpc_ordered: {
    // Build void __kmpc_ordered(ident_t *loc, kmp_int32 global_tid);
    llvm::Type *TypeParams[] = {getIdentTyPointerTy(), CGM.Int32Ty};
    auto *FnTy =
        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg=*/false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, "__kmpc_ordered");
    break;
  }
  case OMPRTL__kmpc_end_ordered: {
    // Build void __kmpc_end_ordered(ident_t *loc, kmp_int32 global_tid);
    llvm::Type *TypeParams[] = {getIdentTyPointerTy(), CGM.Int32Ty};
    auto *FnTy =
        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg=*/false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, "__kmpc_end_ordered");
    break;
  }
  case OMPRTL__kmpc_omp_taskwait: {
    // Build kmp_int32 __kmpc_omp_taskwait(ident_t *loc, kmp_int32 global_tid);
    llvm::Type *TypeParams[] = {getIdentTyPointerTy(), CGM.Int32Ty};
    auto *FnTy =
        llvm::FunctionType::get(CGM.Int32Ty, TypeParams, /*isVarArg=*/false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, "__kmpc_omp_taskwait");
    break;
  }
  case OMPRTL__kmpc_taskgroup: {
    // Build void __kmpc_taskgroup(ident_t *loc, kmp_int32 global_tid);
    llvm::Type *TypeParams[] = {getIdentTyPointerTy(), CGM.Int32Ty};
    auto *FnTy =
        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg=*/false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, "__kmpc_taskgroup");
    break;
  }
  case OMPRTL__kmpc_end_taskgroup: {
    // Build void __kmpc_end_taskgroup(ident_t *loc, kmp_int32 global_tid);
    llvm::Type *TypeParams[] = {getIdentTyPointerTy(), CGM.Int32Ty};
    auto *FnTy =
        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg=*/false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, "__kmpc_end_taskgroup");
    break;
  }
  case OMPRTL__kmpc_push_proc_bind: {
    // Build void __kmpc_push_proc_bind(ident_t *loc, kmp_int32 global_tid,
    // int proc_bind)
    llvm::Type *TypeParams[] = {getIdentTyPointerTy(), CGM.Int32Ty, CGM.IntTy};
    auto *FnTy =
        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg*/ false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, "__kmpc_push_proc_bind");
    break;
  }
  case OMPRTL__kmpc_omp_task_with_deps: {
    // Build kmp_int32 __kmpc_omp_task_with_deps(ident_t *, kmp_int32 gtid,
    // kmp_task_t *new_task, kmp_int32 ndeps, kmp_depend_info_t *dep_list,
    // kmp_int32 ndeps_noalias, kmp_depend_info_t *noalias_dep_list);
    llvm::Type *TypeParams[] = {
        getIdentTyPointerTy(), CGM.Int32Ty, CGM.VoidPtrTy, CGM.Int32Ty,
        CGM.VoidPtrTy,         CGM.Int32Ty, CGM.VoidPtrTy};
    auto *FnTy =
        llvm::FunctionType::get(CGM.Int32Ty, TypeParams, /*isVarArg=*/false);
    RTLFn =
        CGM.CreateRuntimeFunction(FnTy, /*Name=*/"__kmpc_omp_task_with_deps");
    break;
  }
  case OMPRTL__kmpc_omp_wait_deps: {
    // Build void __kmpc_omp_wait_deps(ident_t *, kmp_int32 gtid,
    // kmp_int32 ndeps, kmp_depend_info_t *dep_list, kmp_int32 ndeps_noalias,
    // kmp_depend_info_t *noalias_dep_list);
    llvm::Type *TypeParams[] = {getIdentTyPointerTy(), CGM.Int32Ty,
                                CGM.Int32Ty,           CGM.VoidPtrTy,
                                CGM.Int32Ty,           CGM.VoidPtrTy};
    auto *FnTy =
        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg=*/false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, /*Name=*/"__kmpc_omp_wait_deps");
    break;
  }
  case OMPRTL__kmpc_cancellationpoint: {
    // Build kmp_int32 __kmpc_cancellationpoint(ident_t *loc, kmp_int32
    // global_tid, kmp_int32 cncl_kind)
    llvm::Type *TypeParams[] = {getIdentTyPointerTy(), CGM.Int32Ty, CGM.IntTy};
    auto *FnTy =
        llvm::FunctionType::get(CGM.Int32Ty, TypeParams, /*isVarArg*/ false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, "__kmpc_cancellationpoint");
    break;
  }
  case OMPRTL__kmpc_cancel: {
    // Build kmp_int32 __kmpc_cancel(ident_t *loc, kmp_int32 global_tid,
    // kmp_int32 cncl_kind)
    llvm::Type *TypeParams[] = {getIdentTyPointerTy(), CGM.Int32Ty, CGM.IntTy};
    auto *FnTy =
        llvm::FunctionType::get(CGM.Int32Ty, TypeParams, /*isVarArg*/ false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, "__kmpc_cancel");
    break;
  }
  case OMPRTL__kmpc_push_num_teams: {
    // Build void kmpc_push_num_teams (ident_t loc, kmp_int32 global_tid,
    // kmp_int32 num_teams, kmp_int32 num_threads)
    llvm::Type *TypeParams[] = {getIdentTyPointerTy(), CGM.Int32Ty, CGM.Int32Ty,
        CGM.Int32Ty};
    auto *FnTy =
        llvm::FunctionType::get(CGM.Int32Ty, TypeParams, /*isVarArg*/ false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, "__kmpc_push_num_teams");
    break;
  }
  case OMPRTL__kmpc_fork_teams: {
    // Build void __kmpc_fork_teams(ident_t *loc, kmp_int32 argc, kmpc_micro
    // microtask, ...);
    llvm::Type *TypeParams[] = {getIdentTyPointerTy(), CGM.Int32Ty,
                                getKmpc_MicroPointerTy()};
    auto *FnTy =
        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg*/ true);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, "__kmpc_fork_teams");
    if (auto *F = dyn_cast<llvm::Function>(RTLFn.getCallee())) {
      if (!F->hasMetadata(llvm::LLVMContext::MD_callback)) {
        llvm::LLVMContext &Ctx = F->getContext();
        llvm::MDBuilder MDB(Ctx);
        // Annotate the callback behavior of the __kmpc_fork_teams:
        //  - The callback callee is argument number 2 (microtask).
        //  - The first two arguments of the callback callee are unknown (-1).
        //  - All variadic arguments to the __kmpc_fork_teams are passed to the
        //    callback callee.
        F->addMetadata(
            llvm::LLVMContext::MD_callback,
            *llvm::MDNode::get(Ctx, {MDB.createCallbackEncoding(
                                        2, {-1, -1},
                                        /* VarArgsArePassed */ true)}));
      }
    }
    break;
  }
  case OMPRTL__kmpc_taskloop: {
    // Build void __kmpc_taskloop(ident_t *loc, int gtid, kmp_task_t *task, int
    // if_val, kmp_uint64 *lb, kmp_uint64 *ub, kmp_int64 st, int nogroup, int
    // sched, kmp_uint64 grainsize, void *task_dup);
    llvm::Type *TypeParams[] = {getIdentTyPointerTy(),
                                CGM.IntTy,
                                CGM.VoidPtrTy,
                                CGM.IntTy,
                                CGM.Int64Ty->getPointerTo(),
                                CGM.Int64Ty->getPointerTo(),
                                CGM.Int64Ty,
                                CGM.IntTy,
                                CGM.IntTy,
                                CGM.Int64Ty,
                                CGM.VoidPtrTy};
    auto *FnTy =
        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg=*/false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, /*Name=*/"__kmpc_taskloop");
    break;
  }
  case OMPRTL__kmpc_doacross_init: {
    // Build void __kmpc_doacross_init(ident_t *loc, kmp_int32 gtid, kmp_int32
    // num_dims, struct kmp_dim *dims);
    llvm::Type *TypeParams[] = {getIdentTyPointerTy(),
                                CGM.Int32Ty,
                                CGM.Int32Ty,
                                CGM.VoidPtrTy};
    auto *FnTy =
        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg=*/false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, /*Name=*/"__kmpc_doacross_init");
    break;
  }
  case OMPRTL__kmpc_doacross_fini: {
    // Build void __kmpc_doacross_fini(ident_t *loc, kmp_int32 gtid);
    llvm::Type *TypeParams[] = {getIdentTyPointerTy(), CGM.Int32Ty};
    auto *FnTy =
        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg=*/false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, /*Name=*/"__kmpc_doacross_fini");
    break;
  }
  case OMPRTL__kmpc_doacross_post: {
    // Build void __kmpc_doacross_post(ident_t *loc, kmp_int32 gtid, kmp_int64
    // *vec);
    llvm::Type *TypeParams[] = {getIdentTyPointerTy(), CGM.Int32Ty,
                                CGM.Int64Ty->getPointerTo()};
    auto *FnTy =
        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg=*/false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, /*Name=*/"__kmpc_doacross_post");
    break;
  }
  case OMPRTL__kmpc_doacross_wait: {
    // Build void __kmpc_doacross_wait(ident_t *loc, kmp_int32 gtid, kmp_int64
    // *vec);
    llvm::Type *TypeParams[] = {getIdentTyPointerTy(), CGM.Int32Ty,
                                CGM.Int64Ty->getPointerTo()};
    auto *FnTy =
        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg=*/false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, /*Name=*/"__kmpc_doacross_wait");
    break;
  }
  case OMPRTL__kmpc_task_reduction_init: {
    // Build void *__kmpc_task_reduction_init(int gtid, int num_data, void
    // *data);
    llvm::Type *TypeParams[] = {CGM.IntTy, CGM.IntTy, CGM.VoidPtrTy};
    auto *FnTy =
        llvm::FunctionType::get(CGM.VoidPtrTy, TypeParams, /*isVarArg=*/false);
    RTLFn =
        CGM.CreateRuntimeFunction(FnTy, /*Name=*/"__kmpc_task_reduction_init");
    break;
  }
  case OMPRTL__kmpc_task_reduction_get_th_data: {
    // Build void *__kmpc_task_reduction_get_th_data(int gtid, void *tg, void
    // *d);
    llvm::Type *TypeParams[] = {CGM.IntTy, CGM.VoidPtrTy, CGM.VoidPtrTy};
    auto *FnTy =
        llvm::FunctionType::get(CGM.VoidPtrTy, TypeParams, /*isVarArg=*/false);
    RTLFn = CGM.CreateRuntimeFunction(
        FnTy, /*Name=*/"__kmpc_task_reduction_get_th_data");
    break;
  }
  case OMPRTL__kmpc_push_target_tripcount: {
    // Build void __kmpc_push_target_tripcount(int64_t device_id, kmp_uint64
    // size);
    llvm::Type *TypeParams[] = {CGM.Int64Ty, CGM.Int64Ty};
    llvm::FunctionType *FnTy =
        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg=*/false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, "__kmpc_push_target_tripcount");
    break;
  }
  case OMPRTL__tgt_target: {
    // Build int32_t __tgt_target(int64_t device_id, void *host_ptr, int32_t
    // arg_num, void** args_base, void **args, size_t *arg_sizes, int64_t
    // *arg_types);
    llvm::Type *TypeParams[] = {CGM.Int64Ty,
                                CGM.VoidPtrTy,
                                CGM.Int32Ty,
                                CGM.VoidPtrPtrTy,
                                CGM.VoidPtrPtrTy,
                                CGM.SizeTy->getPointerTo(),
                                CGM.Int64Ty->getPointerTo()};
    auto *FnTy =
        llvm::FunctionType::get(CGM.Int32Ty, TypeParams, /*isVarArg*/ false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, "__tgt_target");
    break;
  }
  case OMPRTL__tgt_target_nowait: {
    // Build int32_t __tgt_target_nowait(int64_t device_id, void *host_ptr,
    // int32_t arg_num, void** args_base, void **args, size_t *arg_sizes,
    // int64_t *arg_types);
    llvm::Type *TypeParams[] = {CGM.Int64Ty,
                                CGM.VoidPtrTy,
                                CGM.Int32Ty,
                                CGM.VoidPtrPtrTy,
                                CGM.VoidPtrPtrTy,
                                CGM.SizeTy->getPointerTo(),
                                CGM.Int64Ty->getPointerTo()};
    auto *FnTy =
        llvm::FunctionType::get(CGM.Int32Ty, TypeParams, /*isVarArg*/ false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, "__tgt_target_nowait");
    break;
  }
  case OMPRTL__tgt_target_teams: {
    // Build int32_t __tgt_target_teams(int64_t device_id, void *host_ptr,
    // int32_t arg_num, void** args_base, void **args, size_t *arg_sizes,
    // int64_t *arg_types, int32_t num_teams, int32_t thread_limit);
    llvm::Type *TypeParams[] = {CGM.Int64Ty,
                                CGM.VoidPtrTy,
                                CGM.Int32Ty,
                                CGM.VoidPtrPtrTy,
                                CGM.VoidPtrPtrTy,
                                CGM.SizeTy->getPointerTo(),
                                CGM.Int64Ty->getPointerTo(),
                                CGM.Int32Ty,
                                CGM.Int32Ty};
    auto *FnTy =
        llvm::FunctionType::get(CGM.Int32Ty, TypeParams, /*isVarArg*/ false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, "__tgt_target_teams");
    break;
  }
  case OMPRTL__tgt_target_teams_nowait: {
    // Build int32_t __tgt_target_teams_nowait(int64_t device_id, void
    // *host_ptr, int32_t arg_num, void** args_base, void **args, size_t
    // *arg_sizes, int64_t *arg_types, int32_t num_teams, int32_t thread_limit);
    llvm::Type *TypeParams[] = {CGM.Int64Ty,
                                CGM.VoidPtrTy,
                                CGM.Int32Ty,
                                CGM.VoidPtrPtrTy,
                                CGM.VoidPtrPtrTy,
                                CGM.SizeTy->getPointerTo(),
                                CGM.Int64Ty->getPointerTo(),
                                CGM.Int32Ty,
                                CGM.Int32Ty};
    auto *FnTy =
        llvm::FunctionType::get(CGM.Int32Ty, TypeParams, /*isVarArg*/ false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, "__tgt_target_teams_nowait");
    break;
  }
  case OMPRTL__tgt_register_lib: {
    // Build void __tgt_register_lib(__tgt_bin_desc *desc);
    QualType ParamTy =
        CGM.getContext().getPointerType(getTgtBinaryDescriptorQTy());
    llvm::Type *TypeParams[] = {CGM.getTypes().ConvertTypeForMem(ParamTy)};
    auto *FnTy =
        llvm::FunctionType::get(CGM.Int32Ty, TypeParams, /*isVarArg*/ false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, "__tgt_register_lib");
    break;
  }
  case OMPRTL__tgt_unregister_lib: {
    // Build void __tgt_unregister_lib(__tgt_bin_desc *desc);
    QualType ParamTy =
        CGM.getContext().getPointerType(getTgtBinaryDescriptorQTy());
    llvm::Type *TypeParams[] = {CGM.getTypes().ConvertTypeForMem(ParamTy)};
    auto *FnTy =
        llvm::FunctionType::get(CGM.Int32Ty, TypeParams, /*isVarArg*/ false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, "__tgt_unregister_lib");
    break;
  }
  case OMPRTL__tgt_target_data_begin: {
    // Build void __tgt_target_data_begin(int64_t device_id, int32_t arg_num,
    // void** args_base, void **args, size_t *arg_sizes, int64_t *arg_types);
    llvm::Type *TypeParams[] = {CGM.Int64Ty,
                                CGM.Int32Ty,
                                CGM.VoidPtrPtrTy,
                                CGM.VoidPtrPtrTy,
                                CGM.SizeTy->getPointerTo(),
                                CGM.Int64Ty->getPointerTo()};
    auto *FnTy =
        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg*/ false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, "__tgt_target_data_begin");
    break;
  }
  case OMPRTL__tgt_target_data_begin_nowait: {
    // Build void __tgt_target_data_begin_nowait(int64_t device_id, int32_t
    // arg_num, void** args_base, void **args, size_t *arg_sizes, int64_t
    // *arg_types);
    llvm::Type *TypeParams[] = {CGM.Int64Ty,
                                CGM.Int32Ty,
                                CGM.VoidPtrPtrTy,
                                CGM.VoidPtrPtrTy,
                                CGM.SizeTy->getPointerTo(),
                                CGM.Int64Ty->getPointerTo()};
    auto *FnTy =
        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg=*/false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, "__tgt_target_data_begin_nowait");
    break;
  }
  case OMPRTL__tgt_target_data_end: {
    // Build void __tgt_target_data_end(int64_t device_id, int32_t arg_num,
    // void** args_base, void **args, size_t *arg_sizes, int64_t *arg_types);
    llvm::Type *TypeParams[] = {CGM.Int64Ty,
                                CGM.Int32Ty,
                                CGM.VoidPtrPtrTy,
                                CGM.VoidPtrPtrTy,
                                CGM.SizeTy->getPointerTo(),
                                CGM.Int64Ty->getPointerTo()};
    auto *FnTy =
        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg*/ false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, "__tgt_target_data_end");
    break;
  }
  case OMPRTL__tgt_target_data_end_nowait: {
    // Build void __tgt_target_data_end_nowait(int64_t device_id, int32_t
    // arg_num, void** args_base, void **args, size_t *arg_sizes, int64_t
    // *arg_types);
    llvm::Type *TypeParams[] = {CGM.Int64Ty,
                                CGM.Int32Ty,
                                CGM.VoidPtrPtrTy,
                                CGM.VoidPtrPtrTy,
                                CGM.SizeTy->getPointerTo(),
                                CGM.Int64Ty->getPointerTo()};
    auto *FnTy =
        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg=*/false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, "__tgt_target_data_end_nowait");
    break;
  }
  case OMPRTL__tgt_target_data_update: {
    // Build void __tgt_target_data_update(int64_t device_id, int32_t arg_num,
    // void** args_base, void **args, size_t *arg_sizes, int64_t *arg_types);
    llvm::Type *TypeParams[] = {CGM.Int64Ty,
                                CGM.Int32Ty,
                                CGM.VoidPtrPtrTy,
                                CGM.VoidPtrPtrTy,
                                CGM.SizeTy->getPointerTo(),
                                CGM.Int64Ty->getPointerTo()};
    auto *FnTy =
        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg*/ false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, "__tgt_target_data_update");
    break;
  }
  case OMPRTL__tgt_target_data_update_nowait: {
    // Build void __tgt_target_data_update_nowait(int64_t device_id, int32_t
    // arg_num, void** args_base, void **args, size_t *arg_sizes, int64_t
    // *arg_types);
    llvm::Type *TypeParams[] = {CGM.Int64Ty,
                                CGM.Int32Ty,
                                CGM.VoidPtrPtrTy,
                                CGM.VoidPtrPtrTy,
                                CGM.SizeTy->getPointerTo(),
                                CGM.Int64Ty->getPointerTo()};
    auto *FnTy =
        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg=*/false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, "__tgt_target_data_update_nowait");
    break;
  }
  }
  assert(RTLFn && "Unable to find OpenMP runtime function");
  return RTLFn;
}

