lib/CodeGen/CGOpenMPRuntime.cpp

//===----- 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)
};

namespace {
LLVM_ENABLE_BITMASK_ENUMS_IN_NAMESPACE();
/// Values for bit flags for marking which requires clauses have been used.
enum OpenMPOffloadingRequiresDirFlags : int64_t {
  /// flag undefined.
  OMP_REQ_UNDEFINED               = 0x000,
  /// no requires clause present.
  OMP_REQ_NONE                    = 0x001,
  /// reverse_offload clause.
  OMP_REQ_REVERSE_OFFLOAD         = 0x002,
  /// unified_address clause.
  OMP_REQ_UNIFIED_ADDRESS         = 0x004,
  /// unified_shared_memory clause.
  OMP_REQ_UNIFIED_SHARED_MEMORY   = 0x008,
  /// dynamic_allocators clause.
  OMP_REQ_DYNAMIC_ALLOCATORS      = 0x010,
  LLVM_MARK_AS_BITMASK_ENUM(/*LargestValue=*/OMP_REQ_DYNAMIC_ALLOCATORS)
};

enum OpenMPOffloadingReservedDeviceIDs {
  /// Device ID if the device was not defined, runtime should get it
  /// from environment variables in the spec.
  OMP_DEVICEID_UNDEF = -1,
};
} // anonymous namespace

/// 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_task_t * __kmpc_omp_target_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,
  // kmp_int64 device_id);
  OMPRTL__kmpc_omp_target_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,
  // Call to void *__kmpc_alloc(int gtid, size_t sz, omp_allocator_handle_t al);
  OMPRTL__kmpc_alloc,
  // Call to void __kmpc_free(int gtid, void *ptr, omp_allocator_handle_t al);
  OMPRTL__kmpc_free,

  //
  // 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, int64_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, int64_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, int64_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, int64_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_requires(int64_t flags);
  OMPRTL__tgt_register_requires,
  // 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, int64_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, int64_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, int64_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, int64_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, int64_t *arg_sizes, int64_t
  // *arg_types);
  OMPRTL__tgt_target_data_update_nowait,
  // Call to int64_t __tgt_mapper_num_components(void *rt_mapper_handle);
  OMPRTL__tgt_mapper_num_components,
  // Call to void __tgt_push_mapper_component(void *rt_mapper_handle, void
  // *base, void *begin, int64_t size, int64_t type);
  OMPRTL__tgt_push_mapper_component,
};

/// 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();
}

bool CGOpenMPRuntime::tryEmitDeclareVariant(const GlobalDecl &NewGD,
                                            const GlobalDecl &OldGD,
                                            llvm::GlobalValue *OrigAddr,
                                            bool IsForDefinition) {
  // Emit at least a definition for the aliasee if the the address of the
  // original function is requested.
  if (IsForDefinition || OrigAddr)
    (void)CGM.GetAddrOfGlobal(NewGD);
  StringRef NewMangledName = CGM.getMangledName(NewGD);
  llvm::GlobalValue *Addr = CGM.GetGlobalValue(NewMangledName);
  if (Addr && !Addr->isDeclaration()) {
    const auto *D = cast<FunctionDecl>(OldGD.getDecl());
    const CGFunctionInfo &FI = CGM.getTypes().arrangeGlobalDeclaration(OldGD);
    llvm::Type *DeclTy = CGM.getTypes().GetFunctionType(FI);

    // Create a reference to the named value.  This ensures that it is emitted
    // if a deferred decl.
    llvm::GlobalValue::LinkageTypes LT = CGM.getFunctionLinkage(OldGD);

    // Create the new alias itself, but don't set a name yet.
    auto *GA =
        llvm::GlobalAlias::create(DeclTy, 0, LT, "", Addr, &CGM.getModule());

    if (OrigAddr) {
      assert(OrigAddr->isDeclaration() && "Expected declaration");

      GA->takeName(OrigAddr);
      OrigAddr->replaceAllUsesWith(
          llvm::ConstantExpr::getBitCast(GA, OrigAddr->getType()));
      OrigAddr->eraseFromParent();
    } else {
      GA->setName(CGM.getMangledName(OldGD));
    }

    // Set attributes which are particular to an alias; this is a
    // specialization of the attributes which may be set on a global function.
    if (D->hasAttr<WeakAttr>() || D->hasAttr<WeakRefAttr>() ||
        D->isWeakImported())
      GA->setLinkage(llvm::Function::WeakAnyLinkage);

    CGM.SetCommonAttributes(OldGD, GA);
    return true;
  }
  return false;
}

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();
  }
  // Emit aliases for the deferred aliasees.
  for (const auto &Pair : DeferredVariantFunction) {
    StringRef MangledName = CGM.getMangledName(Pair.second.second);
    llvm::GlobalValue *Addr = CGM.GetGlobalValue(MangledName);
    // If not able to emit alias, just emit original declaration.
    (void)tryEmitDeclareVariant(Pair.second.first, Pair.second.second, Addr,
                                /*IsForDefinition=*/false);
  }
}