llvm::FunctionCallee
CGOpenMPRuntime::createForStaticInitFunction(unsigned IVSize, bool IVSigned) {
  assert((IVSize == 32 || IVSize == 64) &&
         "IV size is not compatible with the omp runtime");
  StringRef Name = IVSize == 32 ? (IVSigned ? "__kmpc_for_static_init_4"
                                            : "__kmpc_for_static_init_4u")
                                : (IVSigned ? "__kmpc_for_static_init_8"
                                            : "__kmpc_for_static_init_8u");
  llvm::Type *ITy = IVSize == 32 ? CGM.Int32Ty : CGM.Int64Ty;
  auto *PtrTy = llvm::PointerType::getUnqual(ITy);
  llvm::Type *TypeParams[] = {
    getIdentTyPointerTy(),                     // loc
    CGM.Int32Ty,                               // tid
    CGM.Int32Ty,                               // schedtype
    llvm::PointerType::getUnqual(CGM.Int32Ty), // p_lastiter
    PtrTy,                                     // p_lower
    PtrTy,                                     // p_upper
    PtrTy,                                     // p_stride
    ITy,                                       // incr
    ITy                                        // chunk
  };
  auto *FnTy =
      llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg*/ false);
  return CGM.CreateRuntimeFunction(FnTy, Name);
}

llvm::FunctionCallee
CGOpenMPRuntime::createDispatchInitFunction(unsigned IVSize, bool IVSigned) {
  assert((IVSize == 32 || IVSize == 64) &&
         "IV size is not compatible with the omp runtime");
  StringRef Name =
      IVSize == 32
          ? (IVSigned ? "__kmpc_dispatch_init_4" : "__kmpc_dispatch_init_4u")
          : (IVSigned ? "__kmpc_dispatch_init_8" : "__kmpc_dispatch_init_8u");
  llvm::Type *ITy = IVSize == 32 ? CGM.Int32Ty : CGM.Int64Ty;
  llvm::Type *TypeParams[] = { getIdentTyPointerTy(), // loc
                               CGM.Int32Ty,           // tid
                               CGM.Int32Ty,           // schedtype
                               ITy,                   // lower
                               ITy,                   // upper
                               ITy,                   // stride
                               ITy                    // chunk
  };
  auto *FnTy =
      llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg*/ false);
  return CGM.CreateRuntimeFunction(FnTy, Name);
}

llvm::FunctionCallee
CGOpenMPRuntime::createDispatchFiniFunction(unsigned IVSize, bool IVSigned) {
  assert((IVSize == 32 || IVSize == 64) &&
         "IV size is not compatible with the omp runtime");
  StringRef Name =
      IVSize == 32
          ? (IVSigned ? "__kmpc_dispatch_fini_4" : "__kmpc_dispatch_fini_4u")
          : (IVSigned ? "__kmpc_dispatch_fini_8" : "__kmpc_dispatch_fini_8u");
  llvm::Type *TypeParams[] = {
      getIdentTyPointerTy(), // loc
      CGM.Int32Ty,           // tid
  };
  auto *FnTy =
      llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg=*/false);
  return CGM.CreateRuntimeFunction(FnTy, Name);
}

llvm::FunctionCallee
CGOpenMPRuntime::createDispatchNextFunction(unsigned IVSize, bool IVSigned) {
  assert((IVSize == 32 || IVSize == 64) &&
         "IV size is not compatible with the omp runtime");
  StringRef Name =
      IVSize == 32
          ? (IVSigned ? "__kmpc_dispatch_next_4" : "__kmpc_dispatch_next_4u")
          : (IVSigned ? "__kmpc_dispatch_next_8" : "__kmpc_dispatch_next_8u");
  llvm::Type *ITy = IVSize == 32 ? CGM.Int32Ty : CGM.Int64Ty;
  auto *PtrTy = llvm::PointerType::getUnqual(ITy);
  llvm::Type *TypeParams[] = {
    getIdentTyPointerTy(),                     // loc
    CGM.Int32Ty,                               // tid
    llvm::PointerType::getUnqual(CGM.Int32Ty), // p_lastiter
    PtrTy,                                     // p_lower
    PtrTy,                                     // p_upper
    PtrTy                                      // p_stride
  };
  auto *FnTy =
      llvm::FunctionType::get(CGM.Int32Ty, TypeParams, /*isVarArg*/ false);
  return CGM.CreateRuntimeFunction(FnTy, Name);
}

Address CGOpenMPRuntime::getAddrOfDeclareTargetLink(const VarDecl *VD) {
  if (CGM.getLangOpts().OpenMPSimd)
    return Address::invalid();
  llvm::Optional<OMPDeclareTargetDeclAttr::MapTypeTy> Res =
      OMPDeclareTargetDeclAttr::isDeclareTargetDeclaration(VD);
  if (Res && *Res == OMPDeclareTargetDeclAttr::MT_Link) {
    SmallString<64> PtrName;
    {
      llvm::raw_svector_ostream OS(PtrName);
      OS << CGM.getMangledName(GlobalDecl(VD)) << "_decl_tgt_link_ptr";
    }
    llvm::Value *Ptr = CGM.getModule().getNamedValue(PtrName);
    if (!Ptr) {
      QualType PtrTy = CGM.getContext().getPointerType(VD->getType());
      Ptr = getOrCreateInternalVariable(CGM.getTypes().ConvertTypeForMem(PtrTy),
                                        PtrName);
      if (!CGM.getLangOpts().OpenMPIsDevice) {
        auto *GV = cast<llvm::GlobalVariable>(Ptr);
        GV->setLinkage(llvm::GlobalValue::ExternalLinkage);
        GV->setInitializer(CGM.GetAddrOfGlobal(VD));
      }
      CGM.addUsedGlobal(cast<llvm::GlobalValue>(Ptr));
      registerTargetGlobalVariable(VD, cast<llvm::Constant>(Ptr));
    }
    return Address(Ptr, CGM.getContext().getDeclAlign(VD));
  }
  return Address::invalid();
}

llvm::Constant *
CGOpenMPRuntime::getOrCreateThreadPrivateCache(const VarDecl *VD) {
  assert(!CGM.getLangOpts().OpenMPUseTLS ||
         !CGM.getContext().getTargetInfo().isTLSSupported());
  // Lookup the entry, lazily creating it if necessary.
  std::string Suffix = getName({"cache", ""});
  return getOrCreateInternalVariable(
      CGM.Int8PtrPtrTy, Twine(CGM.getMangledName(VD)).concat(Suffix));
}

Address CGOpenMPRuntime::getAddrOfThreadPrivate(CodeGenFunction &CGF,
                                                const VarDecl *VD,
                                                Address VDAddr,
                                                SourceLocation Loc) {
  if (CGM.getLangOpts().OpenMPUseTLS &&
      CGM.getContext().getTargetInfo().isTLSSupported())
    return VDAddr;

  llvm::Type *VarTy = VDAddr.getElementType();
  llvm::Value *Args[] = {emitUpdateLocation(CGF, Loc), getThreadID(CGF, Loc),
                         CGF.Builder.CreatePointerCast(VDAddr.getPointer(),
                                                       CGM.Int8PtrTy),
                         CGM.getSize(CGM.GetTargetTypeStoreSize(VarTy)),
                         getOrCreateThreadPrivateCache(VD)};
  return Address(CGF.EmitRuntimeCall(
      createRuntimeFunction(OMPRTL__kmpc_threadprivate_cached), Args),
                 VDAddr.getAlignment());
}

void CGOpenMPRuntime::emitThreadPrivateVarInit(
    CodeGenFunction &CGF, Address VDAddr, llvm::Value *Ctor,
    llvm::Value *CopyCtor, llvm::Value *Dtor, SourceLocation Loc) {
  // Call kmp_int32 __kmpc_global_thread_num(&loc) to init OpenMP runtime
  // library.
  llvm::Value *OMPLoc = emitUpdateLocation(CGF, Loc);
  CGF.EmitRuntimeCall(createRuntimeFunction(OMPRTL__kmpc_global_thread_num),
                      OMPLoc);
  // Call __kmpc_threadprivate_register(&loc, &var, ctor, cctor/*NULL*/, dtor)
  // to register constructor/destructor for variable.
  llvm::Value *Args[] = {
      OMPLoc, CGF.Builder.CreatePointerCast(VDAddr.getPointer(), CGM.VoidPtrTy),
      Ctor, CopyCtor, Dtor};
  CGF.EmitRuntimeCall(
      createRuntimeFunction(OMPRTL__kmpc_threadprivate_register), Args);
}

llvm::Function *CGOpenMPRuntime::emitThreadPrivateVarDefinition(
    const VarDecl *VD, Address VDAddr, SourceLocation Loc,
    bool PerformInit, CodeGenFunction *CGF) {
  if (CGM.getLangOpts().OpenMPUseTLS &&
      CGM.getContext().getTargetInfo().isTLSSupported())
    return nullptr;

  VD = VD->getDefinition(CGM.getContext());
  if (VD && ThreadPrivateWithDefinition.insert(CGM.getMangledName(VD)).second) {
    QualType ASTTy = VD->getType();

    llvm::Value *Ctor = nullptr, *CopyCtor = nullptr, *Dtor = nullptr;
    const Expr *Init = VD->getAnyInitializer();
    if (CGM.getLangOpts().CPlusPlus && PerformInit) {
      // Generate function that re-emits the declaration's initializer into the
      // threadprivate copy of the variable VD
      CodeGenFunction CtorCGF(CGM);
      FunctionArgList Args;
      ImplicitParamDecl Dst(CGM.getContext(), /*DC=*/nullptr, Loc,
                            /*Id=*/nullptr, CGM.getContext().VoidPtrTy,
                            ImplicitParamDecl::Other);
      Args.push_back(&Dst);

      const auto &FI = CGM.getTypes().arrangeBuiltinFunctionDeclaration(
          CGM.getContext().VoidPtrTy, Args);
      llvm::FunctionType *FTy = CGM.getTypes().GetFunctionType(FI);
      std::string Name = getName({"__kmpc_global_ctor_", ""});
      llvm::Function *Fn =
          CGM.CreateGlobalInitOrDestructFunction(FTy, Name, FI, Loc);
      CtorCGF.StartFunction(GlobalDecl(), CGM.getContext().VoidPtrTy, Fn, FI,
                            Args, Loc, Loc);
      llvm::Value *ArgVal = CtorCGF.EmitLoadOfScalar(
          CtorCGF.GetAddrOfLocalVar(&Dst), /*Volatile=*/false,
          CGM.getContext().VoidPtrTy, Dst.getLocation());
      Address Arg = Address(ArgVal, VDAddr.getAlignment());
      Arg = CtorCGF.Builder.CreateElementBitCast(
          Arg, CtorCGF.ConvertTypeForMem(ASTTy));
      CtorCGF.EmitAnyExprToMem(Init, Arg, Init->getType().getQualifiers(),
                               /*IsInitializer=*/true);
      ArgVal = CtorCGF.EmitLoadOfScalar(
          CtorCGF.GetAddrOfLocalVar(&Dst), /*Volatile=*/false,
          CGM.getContext().VoidPtrTy, Dst.getLocation());
      CtorCGF.Builder.CreateStore(ArgVal, CtorCGF.ReturnValue);
      CtorCGF.FinishFunction();
      Ctor = Fn;
    }
    if (VD->getType().isDestructedType() != QualType::DK_none) {
      // Generate function that emits destructor call for the threadprivate copy
      // of the variable VD
      CodeGenFunction DtorCGF(CGM);
      FunctionArgList Args;
      ImplicitParamDecl Dst(CGM.getContext(), /*DC=*/nullptr, Loc,
                            /*Id=*/nullptr, CGM.getContext().VoidPtrTy,
                            ImplicitParamDecl::Other);
      Args.push_back(&Dst);

      const auto &FI = CGM.getTypes().arrangeBuiltinFunctionDeclaration(
          CGM.getContext().VoidTy, Args);
      llvm::FunctionType *FTy = CGM.getTypes().GetFunctionType(FI);
      std::string Name = getName({"__kmpc_global_dtor_", ""});
      llvm::Function *Fn =
          CGM.CreateGlobalInitOrDestructFunction(FTy, Name, FI, Loc);
      auto NL = ApplyDebugLocation::CreateEmpty(DtorCGF);
      DtorCGF.StartFunction(GlobalDecl(), CGM.getContext().VoidTy, Fn, FI, Args,
                            Loc, Loc);
      // Create a scope with an artificial location for the body of this function.
      auto AL = ApplyDebugLocation::CreateArtificial(DtorCGF);
      llvm::Value *ArgVal = DtorCGF.EmitLoadOfScalar(
          DtorCGF.GetAddrOfLocalVar(&Dst),
          /*Volatile=*/false, CGM.getContext().VoidPtrTy, Dst.getLocation());
      DtorCGF.emitDestroy(Address(ArgVal, VDAddr.getAlignment()), ASTTy,
                          DtorCGF.getDestroyer(ASTTy.isDestructedType()),
                          DtorCGF.needsEHCleanup(ASTTy.isDestructedType()));
      DtorCGF.FinishFunction();
      Dtor = Fn;
    }
    // Do not emit init function if it is not required.
    if (!Ctor && !Dtor)
      return nullptr;

    llvm::Type *CopyCtorTyArgs[] = {CGM.VoidPtrTy, CGM.VoidPtrTy};
    auto *CopyCtorTy = llvm::FunctionType::get(CGM.VoidPtrTy, CopyCtorTyArgs,
                                               /*isVarArg=*/false)
                           ->getPointerTo();
    // Copying constructor for the threadprivate variable.
    // Must be NULL - reserved by runtime, but currently it requires that this
    // parameter is always NULL. Otherwise it fires assertion.
    CopyCtor = llvm::Constant::getNullValue(CopyCtorTy);
    if (Ctor == nullptr) {
      auto *CtorTy = llvm::FunctionType::get(CGM.VoidPtrTy, CGM.VoidPtrTy,
                                             /*isVarArg=*/false)
                         ->getPointerTo();
      Ctor = llvm::Constant::getNullValue(CtorTy);
    }
    if (Dtor == nullptr) {
      auto *DtorTy = llvm::FunctionType::get(CGM.VoidTy, CGM.VoidPtrTy,
                                             /*isVarArg=*/false)
                         ->getPointerTo();
      Dtor = llvm::Constant::getNullValue(DtorTy);
    }
    if (!CGF) {
      auto *InitFunctionTy =
          llvm::FunctionType::get(CGM.VoidTy, /*isVarArg*/ false);
      std::string Name = getName({"__omp_threadprivate_init_", ""});
      llvm::Function *InitFunction = CGM.CreateGlobalInitOrDestructFunction(
          InitFunctionTy, Name, CGM.getTypes().arrangeNullaryFunction());
      CodeGenFunction InitCGF(CGM);
      FunctionArgList ArgList;
      InitCGF.StartFunction(GlobalDecl(), CGM.getContext().VoidTy, InitFunction,
                            CGM.getTypes().arrangeNullaryFunction(), ArgList,
                            Loc, Loc);
      emitThreadPrivateVarInit(InitCGF, VDAddr, Ctor, CopyCtor, Dtor, Loc);
      InitCGF.FinishFunction();
      return InitFunction;
    }
    emitThreadPrivateVarInit(*CGF, VDAddr, Ctor, CopyCtor, Dtor, Loc);
  }
  return nullptr;
}

/// Obtain information that uniquely identifies a target entry. This
/// consists of the file and device IDs as well as line number associated with
/// the relevant entry source location.
static void getTargetEntryUniqueInfo(ASTContext &C, SourceLocation Loc,
                                     unsigned &DeviceID, unsigned &FileID,
                                     unsigned &LineNum) {
  SourceManager &SM = C.getSourceManager();

  // The loc should be always valid and have a file ID (the user cannot use
  // #pragma directives in macros)

  assert(Loc.isValid() && "Source location is expected to be always valid.");

  PresumedLoc PLoc = SM.getPresumedLoc(Loc);
  assert(PLoc.isValid() && "Source location is expected to be always valid.");

  llvm::sys::fs::UniqueID ID;
  if (auto EC = llvm::sys::fs::getUniqueID(PLoc.getFilename(), ID))
    SM.getDiagnostics().Report(diag::err_cannot_open_file)
        << PLoc.getFilename() << EC.message();

  DeviceID = ID.getDevice();
  FileID = ID.getFile();
  LineNum = PLoc.getLine();
}

bool CGOpenMPRuntime::emitDeclareTargetVarDefinition(const VarDecl *VD,
                                                     llvm::GlobalVariable *Addr,
                                                     bool PerformInit) {
  Optional<OMPDeclareTargetDeclAttr::MapTypeTy> Res =
      OMPDeclareTargetDeclAttr::isDeclareTargetDeclaration(VD);
  if (!Res || *Res == OMPDeclareTargetDeclAttr::MT_Link)
    return CGM.getLangOpts().OpenMPIsDevice;
  VD = VD->getDefinition(CGM.getContext());
  if (VD && !DeclareTargetWithDefinition.insert(CGM.getMangledName(VD)).second)
    return CGM.getLangOpts().OpenMPIsDevice;

  QualType ASTTy = VD->getType();

  SourceLocation Loc = VD->getCanonicalDecl()->getBeginLoc();
  // Produce the unique prefix to identify the new target regions. We use
  // the source location of the variable declaration which we know to not
  // conflict with any target region.
  unsigned DeviceID;
  unsigned FileID;
  unsigned Line;
  getTargetEntryUniqueInfo(CGM.getContext(), Loc, DeviceID, FileID, Line);
  SmallString<128> Buffer, Out;
  {
    llvm::raw_svector_ostream OS(Buffer);
    OS << "__omp_offloading_" << llvm::format("_%x", DeviceID)
       << llvm::format("_%x_", FileID) << VD->getName() << "_l" << Line;
  }

  const Expr *Init = VD->getAnyInitializer();
  if (CGM.getLangOpts().CPlusPlus && PerformInit) {
    llvm::Constant *Ctor;
    llvm::Constant *ID;
    if (CGM.getLangOpts().OpenMPIsDevice) {
      // Generate function that re-emits the declaration's initializer into
      // the threadprivate copy of the variable VD
      CodeGenFunction CtorCGF(CGM);

      const CGFunctionInfo &FI = CGM.getTypes().arrangeNullaryFunction();
      llvm::FunctionType *FTy = CGM.getTypes().GetFunctionType(FI);
      llvm::Function *Fn = CGM.CreateGlobalInitOrDestructFunction(
          FTy, Twine(Buffer, "_ctor"), FI, Loc);
      auto NL = ApplyDebugLocation::CreateEmpty(CtorCGF);
      CtorCGF.StartFunction(GlobalDecl(), CGM.getContext().VoidTy, Fn, FI,
                            FunctionArgList(), Loc, Loc);
      auto AL = ApplyDebugLocation::CreateArtificial(CtorCGF);
      CtorCGF.EmitAnyExprToMem(Init,
                               Address(Addr, CGM.getContext().getDeclAlign(VD)),
                               Init->getType().getQualifiers(),
                               /*IsInitializer=*/true);
      CtorCGF.FinishFunction();
      Ctor = Fn;
      ID = llvm::ConstantExpr::getBitCast(Fn, CGM.Int8PtrTy);
      CGM.addUsedGlobal(cast<llvm::GlobalValue>(Ctor));
    } else {
      Ctor = new llvm::GlobalVariable(
          CGM.getModule(), CGM.Int8Ty, /*isConstant=*/true,
          llvm::GlobalValue::PrivateLinkage,
          llvm::Constant::getNullValue(CGM.Int8Ty), Twine(Buffer, "_ctor"));
      ID = Ctor;
    }

    // Register the information for the entry associated with the constructor.
    Out.clear();
    OffloadEntriesInfoManager.registerTargetRegionEntryInfo(
        DeviceID, FileID, Twine(Buffer, "_ctor").toStringRef(Out), Line, Ctor,
        ID, OffloadEntriesInfoManagerTy::OMPTargetRegionEntryCtor);
  }
  if (VD->getType().isDestructedType() != QualType::DK_none) {
    llvm::Constant *Dtor;
    llvm::Constant *ID;
    if (CGM.getLangOpts().OpenMPIsDevice) {
      // Generate function that emits destructor call for the threadprivate
      // copy of the variable VD
      CodeGenFunction DtorCGF(CGM);

      const CGFunctionInfo &FI = CGM.getTypes().arrangeNullaryFunction();
      llvm::FunctionType *FTy = CGM.getTypes().GetFunctionType(FI);
      llvm::Function *Fn = CGM.CreateGlobalInitOrDestructFunction(
          FTy, Twine(Buffer, "_dtor"), FI, Loc);
      auto NL = ApplyDebugLocation::CreateEmpty(DtorCGF);
      DtorCGF.StartFunction(GlobalDecl(), CGM.getContext().VoidTy, Fn, FI,
                            FunctionArgList(), Loc, Loc);
      // Create a scope with an artificial location for the body of this
      // function.
      auto AL = ApplyDebugLocation::CreateArtificial(DtorCGF);
      DtorCGF.emitDestroy(Address(Addr, CGM.getContext().getDeclAlign(VD)),
                          ASTTy, DtorCGF.getDestroyer(ASTTy.isDestructedType()),
                          DtorCGF.needsEHCleanup(ASTTy.isDestructedType()));
      DtorCGF.FinishFunction();
      Dtor = Fn;
      ID = llvm::ConstantExpr::getBitCast(Fn, CGM.Int8PtrTy);
      CGM.addUsedGlobal(cast<llvm::GlobalValue>(Dtor));
    } else {
      Dtor = new llvm::GlobalVariable(
          CGM.getModule(), CGM.Int8Ty, /*isConstant=*/true,
          llvm::GlobalValue::PrivateLinkage,
          llvm::Constant::getNullValue(CGM.Int8Ty), Twine(Buffer, "_dtor"));
      ID = Dtor;
    }
    // Register the information for the entry associated with the destructor.
    Out.clear();
    OffloadEntriesInfoManager.registerTargetRegionEntryInfo(
        DeviceID, FileID, Twine(Buffer, "_dtor").toStringRef(Out), Line, Dtor,
        ID, OffloadEntriesInfoManagerTy::OMPTargetRegionEntryDtor);
  }
  return CGM.getLangOpts().OpenMPIsDevice;
}

Address CGOpenMPRuntime::getAddrOfArtificialThreadPrivate(CodeGenFunction &CGF,
                                                          QualType VarType,
                                                          StringRef Name) {
  std::string Suffix = getName({"artificial", ""});
  std::string CacheSuffix = getName({"cache", ""});
  llvm::Type *VarLVType = CGF.ConvertTypeForMem(VarType);
  llvm::Value *GAddr =
      getOrCreateInternalVariable(VarLVType, Twine(Name).concat(Suffix));
  llvm::Value *Args[] = {
      emitUpdateLocation(CGF, SourceLocation()),
      getThreadID(CGF, SourceLocation()),
      CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(GAddr, CGM.VoidPtrTy),
      CGF.Builder.CreateIntCast(CGF.getTypeSize(VarType), CGM.SizeTy,
                                /*IsSigned=*/false),
      getOrCreateInternalVariable(
          CGM.VoidPtrPtrTy, Twine(Name).concat(Suffix).concat(CacheSuffix))};
  return Address(
      CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
          CGF.EmitRuntimeCall(
              createRuntimeFunction(OMPRTL__kmpc_threadprivate_cached), Args),
          VarLVType->getPointerTo(/*AddrSpace=*/0)),
      CGM.getPointerAlign());
}

void CGOpenMPRuntime::emitOMPIfClause(CodeGenFunction &CGF, const Expr *Cond,
                                      const RegionCodeGenTy &ThenGen,
                                      const RegionCodeGenTy &ElseGen) {
  CodeGenFunction::LexicalScope ConditionScope(CGF, Cond->getSourceRange());

  // If the condition constant folds and can be elided, try to avoid emitting
  // the condition and the dead arm of the if/else.
  bool CondConstant;
  if (CGF.ConstantFoldsToSimpleInteger(Cond, CondConstant)) {
    if (CondConstant)
      ThenGen(CGF);
    else
      ElseGen(CGF);
    return;
  }

  // Otherwise, the condition did not fold, or we couldn't elide it.  Just
  // emit the conditional branch.
  llvm::BasicBlock *ThenBlock = CGF.createBasicBlock("omp_if.then");
  llvm::BasicBlock *ElseBlock = CGF.createBasicBlock("omp_if.else");
  llvm::BasicBlock *ContBlock = CGF.createBasicBlock("omp_if.end");
  CGF.EmitBranchOnBoolExpr(Cond, ThenBlock, ElseBlock, /*TrueCount=*/0);

  // Emit the 'then' code.
  CGF.EmitBlock(ThenBlock);
  ThenGen(CGF);
  CGF.EmitBranch(ContBlock);
  // Emit the 'else' code if present.
  // There is no need to emit line number for unconditional branch.
  (void)ApplyDebugLocation::CreateEmpty(CGF);
  CGF.EmitBlock(ElseBlock);
  ElseGen(CGF);
  // There is no need to emit line number for unconditional branch.
  (void)ApplyDebugLocation::CreateEmpty(CGF);
  CGF.EmitBranch(ContBlock);
  // Emit the continuation block for code after the if.
  CGF.EmitBlock(ContBlock, /*IsFinished=*/true);
}

void CGOpenMPRuntime::emitParallelCall(CodeGenFunction &CGF, SourceLocation Loc,
                                       llvm::Function *OutlinedFn,
                                       ArrayRef<llvm::Value *> CapturedVars,
                                       const Expr *IfCond) {
  if (!CGF.HaveInsertPoint())
    return;
  llvm::Value *RTLoc = emitUpdateLocation(CGF, Loc);
  auto &&ThenGen = [OutlinedFn, CapturedVars, RTLoc](CodeGenFunction &CGF,
                                                     PrePostActionTy &) {
    // Build call __kmpc_fork_call(loc, n, microtask, var1, .., varn);
    CGOpenMPRuntime &RT = CGF.CGM.getOpenMPRuntime();
    llvm::Value *Args[] = {
        RTLoc,
        CGF.Builder.getInt32(CapturedVars.size()), // Number of captured vars
        CGF.Builder.CreateBitCast(OutlinedFn, RT.getKmpc_MicroPointerTy())};
    llvm::SmallVector<llvm::Value *, 16> RealArgs;
    RealArgs.append(std::begin(Args), std::end(Args));
    RealArgs.append(CapturedVars.begin(), CapturedVars.end());

    llvm::FunctionCallee RTLFn =
        RT.createRuntimeFunction(OMPRTL__kmpc_fork_call);
    CGF.EmitRuntimeCall(RTLFn, RealArgs);
  };
  auto &&ElseGen = [OutlinedFn, CapturedVars, RTLoc, Loc](CodeGenFunction &CGF,
                                                          PrePostActionTy &) {
    CGOpenMPRuntime &RT = CGF.CGM.getOpenMPRuntime();
    llvm::Value *ThreadID = RT.getThreadID(CGF, Loc);
    // Build calls:
    // __kmpc_serialized_parallel(&Loc, GTid);
    llvm::Value *Args[] = {RTLoc, ThreadID};
    CGF.EmitRuntimeCall(
        RT.createRuntimeFunction(OMPRTL__kmpc_serialized_parallel), Args);

    // OutlinedFn(&GTid, &zero, CapturedStruct);
    Address ZeroAddr = CGF.CreateDefaultAlignTempAlloca(CGF.Int32Ty,
                                                        /*Name*/ ".zero.addr");
    CGF.InitTempAlloca(ZeroAddr, CGF.Builder.getInt32(/*C*/ 0));
    llvm::SmallVector<llvm::Value *, 16> OutlinedFnArgs;
    // ThreadId for serialized parallels is 0.
    OutlinedFnArgs.push_back(ZeroAddr.getPointer());
    OutlinedFnArgs.push_back(ZeroAddr.getPointer());
    OutlinedFnArgs.append(CapturedVars.begin(), CapturedVars.end());
    RT.emitOutlinedFunctionCall(CGF, Loc, OutlinedFn, OutlinedFnArgs);

    // __kmpc_end_serialized_parallel(&Loc, GTid);
    llvm::Value *EndArgs[] = {RT.emitUpdateLocation(CGF, Loc), ThreadID};
    CGF.EmitRuntimeCall(
        RT.createRuntimeFunction(OMPRTL__kmpc_end_serialized_parallel),
        EndArgs);
  };
  if (IfCond) {
    emitOMPIfClause(CGF, IfCond, ThenGen, ElseGen);
  } else {
    RegionCodeGenTy ThenRCG(ThenGen);
    ThenRCG(CGF);
  }
}

// If we're inside an (outlined) parallel region, use the region info's
// thread-ID variable (it is passed in a first argument of the outlined function
// as "kmp_int32 *gtid"). Otherwise, if we're not inside parallel region, but in
// regular serial code region, get thread ID by calling kmp_int32
// kmpc_global_thread_num(ident_t *loc), stash this thread ID in a temporary and
// return the address of that temp.
Address CGOpenMPRuntime::emitThreadIDAddress(CodeGenFunction &CGF,
                                             SourceLocation Loc) {
  if (auto *OMPRegionInfo =
          dyn_cast_or_null<CGOpenMPRegionInfo>(CGF.CapturedStmtInfo))
    if (OMPRegionInfo->getThreadIDVariable())
      return OMPRegionInfo->getThreadIDVariableLValue(CGF).getAddress();

  llvm::Value *ThreadID = getThreadID(CGF, Loc);
  QualType Int32Ty =
      CGF.getContext().getIntTypeForBitwidth(/*DestWidth*/ 32, /*Signed*/ true);
  Address ThreadIDTemp = CGF.CreateMemTemp(Int32Ty, /*Name*/ ".threadid_temp.");
  CGF.EmitStoreOfScalar(ThreadID,
                        CGF.MakeAddrLValue(ThreadIDTemp, Int32Ty));

  return ThreadIDTemp;
}

llvm::Constant *
CGOpenMPRuntime::getOrCreateInternalVariable(llvm::Type *Ty,
                                             const llvm::Twine &Name) {
  SmallString<256> Buffer;
  llvm::raw_svector_ostream Out(Buffer);
  Out << Name;
  StringRef RuntimeName = Out.str();
  auto &Elem = *InternalVars.try_emplace(RuntimeName, nullptr).first;
  if (Elem.second) {
    assert(Elem.second->getType()->getPointerElementType() == Ty &&
           "OMP internal variable has different type than requested");
    return &*Elem.second;
  }

  return Elem.second = new llvm::GlobalVariable(
             CGM.getModule(), Ty, /*IsConstant*/ false,
             llvm::GlobalValue::CommonLinkage, llvm::Constant::getNullValue(Ty),
             Elem.first());
}

llvm::Value *CGOpenMPRuntime::getCriticalRegionLock(StringRef CriticalName) {
  std::string Prefix = Twine("gomp_critical_user_", CriticalName).str();
  std::string Name = getName({Prefix, "var"});
  return getOrCreateInternalVariable(KmpCriticalNameTy, Name);
}

namespace {
/// Common pre(post)-action for different OpenMP constructs.
class CommonActionTy final : public PrePostActionTy {
  llvm::FunctionCallee EnterCallee;
  ArrayRef<llvm::Value *> EnterArgs;
  llvm::FunctionCallee ExitCallee;
  ArrayRef<llvm::Value *> ExitArgs;
  bool Conditional;
  llvm::BasicBlock *ContBlock = nullptr;

public:
  CommonActionTy(llvm::FunctionCallee EnterCallee,
                 ArrayRef<llvm::Value *> EnterArgs,
                 llvm::FunctionCallee ExitCallee,
                 ArrayRef<llvm::Value *> ExitArgs, bool Conditional = false)
      : EnterCallee(EnterCallee), EnterArgs(EnterArgs), ExitCallee(ExitCallee),
        ExitArgs(ExitArgs), Conditional(Conditional) {}
  void Enter(CodeGenFunction &CGF) override {
    llvm::Value *EnterRes = CGF.EmitRuntimeCall(EnterCallee, EnterArgs);
    if (Conditional) {
      llvm::Value *CallBool = CGF.Builder.CreateIsNotNull(EnterRes);
      auto *ThenBlock = CGF.createBasicBlock("omp_if.then");
      ContBlock = CGF.createBasicBlock("omp_if.end");
      // Generate the branch (If-stmt)
      CGF.Builder.CreateCondBr(CallBool, ThenBlock, ContBlock);
      CGF.EmitBlock(ThenBlock);
    }
  }
  void Done(CodeGenFunction &CGF) {
    // Emit the rest of blocks/branches
    CGF.EmitBranch(ContBlock);
    CGF.EmitBlock(ContBlock, true);
  }
  void Exit(CodeGenFunction &CGF) override {
    CGF.EmitRuntimeCall(ExitCallee, ExitArgs);
  }
};
} // anonymous namespace

void CGOpenMPRuntime::emitCriticalRegion(CodeGenFunction &CGF,
                                         StringRef CriticalName,
                                         const RegionCodeGenTy &CriticalOpGen,
                                         SourceLocation Loc, const Expr *Hint) {
  // __kmpc_critical[_with_hint](ident_t *, gtid, Lock[, hint]);
  // CriticalOpGen();
  // __kmpc_end_critical(ident_t *, gtid, Lock);
  // Prepare arguments and build a call to __kmpc_critical
  if (!CGF.HaveInsertPoint())
    return;
  llvm::Value *Args[] = {emitUpdateLocation(CGF, Loc), getThreadID(CGF, Loc),
                         getCriticalRegionLock(CriticalName)};
  llvm::SmallVector<llvm::Value *, 4> EnterArgs(std::begin(Args),
                                                std::end(Args));
  if (Hint) {
    EnterArgs.push_back(CGF.Builder.CreateIntCast(
        CGF.EmitScalarExpr(Hint), CGM.IntPtrTy, /*isSigned=*/false));
  }
  CommonActionTy Action(
      createRuntimeFunction(Hint ? OMPRTL__kmpc_critical_with_hint
                                 : OMPRTL__kmpc_critical),
      EnterArgs, createRuntimeFunction(OMPRTL__kmpc_end_critical), Args);
  CriticalOpGen.setAction(Action);
  emitInlinedDirective(CGF, OMPD_critical, CriticalOpGen);
}

void CGOpenMPRuntime::emitMasterRegion(CodeGenFunction &CGF,
                                       const RegionCodeGenTy &MasterOpGen,
                                       SourceLocation Loc) {
  if (!CGF.HaveInsertPoint())
    return;
  // if(__kmpc_master(ident_t *, gtid)) {
  //   MasterOpGen();
  //   __kmpc_end_master(ident_t *, gtid);
  // }
  // Prepare arguments and build a call to __kmpc_master
  llvm::Value *Args[] = {emitUpdateLocation(CGF, Loc), getThreadID(CGF, Loc)};
  CommonActionTy Action(createRuntimeFunction(OMPRTL__kmpc_master), Args,
                        createRuntimeFunction(OMPRTL__kmpc_end_master), Args,
                        /*Conditional=*/true);
  MasterOpGen.setAction(Action);
  emitInlinedDirective(CGF, OMPD_master, MasterOpGen);
  Action.Done(CGF);
}

void CGOpenMPRuntime::emitTaskyieldCall(CodeGenFunction &CGF,
                                        SourceLocation Loc) {
  if (!CGF.HaveInsertPoint())
    return;
  // Build call __kmpc_omp_taskyield(loc, thread_id, 0);
  llvm::Value *Args[] = {
      emitUpdateLocation(CGF, Loc), getThreadID(CGF, Loc),
      llvm::ConstantInt::get(CGM.IntTy, /*V=*/0, /*isSigned=*/true)};
  CGF.EmitRuntimeCall(createRuntimeFunction(OMPRTL__kmpc_omp_taskyield), Args);
  if (auto *Region = dyn_cast_or_null<CGOpenMPRegionInfo>(CGF.CapturedStmtInfo))
    Region->emitUntiedSwitch(CGF);
}

void CGOpenMPRuntime::emitTaskgroupRegion(CodeGenFunction &CGF,
                                          const RegionCodeGenTy &TaskgroupOpGen,
                                          SourceLocation Loc) {
  if (!CGF.HaveInsertPoint())
    return;
  // __kmpc_taskgroup(ident_t *, gtid);
  // TaskgroupOpGen();
  // __kmpc_end_taskgroup(ident_t *, gtid);
  // Prepare arguments and build a call to __kmpc_taskgroup
  llvm::Value *Args[] = {emitUpdateLocation(CGF, Loc), getThreadID(CGF, Loc)};
  CommonActionTy Action(createRuntimeFunction(OMPRTL__kmpc_taskgroup), Args,
                        createRuntimeFunction(OMPRTL__kmpc_end_taskgroup),
                        Args);
  TaskgroupOpGen.setAction(Action);
  emitInlinedDirective(CGF, OMPD_taskgroup, TaskgroupOpGen);
}

/// Given an array of pointers to variables, project the address of a
/// given variable.
static Address emitAddrOfVarFromArray(CodeGenFunction &CGF, Address Array,
                                      unsigned Index, const VarDecl *Var) {
  // Pull out the pointer to the variable.
  Address PtrAddr = CGF.Builder.CreateConstArrayGEP(Array, Index);
  llvm::Value *Ptr = CGF.Builder.CreateLoad(PtrAddr);

  Address Addr = Address(Ptr, CGF.getContext().getDeclAlign(Var));
  Addr = CGF.Builder.CreateElementBitCast(
      Addr, CGF.ConvertTypeForMem(Var->getType()));
  return Addr;
}

static llvm::Value *emitCopyprivateCopyFunction(
    CodeGenModule &CGM, llvm::Type *ArgsType,
    ArrayRef<const Expr *> CopyprivateVars, ArrayRef<const Expr *> DestExprs,
    ArrayRef<const Expr *> SrcExprs, ArrayRef<const Expr *> AssignmentOps,
    SourceLocation Loc) {
  ASTContext &C = CGM.getContext();
  // void copy_func(void *LHSArg, void *RHSArg);
  FunctionArgList Args;
  ImplicitParamDecl LHSArg(C, /*DC=*/nullptr, Loc, /*Id=*/nullptr, C.VoidPtrTy,
                           ImplicitParamDecl::Other);
  ImplicitParamDecl RHSArg(C, /*DC=*/nullptr, Loc, /*Id=*/nullptr, C.VoidPtrTy,
                           ImplicitParamDecl::Other);
  Args.push_back(&LHSArg);
  Args.push_back(&RHSArg);
  const auto &CGFI =
      CGM.getTypes().arrangeBuiltinFunctionDeclaration(C.VoidTy, Args);
  std::string Name =
      CGM.getOpenMPRuntime().getName({"omp", "copyprivate", "copy_func"});
  auto *Fn = llvm::Function::Create(CGM.getTypes().GetFunctionType(CGFI),
                                    llvm::GlobalValue::InternalLinkage, Name,
                                    &CGM.getModule());
  CGM.SetInternalFunctionAttributes(GlobalDecl(), Fn, CGFI);
  Fn->setDoesNotRecurse();
  CodeGenFunction CGF(CGM);
  CGF.StartFunction(GlobalDecl(), C.VoidTy, Fn, CGFI, Args, Loc, Loc);
  // Dest = (void*[n])(LHSArg);
  // Src = (void*[n])(RHSArg);
  Address LHS(CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
      CGF.Builder.CreateLoad(CGF.GetAddrOfLocalVar(&LHSArg)),
      ArgsType), CGF.getPointerAlign());
  Address RHS(CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
      CGF.Builder.CreateLoad(CGF.GetAddrOfLocalVar(&RHSArg)),
      ArgsType), CGF.getPointerAlign());
  // *(Type0*)Dst[0] = *(Type0*)Src[0];
  // *(Type1*)Dst[1] = *(Type1*)Src[1];
  // ...
  // *(Typen*)Dst[n] = *(Typen*)Src[n];
  for (unsigned I = 0, E = AssignmentOps.size(); I < E; ++I) {
    const auto *DestVar =
        cast<VarDecl>(cast<DeclRefExpr>(DestExprs[I])->getDecl());
    Address DestAddr = emitAddrOfVarFromArray(CGF, LHS, I, DestVar);

    const auto *SrcVar =
        cast<VarDecl>(cast<DeclRefExpr>(SrcExprs[I])->getDecl());
    Address SrcAddr = emitAddrOfVarFromArray(CGF, RHS, I, SrcVar);

    const auto *VD = cast<DeclRefExpr>(CopyprivateVars[I])->getDecl();
    QualType Type = VD->getType();
    CGF.EmitOMPCopy(Type, DestAddr, SrcAddr, DestVar, SrcVar, AssignmentOps[I]);
  }
  CGF.FinishFunction();
  return Fn;
}

void CGOpenMPRuntime::emitSingleRegion(CodeGenFunction &CGF,
                                       const RegionCodeGenTy &SingleOpGen,
                                       SourceLocation Loc,
                                       ArrayRef<const Expr *> CopyprivateVars,
                                       ArrayRef<const Expr *> SrcExprs,
                                       ArrayRef<const Expr *> DstExprs,
                                       ArrayRef<const Expr *> AssignmentOps) {
  if (!CGF.HaveInsertPoint())
    return;
  assert(CopyprivateVars.size() == SrcExprs.size() &&
         CopyprivateVars.size() == DstExprs.size() &&
         CopyprivateVars.size() == AssignmentOps.size());
  ASTContext &C = CGM.getContext();
  // int32 did_it = 0;
  // if(__kmpc_single(ident_t *, gtid)) {
  //   SingleOpGen();
  //   __kmpc_end_single(ident_t *, gtid);
  //   did_it = 1;
  // }
  // call __kmpc_copyprivate(ident_t *, gtid, <buf_size>, <copyprivate list>,
  // <copy_func>, did_it);

  Address DidIt = Address::invalid();
  if (!CopyprivateVars.empty()) {
    // int32 did_it = 0;
    QualType KmpInt32Ty =
        C.getIntTypeForBitwidth(/*DestWidth=*/32, /*Signed=*/1);
    DidIt = CGF.CreateMemTemp(KmpInt32Ty, ".omp.copyprivate.did_it");
    CGF.Builder.CreateStore(CGF.Builder.getInt32(0), DidIt);
  }
  // Prepare arguments and build a call to __kmpc_single
  llvm::Value *Args[] = {emitUpdateLocation(CGF, Loc), getThreadID(CGF, Loc)};
  CommonActionTy Action(createRuntimeFunction(OMPRTL__kmpc_single), Args,
                        createRuntimeFunction(OMPRTL__kmpc_end_single), Args,
                        /*Conditional=*/true);
  SingleOpGen.setAction(Action);
  emitInlinedDirective(CGF, OMPD_single, SingleOpGen);
  if (DidIt.isValid()) {
    // did_it = 1;
    CGF.Builder.CreateStore(CGF.Builder.getInt32(1), DidIt);
  }
  Action.Done(CGF);
  // call __kmpc_copyprivate(ident_t *, gtid, <buf_size>, <copyprivate list>,
  // <copy_func>, did_it);
  if (DidIt.isValid()) {
    llvm::APInt ArraySize(/*unsigned int numBits=*/32, CopyprivateVars.size());
    QualType CopyprivateArrayTy =
        C.getConstantArrayType(C.VoidPtrTy, ArraySize, ArrayType::Normal,
                               /*IndexTypeQuals=*/0);
    // Create a list of all private variables for copyprivate.
    Address CopyprivateList =
        CGF.CreateMemTemp(CopyprivateArrayTy, ".omp.copyprivate.cpr_list");
    for (unsigned I = 0, E = CopyprivateVars.size(); I < E; ++I) {
      Address Elem = CGF.Builder.CreateConstArrayGEP(CopyprivateList, I);
      CGF.Builder.CreateStore(
          CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
              CGF.EmitLValue(CopyprivateVars[I]).getPointer(), CGF.VoidPtrTy),
          Elem);
    }
    // Build function that copies private values from single region to all other
    // threads in the corresponding parallel region.
    llvm::Value *CpyFn = emitCopyprivateCopyFunction(
        CGM, CGF.ConvertTypeForMem(CopyprivateArrayTy)->getPointerTo(),
        CopyprivateVars, SrcExprs, DstExprs, AssignmentOps, Loc);
    llvm::Value *BufSize = CGF.getTypeSize(CopyprivateArrayTy);
    Address CL =
      CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(CopyprivateList,
                                                      CGF.VoidPtrTy);
    llvm::Value *DidItVal = CGF.Builder.CreateLoad(DidIt);
    llvm::Value *Args[] = {
        emitUpdateLocation(CGF, Loc), // ident_t *<loc>
        getThreadID(CGF, Loc),        // i32 <gtid>
        BufSize,                      // size_t <buf_size>
        CL.getPointer(),              // void *<copyprivate list>
        CpyFn,                        // void (*) (void *, void *) <copy_func>
        DidItVal                      // i32 did_it
    };
    CGF.EmitRuntimeCall(createRuntimeFunction(OMPRTL__kmpc_copyprivate), Args);
  }
}

void CGOpenMPRuntime::emitOrderedRegion(CodeGenFunction &CGF,
                                        const RegionCodeGenTy &OrderedOpGen,
                                        SourceLocation Loc, bool IsThreads) {
  if (!CGF.HaveInsertPoint())
    return;
  // __kmpc_ordered(ident_t *, gtid);
  // OrderedOpGen();
  // __kmpc_end_ordered(ident_t *, gtid);
  // Prepare arguments and build a call to __kmpc_ordered
  if (IsThreads) {
    llvm::Value *Args[] = {emitUpdateLocation(CGF, Loc), getThreadID(CGF, Loc)};
    CommonActionTy Action(createRuntimeFunction(OMPRTL__kmpc_ordered), Args,
                          createRuntimeFunction(OMPRTL__kmpc_end_ordered),
                          Args);
    OrderedOpGen.setAction(Action);
    emitInlinedDirective(CGF, OMPD_ordered, OrderedOpGen);
    return;
  }
  emitInlinedDirective(CGF, OMPD_ordered, OrderedOpGen);
}

unsigned CGOpenMPRuntime::getDefaultFlagsForBarriers(OpenMPDirectiveKind Kind) {
  unsigned Flags;
  if (Kind == OMPD_for)
    Flags = OMP_IDENT_BARRIER_IMPL_FOR;
  else if (Kind == OMPD_sections)
    Flags = OMP_IDENT_BARRIER_IMPL_SECTIONS;
  else if (Kind == OMPD_single)
    Flags = OMP_IDENT_BARRIER_IMPL_SINGLE;
  else if (Kind == OMPD_barrier)
    Flags = OMP_IDENT_BARRIER_EXPL;
  else
    Flags = OMP_IDENT_BARRIER_IMPL;
  return Flags;
}

void CGOpenMPRuntime::emitBarrierCall(CodeGenFunction &CGF, SourceLocation Loc,
                                      OpenMPDirectiveKind Kind, bool EmitChecks,
                                      bool ForceSimpleCall) {
  if (!CGF.HaveInsertPoint())
    return;
  // Build call __kmpc_cancel_barrier(loc, thread_id);
  // Build call __kmpc_barrier(loc, thread_id);
  unsigned Flags = getDefaultFlagsForBarriers(Kind);
  // Build call __kmpc_cancel_barrier(loc, thread_id) or __kmpc_barrier(loc,
  // thread_id);
  llvm::Value *Args[] = {emitUpdateLocation(CGF, Loc, Flags),
                         getThreadID(CGF, Loc)};
  if (auto *OMPRegionInfo =
          dyn_cast_or_null<CGOpenMPRegionInfo>(CGF.CapturedStmtInfo)) {
    if (!ForceSimpleCall && OMPRegionInfo->hasCancel()) {
      llvm::Value *Result = CGF.EmitRuntimeCall(
          createRuntimeFunction(OMPRTL__kmpc_cancel_barrier), Args);
      if (EmitChecks) {
        // if (__kmpc_cancel_barrier()) {
        //   exit from construct;
        // }
        llvm::BasicBlock *ExitBB = CGF.createBasicBlock(".cancel.exit");
        llvm::BasicBlock *ContBB = CGF.createBasicBlock(".cancel.continue");
        llvm::Value *Cmp = CGF.Builder.CreateIsNotNull(Result);
        CGF.Builder.CreateCondBr(Cmp, ExitBB, ContBB);
        CGF.EmitBlock(ExitBB);
        //   exit from construct;
        CodeGenFunction::JumpDest CancelDestination =
            CGF.getOMPCancelDestination(OMPRegionInfo->getDirectiveKind());
        CGF.EmitBranchThroughCleanup(CancelDestination);
        CGF.EmitBlock(ContBB, /*IsFinished=*/true);
      }
      return;
    }
  }
  CGF.EmitRuntimeCall(createRuntimeFunction(OMPRTL__kmpc_barrier), Args);
}

/// Map the OpenMP loop schedule to the runtime enumeration.
static OpenMPSchedType getRuntimeSchedule(OpenMPScheduleClauseKind ScheduleKind,
                                          bool Chunked, bool Ordered) {
  switch (ScheduleKind) {
  case OMPC_SCHEDULE_static:
    return Chunked ? (Ordered ? OMP_ord_static_chunked : OMP_sch_static_chunked)
                   : (Ordered ? OMP_ord_static : OMP_sch_static);
  case OMPC_SCHEDULE_dynamic:
    return Ordered ? OMP_ord_dynamic_chunked : OMP_sch_dynamic_chunked;
  case OMPC_SCHEDULE_guided:
    return Ordered ? OMP_ord_guided_chunked : OMP_sch_guided_chunked;
  case OMPC_SCHEDULE_runtime:
    return Ordered ? OMP_ord_runtime : OMP_sch_runtime;
  case OMPC_SCHEDULE_auto:
    return Ordered ? OMP_ord_auto : OMP_sch_auto;
  case OMPC_SCHEDULE_unknown:
    assert(!Chunked && "chunk was specified but schedule kind not known");
    return Ordered ? OMP_ord_static : OMP_sch_static;
  }
  llvm_unreachable("Unexpected runtime schedule");
}

/// Map the OpenMP distribute schedule to the runtime enumeration.
static OpenMPSchedType
getRuntimeSchedule(OpenMPDistScheduleClauseKind ScheduleKind, bool Chunked) {
  // only static is allowed for dist_schedule
  return Chunked ? OMP_dist_sch_static_chunked : OMP_dist_sch_static;
}

bool CGOpenMPRuntime::isStaticNonchunked(OpenMPScheduleClauseKind ScheduleKind,
                                         bool Chunked) const {
  OpenMPSchedType Schedule =
      getRuntimeSchedule(ScheduleKind, Chunked, /*Ordered=*/false);
  return Schedule == OMP_sch_static;
}

bool CGOpenMPRuntime::isStaticNonchunked(
    OpenMPDistScheduleClauseKind ScheduleKind, bool Chunked) const {
  OpenMPSchedType Schedule = getRuntimeSchedule(ScheduleKind, Chunked);
  return Schedule == OMP_dist_sch_static;
}

bool CGOpenMPRuntime::isStaticChunked(OpenMPScheduleClauseKind ScheduleKind,
                                      bool Chunked) const {
  OpenMPSchedType Schedule =
      getRuntimeSchedule(ScheduleKind, Chunked, /*Ordered=*/false);
  return Schedule == OMP_sch_static_chunked;
}

bool CGOpenMPRuntime::isStaticChunked(
    OpenMPDistScheduleClauseKind ScheduleKind, bool Chunked) const {
  OpenMPSchedType Schedule = getRuntimeSchedule(ScheduleKind, Chunked);
  return Schedule == OMP_dist_sch_static_chunked;
}

bool CGOpenMPRuntime::isDynamic(OpenMPScheduleClauseKind ScheduleKind) const {
  OpenMPSchedType Schedule =
      getRuntimeSchedule(ScheduleKind, /*Chunked=*/false, /*Ordered=*/false);
  assert(Schedule != OMP_sch_static_chunked && "cannot be chunked here");
  return Schedule != OMP_sch_static;
}

static int addMonoNonMonoModifier(OpenMPSchedType Schedule,
                                  OpenMPScheduleClauseModifier M1,
                                  OpenMPScheduleClauseModifier M2) {
  int Modifier = 0;
  switch (M1) {
  case OMPC_SCHEDULE_MODIFIER_monotonic:
    Modifier = OMP_sch_modifier_monotonic;
    break;
  case OMPC_SCHEDULE_MODIFIER_nonmonotonic:
    Modifier = OMP_sch_modifier_nonmonotonic;
    break;
  case OMPC_SCHEDULE_MODIFIER_simd:
    if (Schedule == OMP_sch_static_chunked)
      Schedule = OMP_sch_static_balanced_chunked;
    break;
  case OMPC_SCHEDULE_MODIFIER_last:
  case OMPC_SCHEDULE_MODIFIER_unknown:
    break;
  }
  switch (M2) {
  case OMPC_SCHEDULE_MODIFIER_monotonic:
    Modifier = OMP_sch_modifier_monotonic;
    break;
  case OMPC_SCHEDULE_MODIFIER_nonmonotonic:
    Modifier = OMP_sch_modifier_nonmonotonic;
    break;
  case OMPC_SCHEDULE_MODIFIER_simd:
    if (Schedule == OMP_sch_static_chunked)
      Schedule = OMP_sch_static_balanced_chunked;
    break;
  case OMPC_SCHEDULE_MODIFIER_last:
  case OMPC_SCHEDULE_MODIFIER_unknown:
    break;
  }
  return Schedule | Modifier;
}

void CGOpenMPRuntime::emitForDispatchInit(
    CodeGenFunction &CGF, SourceLocation Loc,
    const OpenMPScheduleTy &ScheduleKind, unsigned IVSize, bool IVSigned,
    bool Ordered, const DispatchRTInput &DispatchValues) {
  if (!CGF.HaveInsertPoint())
    return;
  OpenMPSchedType Schedule = getRuntimeSchedule(
      ScheduleKind.Schedule, DispatchValues.Chunk != nullptr, Ordered);
  assert(Ordered ||
         (Schedule != OMP_sch_static && Schedule != OMP_sch_static_chunked &&
          Schedule != OMP_ord_static && Schedule != OMP_ord_static_chunked &&
          Schedule != OMP_sch_static_balanced_chunked));
  // Call __kmpc_dispatch_init(
  //          ident_t *loc, kmp_int32 tid, kmp_int32 schedule,
  //          kmp_int[32|64] lower, kmp_int[32|64] upper,
  //          kmp_int[32|64] stride, kmp_int[32|64] chunk);

  // If the Chunk was not specified in the clause - use default value 1.
  llvm::Value *Chunk = DispatchValues.Chunk ? DispatchValues.Chunk
                                            : CGF.Builder.getIntN(IVSize, 1);
  llvm::Value *Args[] = {
      emitUpdateLocation(CGF, Loc), getThreadID(CGF, Loc),
      CGF.Builder.getInt32(addMonoNonMonoModifier(
          Schedule, ScheduleKind.M1, ScheduleKind.M2)), // Schedule type
      DispatchValues.LB,                                // Lower
      DispatchValues.UB,                                // Upper
      CGF.Builder.getIntN(IVSize, 1),                   // Stride
      Chunk                                             // Chunk
  };
  CGF.EmitRuntimeCall(createDispatchInitFunction(IVSize, IVSigned), Args);
}

static void emitForStaticInitCall(
    CodeGenFunction &CGF, llvm::Value *UpdateLocation, llvm::Value *ThreadId,
    llvm::FunctionCallee ForStaticInitFunction, OpenMPSchedType Schedule,
    OpenMPScheduleClauseModifier M1, OpenMPScheduleClauseModifier M2,
    const CGOpenMPRuntime::StaticRTInput &Values) {
  if (!CGF.HaveInsertPoint())
    return;

  assert(!Values.Ordered);
  assert(Schedule == OMP_sch_static || Schedule == OMP_sch_static_chunked ||
         Schedule == OMP_sch_static_balanced_chunked ||
         Schedule == OMP_ord_static || Schedule == OMP_ord_static_chunked ||
         Schedule == OMP_dist_sch_static ||
         Schedule == OMP_dist_sch_static_chunked);

  // Call __kmpc_for_static_init(
  //          ident_t *loc, kmp_int32 tid, kmp_int32 schedtype,
  //          kmp_int32 *p_lastiter, kmp_int[32|64] *p_lower,
  //          kmp_int[32|64] *p_upper, kmp_int[32|64] *p_stride,
  //          kmp_int[32|64] incr, kmp_int[32|64] chunk);
  llvm::Value *Chunk = Values.Chunk;
  if (Chunk == nullptr) {
    assert((Schedule == OMP_sch_static || Schedule == OMP_ord_static ||
            Schedule == OMP_dist_sch_static) &&
           "expected static non-chunked schedule");
    // If the Chunk was not specified in the clause - use default value 1.
    Chunk = CGF.Builder.getIntN(Values.IVSize, 1);
  } else {
    assert((Schedule == OMP_sch_static_chunked ||
            Schedule == OMP_sch_static_balanced_chunked ||
            Schedule == OMP_ord_static_chunked ||
            Schedule == OMP_dist_sch_static_chunked) &&
           "expected static chunked schedule");
  }
  llvm::Value *Args[] = {
      UpdateLocation,
      ThreadId,
      CGF.Builder.getInt32(addMonoNonMonoModifier(Schedule, M1,
                                                  M2)), // Schedule type
      Values.IL.getPointer(),                           // &isLastIter
      Values.LB.getPointer(),                           // &LB
      Values.UB.getPointer(),                           // &UB
      Values.ST.getPointer(),                           // &Stride
      CGF.Builder.getIntN(Values.IVSize, 1),            // Incr
      Chunk                                             // Chunk
  };
  CGF.EmitRuntimeCall(ForStaticInitFunction, Args);
}

void CGOpenMPRuntime::emitForStaticInit(CodeGenFunction &CGF,
                                        SourceLocation Loc,
                                        OpenMPDirectiveKind DKind,
                                        const OpenMPScheduleTy &ScheduleKind,
                                        const StaticRTInput &Values) {
  OpenMPSchedType ScheduleNum = getRuntimeSchedule(
      ScheduleKind.Schedule, Values.Chunk != nullptr, Values.Ordered);
  assert(isOpenMPWorksharingDirective(DKind) &&
         "Expected loop-based or sections-based directive.");
  llvm::Value *UpdatedLocation = emitUpdateLocation(CGF, Loc,
                                             isOpenMPLoopDirective(DKind)
                                                 ? OMP_IDENT_WORK_LOOP
                                                 : OMP_IDENT_WORK_SECTIONS);
  llvm::Value *ThreadId = getThreadID(CGF, Loc);
  llvm::FunctionCallee StaticInitFunction =
      createForStaticInitFunction(Values.IVSize, Values.IVSigned);
  emitForStaticInitCall(CGF, UpdatedLocation, ThreadId, StaticInitFunction,
                        ScheduleNum, ScheduleKind.M1, ScheduleKind.M2, Values);
}

void CGOpenMPRuntime::emitDistributeStaticInit(
    CodeGenFunction &CGF, SourceLocation Loc,
    OpenMPDistScheduleClauseKind SchedKind,
    const CGOpenMPRuntime::StaticRTInput &Values) {
  OpenMPSchedType ScheduleNum =
      getRuntimeSchedule(SchedKind, Values.Chunk != nullptr);
  llvm::Value *UpdatedLocation =
      emitUpdateLocation(CGF, Loc, OMP_IDENT_WORK_DISTRIBUTE);
  llvm::Value *ThreadId = getThreadID(CGF, Loc);
  llvm::FunctionCallee StaticInitFunction =
      createForStaticInitFunction(Values.IVSize, Values.IVSigned);
  emitForStaticInitCall(CGF, UpdatedLocation, ThreadId, StaticInitFunction,
                        ScheduleNum, OMPC_SCHEDULE_MODIFIER_unknown,
                        OMPC_SCHEDULE_MODIFIER_unknown, Values);
}

void CGOpenMPRuntime::emitForStaticFinish(CodeGenFunction &CGF,
                                          SourceLocation Loc,
                                          OpenMPDirectiveKind DKind) {
  if (!CGF.HaveInsertPoint())
    return;
  // Call __kmpc_for_static_fini(ident_t *loc, kmp_int32 tid);
  llvm::Value *Args[] = {
      emitUpdateLocation(CGF, Loc,
                         isOpenMPDistributeDirective(DKind)
                             ? OMP_IDENT_WORK_DISTRIBUTE
                             : isOpenMPLoopDirective(DKind)
                                   ? OMP_IDENT_WORK_LOOP
                                   : OMP_IDENT_WORK_SECTIONS),
      getThreadID(CGF, Loc)};
  CGF.EmitRuntimeCall(createRuntimeFunction(OMPRTL__kmpc_for_static_fini),
                      Args);
}

void CGOpenMPRuntime::emitForOrderedIterationEnd(CodeGenFunction &CGF,
                                                 SourceLocation Loc,
                                                 unsigned IVSize,
                                                 bool IVSigned) {
  if (!CGF.HaveInsertPoint())
    return;
  // Call __kmpc_for_dynamic_fini_(4|8)[u](ident_t *loc, kmp_int32 tid);
  llvm::Value *Args[] = {emitUpdateLocation(CGF, Loc), getThreadID(CGF, Loc)};
  CGF.EmitRuntimeCall(createDispatchFiniFunction(IVSize, IVSigned), Args);
}

llvm::Value *CGOpenMPRuntime::emitForNext(CodeGenFunction &CGF,
                                          SourceLocation Loc, unsigned IVSize,
                                          bool IVSigned, Address IL,
                                          Address LB, Address UB,
                                          Address ST) {
  // Call __kmpc_dispatch_next(
  //          ident_t *loc, kmp_int32 tid, kmp_int32 *p_lastiter,
  //          kmp_int[32|64] *p_lower, kmp_int[32|64] *p_upper,
  //          kmp_int[32|64] *p_stride);
  llvm::Value *Args[] = {
      emitUpdateLocation(CGF, Loc),
      getThreadID(CGF, Loc),
      IL.getPointer(), // &isLastIter
      LB.getPointer(), // &Lower
      UB.getPointer(), // &Upper
      ST.getPointer()  // &Stride
  };
  llvm::Value *Call =
      CGF.EmitRuntimeCall(createDispatchNextFunction(IVSize, IVSigned), Args);
  return CGF.EmitScalarConversion(
      Call, CGF.getContext().getIntTypeForBitwidth(32, /*Signed=*/1),
      CGF.getContext().BoolTy, Loc);
}

void CGOpenMPRuntime::emitNumThreadsClause(CodeGenFunction &CGF,
                                           llvm::Value *NumThreads,
                                           SourceLocation Loc) {
  if (!CGF.HaveInsertPoint())
    return;
  // Build call __kmpc_push_num_threads(&loc, global_tid, num_threads)
  llvm::Value *Args[] = {
      emitUpdateLocation(CGF, Loc), getThreadID(CGF, Loc),
      CGF.Builder.CreateIntCast(NumThreads, CGF.Int32Ty, /*isSigned*/ true)};
  CGF.EmitRuntimeCall(createRuntimeFunction(OMPRTL__kmpc_push_num_threads),
                      Args);
}

void CGOpenMPRuntime::emitProcBindClause(CodeGenFunction &CGF,
                                         OpenMPProcBindClauseKind ProcBind,
                                         SourceLocation Loc) {
  if (!CGF.HaveInsertPoint())
    return;
  // Constants for proc bind value accepted by the runtime.
  enum ProcBindTy {
    ProcBindFalse = 0,
    ProcBindTrue,
    ProcBindMaster,
    ProcBindClose,
    ProcBindSpread,
    ProcBindIntel,
    ProcBindDefault
  } RuntimeProcBind;
  switch (ProcBind) {
  case OMPC_PROC_BIND_master:
    RuntimeProcBind = ProcBindMaster;
    break;
  case OMPC_PROC_BIND_close:
    RuntimeProcBind = ProcBindClose;
    break;
  case OMPC_PROC_BIND_spread:
    RuntimeProcBind = ProcBindSpread;
    break;
  case OMPC_PROC_BIND_unknown:
    llvm_unreachable("Unsupported proc_bind value.");
  }
  // Build call __kmpc_push_proc_bind(&loc, global_tid, proc_bind)
  llvm::Value *Args[] = {
      emitUpdateLocation(CGF, Loc), getThreadID(CGF, Loc),
      llvm::ConstantInt::get(CGM.IntTy, RuntimeProcBind, /*isSigned=*/true)};
  CGF.EmitRuntimeCall(createRuntimeFunction(OMPRTL__kmpc_push_proc_bind), Args);
}

void CGOpenMPRuntime::emitFlush(CodeGenFunction &CGF, ArrayRef<const Expr *>,
                                SourceLocation Loc) {
  if (!CGF.HaveInsertPoint())
    return;
  // Build call void __kmpc_flush(ident_t *loc)
  CGF.EmitRuntimeCall(createRuntimeFunction(OMPRTL__kmpc_flush),
                      emitUpdateLocation(CGF, Loc));
}

namespace {
/// Indexes of fields for type kmp_task_t.
enum KmpTaskTFields {
  /// List of shared variables.
  KmpTaskTShareds,
  /// Task routine.
  KmpTaskTRoutine,
  /// Partition id for the untied tasks.
  KmpTaskTPartId,
  /// Function with call of destructors for private variables.
  Data1,
  /// Task priority.
  Data2,
  /// (Taskloops only) Lower bound.
  KmpTaskTLowerBound,
  /// (Taskloops only) Upper bound.
  KmpTaskTUpperBound,
  /// (Taskloops only) Stride.
  KmpTaskTStride,
  /// (Taskloops only) Is last iteration flag.
  KmpTaskTLastIter,
  /// (Taskloops only) Reduction data.
  KmpTaskTReductions,
};
} // anonymous namespace

bool CGOpenMPRuntime::OffloadEntriesInfoManagerTy::empty() const {
  return OffloadEntriesTargetRegion.empty() &&
         OffloadEntriesDeviceGlobalVar.empty();
}

/// Initialize target region entry.
void CGOpenMPRuntime::OffloadEntriesInfoManagerTy::
    initializeTargetRegionEntryInfo(unsigned DeviceID, unsigned FileID,
                                    StringRef ParentName, unsigned LineNum,
                                    unsigned Order) {
  assert(CGM.getLangOpts().OpenMPIsDevice && "Initialization of entries is "
                                             "only required for the device "
                                             "code generation.");
  OffloadEntriesTargetRegion[DeviceID][FileID][ParentName][LineNum] =
      OffloadEntryInfoTargetRegion(Order, /*Addr=*/nullptr, /*ID=*/nullptr,
                                   OMPTargetRegionEntryTargetRegion);
  ++OffloadingEntriesNum;
}

void CGOpenMPRuntime::OffloadEntriesInfoManagerTy::
    registerTargetRegionEntryInfo(unsigned DeviceID, unsigned FileID,
                                  StringRef ParentName, unsigned LineNum,
                                  llvm::Constant *Addr, llvm::Constant *ID,
                                  OMPTargetRegionEntryKind Flags) {
  // If we are emitting code for a target, the entry is already initialized,
  // only has to be registered.
  if (CGM.getLangOpts().OpenMPIsDevice) {
    if (!hasTargetRegionEntryInfo(DeviceID, FileID, ParentName, LineNum)) {
      unsigned DiagID = CGM.getDiags().getCustomDiagID(
          DiagnosticsEngine::Error,
          "Unable to find target region on line '%0' in the device code.");
      CGM.getDiags().Report(DiagID) << LineNum;
      return;
    }
    auto &Entry =
        OffloadEntriesTargetRegion[DeviceID][FileID][ParentName][LineNum];
    assert(Entry.isValid() && "Entry not initialized!");
    Entry.setAddress(Addr);
    Entry.setID(ID);
    Entry.setFlags(Flags);
  } else {
    OffloadEntryInfoTargetRegion Entry(OffloadingEntriesNum, Addr, ID, Flags);
    OffloadEntriesTargetRegion[DeviceID][FileID][ParentName][LineNum] = Entry;
    ++OffloadingEntriesNum;
  }
}

bool CGOpenMPRuntime::OffloadEntriesInfoManagerTy::hasTargetRegionEntryInfo(
    unsigned DeviceID, unsigned FileID, StringRef ParentName,
    unsigned LineNum) const {
  auto PerDevice = OffloadEntriesTargetRegion.find(DeviceID);
  if (PerDevice == OffloadEntriesTargetRegion.end())
    return false;
  auto PerFile = PerDevice->second.find(FileID);
  if (PerFile == PerDevice->second.end())
    return false;
  auto PerParentName = PerFile->second.find(ParentName);
  if (PerParentName == PerFile->second.end())
    return false;
  auto PerLine = PerParentName->second.find(LineNum);
  if (PerLine == PerParentName->second.end())
    return false;
  // Fail if this entry is already registered.
  if (PerLine->second.getAddress() || PerLine->second.getID())
    return false;
  return true;
}

void CGOpenMPRuntime::OffloadEntriesInfoManagerTy::actOnTargetRegionEntriesInfo(
    const OffloadTargetRegionEntryInfoActTy &Action) {
  // Scan all target region entries and perform the provided action.
  for (const auto &D : OffloadEntriesTargetRegion)
    for (const auto &F : D.second)
      for (const auto &P : F.second)
        for (const auto &L : P.second)
          Action(D.first, F.first, P.first(), L.first, L.second);
}

void CGOpenMPRuntime::OffloadEntriesInfoManagerTy::
    initializeDeviceGlobalVarEntryInfo(StringRef Name,
                                       OMPTargetGlobalVarEntryKind Flags,
                                       unsigned Order) {
  assert(CGM.getLangOpts().OpenMPIsDevice && "Initialization of entries is "
                                             "only required for the device "
                                             "code generation.");
  OffloadEntriesDeviceGlobalVar.try_emplace(Name, Order, Flags);
  ++OffloadingEntriesNum;
}

void CGOpenMPRuntime::OffloadEntriesInfoManagerTy::
    registerDeviceGlobalVarEntryInfo(StringRef VarName, llvm::Constant *Addr,
                                     CharUnits VarSize,
                                     OMPTargetGlobalVarEntryKind Flags,
                                     llvm::GlobalValue::LinkageTypes Linkage) {
  if (CGM.getLangOpts().OpenMPIsDevice) {
    auto &Entry = OffloadEntriesDeviceGlobalVar[VarName];
    assert(Entry.isValid() && Entry.getFlags() == Flags &&
           "Entry not initialized!");
    assert((!Entry.getAddress() || Entry.getAddress() == Addr) &&
           "Resetting with the new address.");
    if (Entry.getAddress() && hasDeviceGlobalVarEntryInfo(VarName))
      return;
    Entry.setAddress(Addr);
    Entry.setVarSize(VarSize);
    Entry.setLinkage(Linkage);
  } else {
    if (hasDeviceGlobalVarEntryInfo(VarName))
      return;
    OffloadEntriesDeviceGlobalVar.try_emplace(
        VarName, OffloadingEntriesNum, Addr, VarSize, Flags, Linkage);
    ++OffloadingEntriesNum;
  }
}

void CGOpenMPRuntime::OffloadEntriesInfoManagerTy::
    actOnDeviceGlobalVarEntriesInfo(
        const OffloadDeviceGlobalVarEntryInfoActTy &Action) {
  // Scan all target region entries and perform the provided action.
  for (const auto &E : OffloadEntriesDeviceGlobalVar)
    Action(E.getKey(), E.getValue());
}

llvm::Function *
CGOpenMPRuntime::createOffloadingBinaryDescriptorRegistration() {
  // If we don't have entries or if we are emitting code for the device, we
  // don't need to do anything.
  if (CGM.getLangOpts().OpenMPIsDevice || OffloadEntriesInfoManager.empty())
    return nullptr;

  llvm::Module &M = CGM.getModule();
  ASTContext &C = CGM.getContext();

  // Get list of devices we care about
  const std::vector<llvm::Triple> &Devices = CGM.getLangOpts().OMPTargetTriples;

  // We should be creating an offloading descriptor only if there are devices
  // specified.
  assert(!Devices.empty() && "No OpenMP offloading devices??");

  // Create the external variables that will point to the begin and end of the
  // host entries section. These will be defined by the linker.
  llvm::Type *OffloadEntryTy =
      CGM.getTypes().ConvertTypeForMem(getTgtOffloadEntryQTy());
  std::string EntriesBeginName = getName({"omp_offloading", "entries_begin"});
  auto *HostEntriesBegin = new llvm::GlobalVariable(
      M, OffloadEntryTy, /*isConstant=*/true,
      llvm::GlobalValue::ExternalLinkage, /*Initializer=*/nullptr,
      EntriesBeginName);
  std::string EntriesEndName = getName({"omp_offloading", "entries_end"});
  auto *HostEntriesEnd =
      new llvm::GlobalVariable(M, OffloadEntryTy, /*isConstant=*/true,
                               llvm::GlobalValue::ExternalLinkage,
                               /*Initializer=*/nullptr, EntriesEndName);

  // Create all device images
  auto *DeviceImageTy = cast<llvm::StructType>(
      CGM.getTypes().ConvertTypeForMem(getTgtDeviceImageQTy()));
  ConstantInitBuilder DeviceImagesBuilder(CGM);
  ConstantArrayBuilder DeviceImagesEntries =
      DeviceImagesBuilder.beginArray(DeviceImageTy);

  for (const llvm::Triple &Device : Devices) {
    StringRef T = Device.getTriple();
    std::string BeginName = getName({"omp_offloading", "img_start", ""});
    auto *ImgBegin = new llvm::GlobalVariable(
        M, CGM.Int8Ty, /*isConstant=*/true,
        llvm::GlobalValue::ExternalWeakLinkage,
        /*Initializer=*/nullptr, Twine(BeginName).concat(T));
    std::string EndName = getName({"omp_offloading", "img_end", ""});
    auto *ImgEnd = new llvm::GlobalVariable(
        M, CGM.Int8Ty, /*isConstant=*/true,
        llvm::GlobalValue::ExternalWeakLinkage,
        /*Initializer=*/nullptr, Twine(EndName).concat(T));

    llvm::Constant *Data[] = {ImgBegin, ImgEnd, HostEntriesBegin,
                              HostEntriesEnd};
    createConstantGlobalStructAndAddToParent(CGM, getTgtDeviceImageQTy(), Data,
                                             DeviceImagesEntries);
  }

  // Create device images global array.
  std::string ImagesName = getName({"omp_offloading", "device_images"});
  llvm::GlobalVariable *DeviceImages =
      DeviceImagesEntries.finishAndCreateGlobal(ImagesName,
                                                CGM.getPointerAlign(),
                                                /*isConstant=*/true);
  DeviceImages->setUnnamedAddr(llvm::GlobalValue::UnnamedAddr::Global);

  // This is a Zero array to be used in the creation of the constant expressions
  llvm::Constant *Index[] = {llvm::Constant::getNullValue(CGM.Int32Ty),
                             llvm::Constant::getNullValue(CGM.Int32Ty)};

  // Create the target region descriptor.
  llvm::Constant *Data[] = {
      llvm::ConstantInt::get(CGM.Int32Ty, Devices.size()),
      llvm::ConstantExpr::getGetElementPtr(DeviceImages->getValueType(),
                                           DeviceImages, Index),
      HostEntriesBegin, HostEntriesEnd};
  std::string Descriptor = getName({"omp_offloading", "descriptor"});
  llvm::GlobalVariable *Desc = createGlobalStruct(
      CGM, getTgtBinaryDescriptorQTy(), /*IsConstant=*/true, Data, Descriptor);

  // Emit code to register or unregister the descriptor at execution
  // startup or closing, respectively.

  llvm::Function *UnRegFn;
  {
    FunctionArgList Args;
    ImplicitParamDecl DummyPtr(C, C.VoidPtrTy, ImplicitParamDecl::Other);
    Args.push_back(&DummyPtr);

    CodeGenFunction CGF(CGM);
    // Disable debug info for global (de-)initializer because they are not part
    // of some particular construct.
    CGF.disableDebugInfo();
    const auto &FI =
        CGM.getTypes().arrangeBuiltinFunctionDeclaration(C.VoidTy, Args);
    llvm::FunctionType *FTy = CGM.getTypes().GetFunctionType(FI);
    std::string UnregName = getName({"omp_offloading", "descriptor_unreg"});
    UnRegFn = CGM.CreateGlobalInitOrDestructFunction(FTy, UnregName, FI);
    CGF.StartFunction(GlobalDecl(), C.VoidTy, UnRegFn, FI, Args);
    CGF.EmitRuntimeCall(createRuntimeFunction(OMPRTL__tgt_unregister_lib),
                        Desc);
    CGF.FinishFunction();
  }
  llvm::Function *RegFn;
  {
    CodeGenFunction CGF(CGM);
    // Disable debug info for global (de-)initializer because they are not part
    // of some particular construct.
    CGF.disableDebugInfo();
    const auto &FI = CGM.getTypes().arrangeNullaryFunction();
    llvm::FunctionType *FTy = CGM.getTypes().GetFunctionType(FI);

    // Encode offload target triples into the registration function name. It
    // will serve as a comdat key for the registration/unregistration code for
    // this particular combination of offloading targets.
    SmallVector<StringRef, 4U> RegFnNameParts(Devices.size() + 2U);
    RegFnNameParts[0] = "omp_offloading";
    RegFnNameParts[1] = "descriptor_reg";
    llvm::transform(Devices, std::next(RegFnNameParts.begin(), 2),
                    [](const llvm::Triple &T) -> const std::string& {
                      return T.getTriple();
                    });
    llvm::sort(std::next(RegFnNameParts.begin(), 2), RegFnNameParts.end());
    std::string Descriptor = getName(RegFnNameParts);
    RegFn = CGM.CreateGlobalInitOrDestructFunction(FTy, Descriptor, FI);
    CGF.StartFunction(GlobalDecl(), C.VoidTy, RegFn, FI, FunctionArgList());
    CGF.EmitRuntimeCall(createRuntimeFunction(OMPRTL__tgt_register_lib), Desc);
    // Create a variable to drive the registration and unregistration of the
    // descriptor, so we can reuse the logic that emits Ctors and Dtors.
    ImplicitParamDecl RegUnregVar(C, C.getTranslationUnitDecl(),
                                  SourceLocation(), nullptr, C.CharTy,
                                  ImplicitParamDecl::Other);
    CGM.getCXXABI().registerGlobalDtor(CGF, RegUnregVar, UnRegFn, Desc);
    CGF.FinishFunction();
  }
  if (CGM.supportsCOMDAT()) {
    // It is sufficient to call registration function only once, so create a
    // COMDAT group for registration/unregistration functions and associated
    // data. That would reduce startup time and code size. Registration
    // function serves as a COMDAT group key.
    llvm::Comdat *ComdatKey = M.getOrInsertComdat(RegFn->getName());
    RegFn->setLinkage(llvm::GlobalValue::LinkOnceAnyLinkage);
    RegFn->setVisibility(llvm::GlobalValue::HiddenVisibility);
    RegFn->setComdat(ComdatKey);
    UnRegFn->setComdat(ComdatKey);
    DeviceImages->setComdat(ComdatKey);
    Desc->setComdat(ComdatKey);
  }
  return RegFn;
}

void CGOpenMPRuntime::createOffloadEntry(
    llvm::Constant *ID, llvm::Constant *Addr, uint64_t Size, int32_t Flags,
    llvm::GlobalValue::LinkageTypes Linkage) {
  StringRef Name = Addr->getName();
  llvm::Module &M = CGM.getModule();
  llvm::LLVMContext &C = M.getContext();

  // Create constant string with the name.
  llvm::Constant *StrPtrInit = llvm::ConstantDataArray::getString(C, Name);

  std::string StringName = getName({"omp_offloading", "entry_name"});
  auto *Str = new llvm::GlobalVariable(
      M, StrPtrInit->getType(), /*isConstant=*/true,
      llvm::GlobalValue::InternalLinkage, StrPtrInit, StringName);
  Str->setUnnamedAddr(llvm::GlobalValue::UnnamedAddr::Global);

  llvm::Constant *Data[] = {llvm::ConstantExpr::getBitCast(ID, CGM.VoidPtrTy),
                            llvm::ConstantExpr::getBitCast(Str, CGM.Int8PtrTy),
                            llvm::ConstantInt::get(CGM.SizeTy, Size),
                            llvm::ConstantInt::get(CGM.Int32Ty, Flags),
                            llvm::ConstantInt::get(CGM.Int32Ty, 0)};
  std::string EntryName = getName({"omp_offloading", "entry", ""});
  llvm::GlobalVariable *Entry = createGlobalStruct(
      CGM, getTgtOffloadEntryQTy(), /*IsConstant=*/true, Data,
      Twine(EntryName).concat(Name), llvm::GlobalValue::WeakAnyLinkage);

  // The entry has to be created in the section the linker expects it to be.
  std::string Section = getName({"omp_offloading", "entries"});
  Entry->setSection(Section);
}

void CGOpenMPRuntime::createOffloadEntriesAndInfoMetadata() {
  // Emit the offloading entries and metadata so that the device codegen side
  // can easily figure out what to emit. The produced metadata looks like
  // this:
  //
  // !omp_offload.info = !{!1, ...}
  //
  // Right now we only generate metadata for function that contain target
  // regions.

  // If we do not have entries, we don't need to do anything.
  if (OffloadEntriesInfoManager.empty())
    return;

  llvm::Module &M = CGM.getModule();
  llvm::LLVMContext &C = M.getContext();
  SmallVector<const OffloadEntriesInfoManagerTy::OffloadEntryInfo *, 16>
      OrderedEntries(OffloadEntriesInfoManager.size());
  llvm::SmallVector<StringRef, 16> ParentFunctions(
      OffloadEntriesInfoManager.size());

  // Auxiliary methods to create metadata values and strings.
  auto &&GetMDInt = [this](unsigned V) {
    return llvm::ConstantAsMetadata::get(
        llvm::ConstantInt::get(CGM.Int32Ty, V));
  };

  auto &&GetMDString = [&C](StringRef V) { return llvm::MDString::get(C, V); };

  // Create the offloading info metadata node.
  llvm::NamedMDNode *MD = M.getOrInsertNamedMetadata("omp_offload.info");

  // Create function that emits metadata for each target region entry;
  auto &&TargetRegionMetadataEmitter =
      [&C, MD, &OrderedEntries, &ParentFunctions, &GetMDInt, &GetMDString](
          unsigned DeviceID, unsigned FileID, StringRef ParentName,
          unsigned Line,
          const OffloadEntriesInfoManagerTy::OffloadEntryInfoTargetRegion &E) {
        // Generate metadata for target regions. Each entry of this metadata
        // contains:
        // - Entry 0 -> Kind of this type of metadata (0).
        // - Entry 1 -> Device ID of the file where the entry was identified.
        // - Entry 2 -> File ID of the file where the entry was identified.
        // - Entry 3 -> Mangled name of the function where the entry was
        // identified.
        // - Entry 4 -> Line in the file where the entry was identified.
        // - Entry 5 -> Order the entry was created.
        // The first element of the metadata node is the kind.
        llvm::Metadata *Ops[] = {GetMDInt(E.getKind()), GetMDInt(DeviceID),
                                 GetMDInt(FileID),      GetMDString(ParentName),
                                 GetMDInt(Line),        GetMDInt(E.getOrder())};

        // Save this entry in the right position of the ordered entries array.
        OrderedEntries[E.getOrder()] = &E;
        ParentFunctions[E.getOrder()] = ParentName;

        // Add metadata to the named metadata node.
        MD->addOperand(llvm::MDNode::get(C, Ops));
      };

  OffloadEntriesInfoManager.actOnTargetRegionEntriesInfo(
      TargetRegionMetadataEmitter);

  // Create function that emits metadata for each device global variable entry;
  auto &&DeviceGlobalVarMetadataEmitter =
      [&C, &OrderedEntries, &GetMDInt, &GetMDString,
       MD](StringRef MangledName,
           const OffloadEntriesInfoManagerTy::OffloadEntryInfoDeviceGlobalVar
               &E) {
        // Generate metadata for global variables. Each entry of this metadata
        // contains:
        // - Entry 0 -> Kind of this type of metadata (1).
        // - Entry 1 -> Mangled name of the variable.
        // - Entry 2 -> Declare target kind.
        // - Entry 3 -> Order the entry was created.
        // The first element of the metadata node is the kind.
        llvm::Metadata *Ops[] = {
            GetMDInt(E.getKind()), GetMDString(MangledName),
            GetMDInt(E.getFlags()), GetMDInt(E.getOrder())};

        // Save this entry in the right position of the ordered entries array.
        OrderedEntries[E.getOrder()] = &E;

        // Add metadata to the named metadata node.
        MD->addOperand(llvm::MDNode::get(C, Ops));
      };

  OffloadEntriesInfoManager.actOnDeviceGlobalVarEntriesInfo(
      DeviceGlobalVarMetadataEmitter);

  for (const auto *E : OrderedEntries) {
    assert(E && "All ordered entries must exist!");
    if (const auto *CE =
            dyn_cast<OffloadEntriesInfoManagerTy::OffloadEntryInfoTargetRegion>(
                E)) {
      if (!CE->getID() || !CE->getAddress()) {
        // Do not blame the entry if the parent funtion is not emitted.
        StringRef FnName = ParentFunctions[CE->getOrder()];
        if (!CGM.GetGlobalValue(FnName))
          continue;
        unsigned DiagID = CGM.getDiags().getCustomDiagID(
            DiagnosticsEngine::Error,
            "Offloading entry for target region is incorrect: either the "
            "address or the ID is invalid.");
        CGM.getDiags().Report(DiagID);
        continue;
      }
      createOffloadEntry(CE->getID(), CE->getAddress(), /*Size=*/0,
                         CE->getFlags(), llvm::GlobalValue::WeakAnyLinkage);
    } else if (const auto *CE =
                   dyn_cast<OffloadEntriesInfoManagerTy::
                                OffloadEntryInfoDeviceGlobalVar>(E)) {
      OffloadEntriesInfoManagerTy::OMPTargetGlobalVarEntryKind Flags =
          static_cast<OffloadEntriesInfoManagerTy::OMPTargetGlobalVarEntryKind>(
              CE->getFlags());
      switch (Flags) {
      case OffloadEntriesInfoManagerTy::OMPTargetGlobalVarEntryTo: {
        if (!CE->getAddress()) {
          unsigned DiagID = CGM.getDiags().getCustomDiagID(
              DiagnosticsEngine::Error,
              "Offloading entry for declare target variable is incorrect: the "
              "address is invalid.");
          CGM.getDiags().Report(DiagID);
          continue;
        }
        // The vaiable has no definition - no need to add the entry.
        if (CE->getVarSize().isZero())
          continue;
        break;
      }
      case OffloadEntriesInfoManagerTy::OMPTargetGlobalVarEntryLink:
        assert(((CGM.getLangOpts().OpenMPIsDevice && !CE->getAddress()) ||
                (!CGM.getLangOpts().OpenMPIsDevice && CE->getAddress())) &&
               "Declaret target link address is set.");
        if (CGM.getLangOpts().OpenMPIsDevice)
          continue;
        if (!CE->getAddress()) {
          unsigned DiagID = CGM.getDiags().getCustomDiagID(
              DiagnosticsEngine::Error,
              "Offloading entry for declare target variable is incorrect: the "
              "address is invalid.");
          CGM.getDiags().Report(DiagID);
          continue;
        }
        break;
      }
      createOffloadEntry(CE->getAddress(), CE->getAddress(),
                         CE->getVarSize().getQuantity(), Flags,
                         CE->getLinkage());
    } else {
      llvm_unreachable("Unsupported entry kind.");
    }
  }
}

/// Loads all the offload entries information from the host IR
/// metadata.
void CGOpenMPRuntime::loadOffloadInfoMetadata() {
  // If we are in target mode, load the metadata from the host IR. This code has
  // to match the metadaata creation in createOffloadEntriesAndInfoMetadata().

  if (!CGM.getLangOpts().OpenMPIsDevice)
    return;

  if (CGM.getLangOpts().OMPHostIRFile.empty())
    return;

  auto Buf = llvm::MemoryBuffer::getFile(CGM.getLangOpts().OMPHostIRFile);
  if (auto EC = Buf.getError()) {
    CGM.getDiags().Report(diag::err_cannot_open_file)
        << CGM.getLangOpts().OMPHostIRFile << EC.message();
    return;
  }

  llvm::LLVMContext C;
  auto ME = expectedToErrorOrAndEmitErrors(
      C, llvm::parseBitcodeFile(Buf.get()->getMemBufferRef(), C));

  if (auto EC = ME.getError()) {
    unsigned DiagID = CGM.getDiags().getCustomDiagID(
        DiagnosticsEngine::Error, "Unable to parse host IR file '%0':'%1'");
    CGM.getDiags().Report(DiagID)
        << CGM.getLangOpts().OMPHostIRFile << EC.message();
    return;
  }

  llvm::NamedMDNode *MD = ME.get()->getNamedMetadata("omp_offload.info");
  if (!MD)
    return;

  for (llvm::MDNode *MN : MD->operands()) {
    auto &&GetMDInt = [MN](unsigned Idx) {
      auto *V = cast<llvm::ConstantAsMetadata>(MN->getOperand(Idx));
      return cast<llvm::ConstantInt>(V->getValue())->getZExtValue();
    };

    auto &&GetMDString = [MN](unsigned Idx) {
      auto *V = cast<llvm::MDString>(MN->getOperand(Idx));
      return V->getString();
    };

    switch (GetMDInt(0)) {
    default:
      llvm_unreachable("Unexpected metadata!");
      break;
    case OffloadEntriesInfoManagerTy::OffloadEntryInfo::
        OffloadingEntryInfoTargetRegion:
      OffloadEntriesInfoManager.initializeTargetRegionEntryInfo(
          /*DeviceID=*/GetMDInt(1), /*FileID=*/GetMDInt(2),
          /*ParentName=*/GetMDString(3), /*Line=*/GetMDInt(4),
          /*Order=*/GetMDInt(5));
      break;
    case OffloadEntriesInfoManagerTy::OffloadEntryInfo::
        OffloadingEntryInfoDeviceGlobalVar:
      OffloadEntriesInfoManager.initializeDeviceGlobalVarEntryInfo(
          /*MangledName=*/GetMDString(1),
          static_cast<OffloadEntriesInfoManagerTy::OMPTargetGlobalVarEntryKind>(
              /*Flags=*/GetMDInt(2)),
          /*Order=*/GetMDInt(3));
      break;
    }
  }
}

void CGOpenMPRuntime::emitKmpRoutineEntryT(QualType KmpInt32Ty) {
  if (!KmpRoutineEntryPtrTy) {
    // Build typedef kmp_int32 (* kmp_routine_entry_t)(kmp_int32, void *); type.
    ASTContext &C = CGM.getContext();
    QualType KmpRoutineEntryTyArgs[] = {KmpInt32Ty, C.VoidPtrTy};
    FunctionProtoType::ExtProtoInfo EPI;
    KmpRoutineEntryPtrQTy = C.getPointerType(
        C.getFunctionType(KmpInt32Ty, KmpRoutineEntryTyArgs, EPI));
    KmpRoutineEntryPtrTy = CGM.getTypes().ConvertType(KmpRoutineEntryPtrQTy);
  }
}

QualType CGOpenMPRuntime::getTgtOffloadEntryQTy() {
  // Make sure the type of the entry is already created. This is the type we
  // have to create:
  // struct __tgt_offload_entry{
  //   void      *addr;       // Pointer to the offload entry info.
  //                          // (function or global)
  //   char      *name;       // Name of the function or global.
  //   size_t     size;       // Size of the entry info (0 if it a function).
  //   int32_t    flags;      // Flags associated with the entry, e.g. 'link'.
  //   int32_t    reserved;   // Reserved, to use by the runtime library.
  // };
  if (TgtOffloadEntryQTy.isNull()) {
    ASTContext &C = CGM.getContext();
    RecordDecl *RD = C.buildImplicitRecord("__tgt_offload_entry");
    RD->startDefinition();
    addFieldToRecordDecl(C, RD, C.VoidPtrTy);
    addFieldToRecordDecl(C, RD, C.getPointerType(C.CharTy));
    addFieldToRecordDecl(C, RD, C.getSizeType());
    addFieldToRecordDecl(
        C, RD, C.getIntTypeForBitwidth(/*DestWidth=*/32, /*Signed=*/true));
    addFieldToRecordDecl(
        C, RD, C.getIntTypeForBitwidth(/*DestWidth=*/32, /*Signed=*/true));
    RD->completeDefinition();
    RD->addAttr(PackedAttr::CreateImplicit(C));
    TgtOffloadEntryQTy = C.getRecordType(RD);
  }
  return TgtOffloadEntryQTy;
}

QualType CGOpenMPRuntime::getTgtDeviceImageQTy() {
  // These are the types we need to build:
  // struct __tgt_device_image{
  // void   *ImageStart;       // Pointer to the target code start.
  // void   *ImageEnd;         // Pointer to the target code end.
  // // We also add the host entries to the device image, as it may be useful
  // // for the target runtime to have access to that information.
  // __tgt_offload_entry  *EntriesBegin;   // Begin of the table with all
  //                                       // the entries.
  // __tgt_offload_entry  *EntriesEnd;     // End of the table with all the
  //                                       // entries (non inclusive).
  // };
  if (TgtDeviceImageQTy.isNull()) {
    ASTContext &C = CGM.getContext();
    RecordDecl *RD = C.buildImplicitRecord("__tgt_device_image");
    RD->startDefinition();
    addFieldToRecordDecl(C, RD, C.VoidPtrTy);
    addFieldToRecordDecl(C, RD, C.VoidPtrTy);
    addFieldToRecordDecl(C, RD, C.getPointerType(getTgtOffloadEntryQTy()));
    addFieldToRecordDecl(C, RD, C.getPointerType(getTgtOffloadEntryQTy()));
    RD->completeDefinition();
    TgtDeviceImageQTy = C.getRecordType(RD);
  }
  return TgtDeviceImageQTy;
}

QualType CGOpenMPRuntime::getTgtBinaryDescriptorQTy() {
  // struct __tgt_bin_desc{
  //   int32_t              NumDevices;      // Number of devices supported.
  //   __tgt_device_image   *DeviceImages;   // Arrays of device images
  //                                         // (one per device).
  //   __tgt_offload_entry  *EntriesBegin;   // Begin of the table with all the
  //                                         // entries.
  //   __tgt_offload_entry  *EntriesEnd;     // End of the table with all the
  //                                         // entries (non inclusive).
  // };
  if (TgtBinaryDescriptorQTy.isNull()) {
    ASTContext &C = CGM.getContext();
    RecordDecl *RD = C.buildImplicitRecord("__tgt_bin_desc");
    RD->startDefinition();
    addFieldToRecordDecl(
        C, RD, C.getIntTypeForBitwidth(/*DestWidth=*/32, /*Signed=*/true));
    addFieldToRecordDecl(C, RD, C.getPointerType(getTgtDeviceImageQTy()));
    addFieldToRecordDecl(C, RD, C.getPointerType(getTgtOffloadEntryQTy()));
    addFieldToRecordDecl(C, RD, C.getPointerType(getTgtOffloadEntryQTy()));
    RD->completeDefinition();
    TgtBinaryDescriptorQTy = C.getRecordType(RD);
  }
  return TgtBinaryDescriptorQTy;
}

namespace {
struct PrivateHelpersTy {
  PrivateHelpersTy(const VarDecl *Original, const VarDecl *PrivateCopy,
                   const VarDecl *PrivateElemInit)
      : Original(Original), PrivateCopy(PrivateCopy),
        PrivateElemInit(PrivateElemInit) {}
  const VarDecl *Original;
  const VarDecl *PrivateCopy;
  const VarDecl *PrivateElemInit;
};
typedef std::pair<CharUnits /*Align*/, PrivateHelpersTy> PrivateDataTy;
} // anonymous namespace

static RecordDecl *
createPrivatesRecordDecl(CodeGenModule &CGM, ArrayRef<PrivateDataTy> Privates) {
  if (!Privates.empty()) {
    ASTContext &C = CGM.getContext();
    // Build struct .kmp_privates_t. {
    //         /*  private vars  */
    //       };
    RecordDecl *RD = C.buildImplicitRecord(".kmp_privates.t");
    RD->startDefinition();
    for (const auto &Pair : Privates) {
      const VarDecl *VD = Pair.second.Original;
      QualType Type = VD->getType().getNonReferenceType();
      FieldDecl *FD = addFieldToRecordDecl(C, RD, Type);
      if (VD->hasAttrs()) {
        for (specific_attr_iterator<AlignedAttr> I(VD->getAttrs().begin()),
             E(VD->getAttrs().end());
             I != E; ++I)
          FD->addAttr(*I);
      }
    }
    RD->completeDefinition();
    return RD;
  }
  return nullptr;
}

static RecordDecl *
createKmpTaskTRecordDecl(CodeGenModule &CGM, OpenMPDirectiveKind Kind,
                         QualType KmpInt32Ty,
                         QualType KmpRoutineEntryPointerQTy) {
  ASTContext &C = CGM.getContext();
  // Build struct kmp_task_t {
  //         void *              shareds;
  //         kmp_routine_entry_t routine;
  //         kmp_int32           part_id;
  //         kmp_cmplrdata_t data1;
  //         kmp_cmplrdata_t data2;
  // For taskloops additional fields:
  //         kmp_uint64          lb;
  //         kmp_uint64          ub;
  //         kmp_int64           st;
  //         kmp_int32           liter;
  //         void *              reductions;
  //       };
  RecordDecl *UD = C.buildImplicitRecord("kmp_cmplrdata_t", TTK_Union);
  UD->startDefinition();
  addFieldToRecordDecl(C, UD, KmpInt32Ty);
  addFieldToRecordDecl(C, UD, KmpRoutineEntryPointerQTy);
  UD->completeDefinition();
  QualType KmpCmplrdataTy = C.getRecordType(UD);
  RecordDecl *RD = C.buildImplicitRecord("kmp_task_t");
  RD->startDefinition();
  addFieldToRecordDecl(C, RD, C.VoidPtrTy);
  addFieldToRecordDecl(C, RD, KmpRoutineEntryPointerQTy);
  addFieldToRecordDecl(C, RD, KmpInt32Ty);
  addFieldToRecordDecl(C, RD, KmpCmplrdataTy);
  addFieldToRecordDecl(C, RD, KmpCmplrdataTy);
  if (isOpenMPTaskLoopDirective(Kind)) {
    QualType KmpUInt64Ty =
        CGM.getContext().getIntTypeForBitwidth(/*DestWidth=*/64, /*Signed=*/0);
    QualType KmpInt64Ty =
        CGM.getContext().getIntTypeForBitwidth(/*DestWidth=*/64, /*Signed=*/1);
    addFieldToRecordDecl(C, RD, KmpUInt64Ty);
    addFieldToRecordDecl(C, RD, KmpUInt64Ty);
    addFieldToRecordDecl(C, RD, KmpInt64Ty);
    addFieldToRecordDecl(C, RD, KmpInt32Ty);
    addFieldToRecordDecl(C, RD, C.VoidPtrTy);
  }
  RD->completeDefinition();
  return RD;
}

static RecordDecl *
createKmpTaskTWithPrivatesRecordDecl(CodeGenModule &CGM, QualType KmpTaskTQTy,
                                     ArrayRef<PrivateDataTy> Privates) {
  ASTContext &C = CGM.getContext();
  // Build struct kmp_task_t_with_privates {
  //         kmp_task_t task_data;
  //         .kmp_privates_t. privates;
  //       };
  RecordDecl *RD = C.buildImplicitRecord("kmp_task_t_with_privates");
  RD->startDefinition();
  addFieldToRecordDecl(C, RD, KmpTaskTQTy);
  if (const RecordDecl *PrivateRD = createPrivatesRecordDecl(CGM, Privates))
    addFieldToRecordDecl(C, RD, C.getRecordType(PrivateRD));
  RD->completeDefinition();
  return RD;
}

/// Emit a proxy function which accepts kmp_task_t as the second
/// argument.
/// \code
/// kmp_int32 .omp_task_entry.(kmp_int32 gtid, kmp_task_t *tt) {
///   TaskFunction(gtid, tt->part_id, &tt->privates, task_privates_map, tt,
///   For taskloops:
///   tt->task_data.lb, tt->task_data.ub, tt->task_data.st, tt->task_data.liter,
///   tt->reductions, tt->shareds);
///   return 0;
/// }
/// \endcode
static llvm::Function *
emitProxyTaskFunction(CodeGenModule &CGM, SourceLocation Loc,
                      OpenMPDirectiveKind Kind, QualType KmpInt32Ty,
                      QualType KmpTaskTWithPrivatesPtrQTy,
                      QualType KmpTaskTWithPrivatesQTy, QualType KmpTaskTQTy,
                      QualType SharedsPtrTy, llvm::Function *TaskFunction,
                      llvm::Value *TaskPrivatesMap) {
  ASTContext &C = CGM.getContext();
  FunctionArgList Args;
  ImplicitParamDecl GtidArg(C, /*DC=*/nullptr, Loc, /*Id=*/nullptr, KmpInt32Ty,
                            ImplicitParamDecl::Other);
  ImplicitParamDecl TaskTypeArg(C, /*DC=*/nullptr, Loc, /*Id=*/nullptr,
                                KmpTaskTWithPrivatesPtrQTy.withRestrict(),
                                ImplicitParamDecl::Other);
  Args.push_back(&GtidArg);
  Args.push_back(&TaskTypeArg);
  const auto &TaskEntryFnInfo =
      CGM.getTypes().arrangeBuiltinFunctionDeclaration(KmpInt32Ty, Args);
  llvm::FunctionType *TaskEntryTy =
      CGM.getTypes().GetFunctionType(TaskEntryFnInfo);
  std::string Name = CGM.getOpenMPRuntime().getName({"omp_task_entry", ""});
  auto *TaskEntry = llvm::Function::Create(
      TaskEntryTy, llvm::GlobalValue::InternalLinkage, Name, &CGM.getModule());
  CGM.SetInternalFunctionAttributes(GlobalDecl(), TaskEntry, TaskEntryFnInfo);
  TaskEntry->setDoesNotRecurse();
  CodeGenFunction CGF(CGM);
  CGF.StartFunction(GlobalDecl(), KmpInt32Ty, TaskEntry, TaskEntryFnInfo, Args,
                    Loc, Loc);

  // TaskFunction(gtid, tt->task_data.part_id, &tt->privates, task_privates_map,
  // tt,
  // For taskloops:
  // tt->task_data.lb, tt->task_data.ub, tt->task_data.st, tt->task_data.liter,
  // tt->task_data.shareds);
  llvm::Value *GtidParam = CGF.EmitLoadOfScalar(
      CGF.GetAddrOfLocalVar(&GtidArg), /*Volatile=*/false, KmpInt32Ty, Loc);
  LValue TDBase = CGF.EmitLoadOfPointerLValue(
      CGF.GetAddrOfLocalVar(&TaskTypeArg),
      KmpTaskTWithPrivatesPtrQTy->castAs<PointerType>());
  const auto *KmpTaskTWithPrivatesQTyRD =
      cast<RecordDecl>(KmpTaskTWithPrivatesQTy->getAsTagDecl());
  LValue Base =
      CGF.EmitLValueForField(TDBase, *KmpTaskTWithPrivatesQTyRD->field_begin());
  const auto *KmpTaskTQTyRD = cast<RecordDecl>(KmpTaskTQTy->getAsTagDecl());
  auto PartIdFI = std::next(KmpTaskTQTyRD->field_begin(), KmpTaskTPartId);
  LValue PartIdLVal = CGF.EmitLValueForField(Base, *PartIdFI);
  llvm::Value *PartidParam = PartIdLVal.getPointer();

  auto SharedsFI = std::next(KmpTaskTQTyRD->field_begin(), KmpTaskTShareds);
  LValue SharedsLVal = CGF.EmitLValueForField(Base, *SharedsFI);
  llvm::Value *SharedsParam = CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
      CGF.EmitLoadOfScalar(SharedsLVal, Loc),
      CGF.ConvertTypeForMem(SharedsPtrTy));

  auto PrivatesFI = std::next(KmpTaskTWithPrivatesQTyRD->field_begin(), 1);
  llvm::Value *PrivatesParam;
  if (PrivatesFI != KmpTaskTWithPrivatesQTyRD->field_end()) {
    LValue PrivatesLVal = CGF.EmitLValueForField(TDBase, *PrivatesFI);
    PrivatesParam = CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
        PrivatesLVal.getPointer(), CGF.VoidPtrTy);
  } else {
    PrivatesParam = llvm::ConstantPointerNull::get(CGF.VoidPtrTy);
  }

  llvm::Value *CommonArgs[] = {GtidParam, PartidParam, PrivatesParam,
                               TaskPrivatesMap,
                               CGF.Builder
                                   .CreatePointerBitCastOrAddrSpaceCast(
                                       TDBase.getAddress(), CGF.VoidPtrTy)
                                   .getPointer()};
  SmallVector<llvm::Value *, 16> CallArgs(std::begin(CommonArgs),
                                          std::end(CommonArgs));
  if (isOpenMPTaskLoopDirective(Kind)) {
    auto LBFI = std::next(KmpTaskTQTyRD->field_begin(), KmpTaskTLowerBound);
    LValue LBLVal = CGF.EmitLValueForField(Base, *LBFI);
    llvm::Value *LBParam = CGF.EmitLoadOfScalar(LBLVal, Loc);
    auto UBFI = std::next(KmpTaskTQTyRD->field_begin(), KmpTaskTUpperBound);
    LValue UBLVal = CGF.EmitLValueForField(Base, *UBFI);
    llvm::Value *UBParam = CGF.EmitLoadOfScalar(UBLVal, Loc);
    auto StFI = std::next(KmpTaskTQTyRD->field_begin(), KmpTaskTStride);
    LValue StLVal = CGF.EmitLValueForField(Base, *StFI);
    llvm::Value *StParam = CGF.EmitLoadOfScalar(StLVal, Loc);
    auto LIFI = std::next(KmpTaskTQTyRD->field_begin(), KmpTaskTLastIter);
    LValue LILVal = CGF.EmitLValueForField(Base, *LIFI);
    llvm::Value *LIParam = CGF.EmitLoadOfScalar(LILVal, Loc);
    auto RFI = std::next(KmpTaskTQTyRD->field_begin(), KmpTaskTReductions);
    LValue RLVal = CGF.EmitLValueForField(Base, *RFI);
    llvm::Value *RParam = CGF.EmitLoadOfScalar(RLVal, Loc);
    CallArgs.push_back(LBParam);
    CallArgs.push_back(UBParam);
    CallArgs.push_back(StParam);
    CallArgs.push_back(LIParam);
    CallArgs.push_back(RParam);
  }
  CallArgs.push_back(SharedsParam);

  CGM.getOpenMPRuntime().emitOutlinedFunctionCall(CGF, Loc, TaskFunction,
                                                  CallArgs);
  CGF.EmitStoreThroughLValue(RValue::get(CGF.Builder.getInt32(/*C=*/0)),
                             CGF.MakeAddrLValue(CGF.ReturnValue, KmpInt32Ty));
  CGF.FinishFunction();
  return TaskEntry;
}

static llvm::Value *emitDestructorsFunction(CodeGenModule &CGM,
                                            SourceLocation Loc,
                                            QualType KmpInt32Ty,
                                            QualType KmpTaskTWithPrivatesPtrQTy,
                                            QualType KmpTaskTWithPrivatesQTy) {
  ASTContext &C = CGM.getContext();
  FunctionArgList Args;
  ImplicitParamDecl GtidArg(C, /*DC=*/nullptr, Loc, /*Id=*/nullptr, KmpInt32Ty,
                            ImplicitParamDecl::Other);
  ImplicitParamDecl TaskTypeArg(C, /*DC=*/nullptr, Loc, /*Id=*/nullptr,
                                KmpTaskTWithPrivatesPtrQTy.withRestrict(),
                                ImplicitParamDecl::Other);
  Args.push_back(&GtidArg);
  Args.push_back(&TaskTypeArg);
  const auto &DestructorFnInfo =
      CGM.getTypes().arrangeBuiltinFunctionDeclaration(KmpInt32Ty, Args);
  llvm::FunctionType *DestructorFnTy =
      CGM.getTypes().GetFunctionType(DestructorFnInfo);
  std::string Name =
      CGM.getOpenMPRuntime().getName({"omp_task_destructor", ""});
  auto *DestructorFn =
      llvm::Function::Create(DestructorFnTy, llvm::GlobalValue::InternalLinkage,
                             Name, &CGM.getModule());
  CGM.SetInternalFunctionAttributes(GlobalDecl(), DestructorFn,
                                    DestructorFnInfo);
  DestructorFn->setDoesNotRecurse();
  CodeGenFunction CGF(CGM);
  CGF.StartFunction(GlobalDecl(), KmpInt32Ty, DestructorFn, DestructorFnInfo,
                    Args, Loc, Loc);

  LValue Base = CGF.EmitLoadOfPointerLValue(
      CGF.GetAddrOfLocalVar(&TaskTypeArg),
      KmpTaskTWithPrivatesPtrQTy->castAs<PointerType>());
  const auto *KmpTaskTWithPrivatesQTyRD =
      cast<RecordDecl>(KmpTaskTWithPrivatesQTy->getAsTagDecl());
  auto FI = std::next(KmpTaskTWithPrivatesQTyRD->field_begin());
  Base = CGF.EmitLValueForField(Base, *FI);
  for (const auto *Field :
       cast<RecordDecl>(FI->getType()->getAsTagDecl())->fields()) {
    if (QualType::DestructionKind DtorKind =
            Field->getType().isDestructedType()) {
      LValue FieldLValue = CGF.EmitLValueForField(Base, Field);
      CGF.pushDestroy(DtorKind, FieldLValue.getAddress(), Field->getType());
    }
  }
  CGF.FinishFunction();
  return DestructorFn;
}

/// Emit a privates mapping function for correct handling of private and
/// firstprivate variables.
/// \code
/// void .omp_task_privates_map.(const .privates. *noalias privs, <ty1>
/// **noalias priv1,...,  <tyn> **noalias privn) {
///   *priv1 = &.privates.priv1;
///   ...;
///   *privn = &.privates.privn;
/// }
/// \endcode
static llvm::Value *
emitTaskPrivateMappingFunction(CodeGenModule &CGM, SourceLocation Loc,
                               ArrayRef<const Expr *> PrivateVars,
                               ArrayRef<const Expr *> FirstprivateVars,
                               ArrayRef<const Expr *> LastprivateVars,
                               QualType PrivatesQTy,
                               ArrayRef<PrivateDataTy> Privates) {
  ASTContext &C = CGM.getContext();
  FunctionArgList Args;
  ImplicitParamDecl TaskPrivatesArg(
      C, /*DC=*/nullptr, Loc, /*Id=*/nullptr,
      C.getPointerType(PrivatesQTy).withConst().withRestrict(),
      ImplicitParamDecl::Other);
  Args.push_back(&TaskPrivatesArg);
  llvm::DenseMap<const VarDecl *, unsigned> PrivateVarsPos;
  unsigned Counter = 1;
  for (const Expr *E : PrivateVars) {
    Args.push_back(ImplicitParamDecl::Create(
        C, /*DC=*/nullptr, Loc, /*Id=*/nullptr,
        C.getPointerType(C.getPointerType(E->getType()))
            .withConst()
            .withRestrict(),
        ImplicitParamDecl::Other));
    const auto *VD = cast<VarDecl>(cast<DeclRefExpr>(E)->getDecl());
    PrivateVarsPos[VD] = Counter;
    ++Counter;
  }
  for (const Expr *E : FirstprivateVars) {
    Args.push_back(ImplicitParamDecl::Create(
        C, /*DC=*/nullptr, Loc, /*Id=*/nullptr,
        C.getPointerType(C.getPointerType(E->getType()))
            .withConst()
            .withRestrict(),
        ImplicitParamDecl::Other));
    const auto *VD = cast<VarDecl>(cast<DeclRefExpr>(E)->getDecl());
    PrivateVarsPos[VD] = Counter;
    ++Counter;
  }
  for (const Expr *E : LastprivateVars) {
    Args.push_back(ImplicitParamDecl::Create(
        C, /*DC=*/nullptr, Loc, /*Id=*/nullptr,
        C.getPointerType(C.getPointerType(E->getType()))
            .withConst()
            .withRestrict(),
        ImplicitParamDecl::Other));
    const auto *VD = cast<VarDecl>(cast<DeclRefExpr>(E)->getDecl());
    PrivateVarsPos[VD] = Counter;
    ++Counter;
  }
  const auto &TaskPrivatesMapFnInfo =
      CGM.getTypes().arrangeBuiltinFunctionDeclaration(C.VoidTy, Args);
  llvm::FunctionType *TaskPrivatesMapTy =
      CGM.getTypes().GetFunctionType(TaskPrivatesMapFnInfo);
  std::string Name =
      CGM.getOpenMPRuntime().getName({"omp_task_privates_map", ""});
  auto *TaskPrivatesMap = llvm::Function::Create(
      TaskPrivatesMapTy, llvm::GlobalValue::InternalLinkage, Name,
      &CGM.getModule());
  CGM.SetInternalFunctionAttributes(GlobalDecl(), TaskPrivatesMap,
                                    TaskPrivatesMapFnInfo);
  TaskPrivatesMap->removeFnAttr(llvm::Attribute::NoInline);
  TaskPrivatesMap->removeFnAttr(llvm::Attribute::OptimizeNone);
  TaskPrivatesMap->addFnAttr(llvm::Attribute::AlwaysInline);
  CodeGenFunction CGF(CGM);
  CGF.StartFunction(GlobalDecl(), C.VoidTy, TaskPrivatesMap,
                    TaskPrivatesMapFnInfo, Args, Loc, Loc);

  // *privi = &.privates.privi;
  LValue Base = CGF.EmitLoadOfPointerLValue(
      CGF.GetAddrOfLocalVar(&TaskPrivatesArg),
      TaskPrivatesArg.getType()->castAs<PointerType>());
  const auto *PrivatesQTyRD = cast<RecordDecl>(PrivatesQTy->getAsTagDecl());
  Counter = 0;
  for (const FieldDecl *Field : PrivatesQTyRD->fields()) {
    LValue FieldLVal = CGF.EmitLValueForField(Base, Field);
    const VarDecl *VD = Args[PrivateVarsPos[Privates[Counter].second.Original]];
    LValue RefLVal =
        CGF.MakeAddrLValue(CGF.GetAddrOfLocalVar(VD), VD->getType());
    LValue RefLoadLVal = CGF.EmitLoadOfPointerLValue(
        RefLVal.getAddress(), RefLVal.getType()->castAs<PointerType>());
    CGF.EmitStoreOfScalar(FieldLVal.getPointer(), RefLoadLVal);
    ++Counter;
  }
  CGF.FinishFunction();
  return TaskPrivatesMap;
}

static bool stable_sort_comparator(const PrivateDataTy P1,
                                   const PrivateDataTy P2) {
  return P1.first > P2.first;
}

/// Emit initialization for private variables in task-based directives.
static void emitPrivatesInit(CodeGenFunction &CGF,
                             const OMPExecutableDirective &D,
                             Address KmpTaskSharedsPtr, LValue TDBase,
                             const RecordDecl *KmpTaskTWithPrivatesQTyRD,
                             QualType SharedsTy, QualType SharedsPtrTy,
                             const OMPTaskDataTy &Data,
                             ArrayRef<PrivateDataTy> Privates, bool ForDup) {
  ASTContext &C = CGF.getContext();
  auto FI = std::next(KmpTaskTWithPrivatesQTyRD->