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);
  if (CGM.getLangOpts().Optimize) {
    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);
  }
  auto I = FunctionUDMMap.find(CGF.CurFn);
  if (I != FunctionUDMMap.end()) {
    for(auto *D : I->second)
      UDMMap.erase(D);
    FunctionUDMMap.erase(I);
  }
}

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_target_task_alloc: {
    // Build kmp_task_t *__kmpc_omp_target_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, kmp_int64 device_id);
    assert(KmpRoutineEntryPtrTy != nullptr &&
           "Type kmp_routine_entry_t must be created.");
    llvm::Type *TypeParams[] = {getIdentTyPointerTy(), CGM.Int32Ty, CGM.Int32Ty,
                                CGM.SizeTy, CGM.SizeTy, KmpRoutineEntryPtrTy,
                                CGM.Int64Ty};
    // 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_target_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_alloc: {
    // Build to void *__kmpc_alloc(int gtid, size_t sz, omp_allocator_handle_t
    // al); omp_allocator_handle_t type is void *.
    llvm::Type *TypeParams[] = {CGM.IntTy, CGM.SizeTy, CGM.VoidPtrTy};
    auto *FnTy =
        llvm::FunctionType::get(CGM.VoidPtrTy, TypeParams, /*isVarArg=*/false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, /*Name=*/"__kmpc_alloc");
    break;
  }
  case OMPRTL__kmpc_free: {
    // Build to void __kmpc_free(int gtid, void *ptr, omp_allocator_handle_t
    // al); omp_allocator_handle_t type is void *.
    llvm::Type *TypeParams[] = {CGM.IntTy, CGM.VoidPtrTy, CGM.VoidPtrTy};
    auto *FnTy =
        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg=*/false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, /*Name=*/"__kmpc_free");
    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, int64_t *arg_sizes, int64_t
    // *arg_types);
    llvm::Type *TypeParams[] = {CGM.Int64Ty,
                                CGM.VoidPtrTy,
                                CGM.Int32Ty,
                                CGM.VoidPtrPtrTy,
                                CGM.VoidPtrPtrTy,
                                CGM.Int64Ty->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, int64_t *arg_sizes,
    // int64_t *arg_types);
    llvm::Type *TypeParams[] = {CGM.Int64Ty,
                                CGM.VoidPtrTy,
                                CGM.Int32Ty,
                                CGM.VoidPtrPtrTy,
                                CGM.VoidPtrPtrTy,
                                CGM.Int64Ty->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, int64_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.Int64Ty->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, int64_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.Int64Ty->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_requires: {
    // Build void __tgt_register_requires(int64_t flags);
    llvm::Type *TypeParams[] = {CGM.Int64Ty};
    auto *FnTy =
        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg*/ false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, "__tgt_register_requires");
    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, int64_t *arg_sizes, int64_t *arg_types);
    llvm::Type *TypeParams[] = {CGM.Int64Ty,
                                CGM.Int32Ty,
                                CGM.VoidPtrPtrTy,
                                CGM.VoidPtrPtrTy,
                                CGM.Int64Ty->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, int64_t *arg_sizes, int64_t
    // *arg_types);
    llvm::Type *TypeParams[] = {CGM.Int64Ty,
                                CGM.Int32Ty,
                                CGM.VoidPtrPtrTy,
                                CGM.VoidPtrPtrTy,
                                CGM.Int64Ty->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, int64_t *arg_sizes, int64_t *arg_types);
    llvm::Type *TypeParams[] = {CGM.Int64Ty,
                                CGM.Int32Ty,
                                CGM.VoidPtrPtrTy,
                                CGM.VoidPtrPtrTy,
                                CGM.Int64Ty->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, int64_t *arg_sizes, int64_t
    // *arg_types);
    llvm::Type *TypeParams[] = {CGM.Int64Ty,
                                CGM.Int32Ty,
                                CGM.VoidPtrPtrTy,
                                CGM.VoidPtrPtrTy,
                                CGM.Int64Ty->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, int64_t *arg_sizes, int64_t *arg_types);
    llvm::Type *TypeParams[] = {CGM.Int64Ty,
                                CGM.Int32Ty,
                                CGM.VoidPtrPtrTy,
                                CGM.VoidPtrPtrTy,
                                CGM.Int64Ty->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, int64_t *arg_sizes, int64_t
    // *arg_types);
    llvm::Type *TypeParams[] = {CGM.Int64Ty,
                                CGM.Int32Ty,
                                CGM.VoidPtrPtrTy,
                                CGM.VoidPtrPtrTy,
                                CGM.Int64Ty->getPointerTo(),
                                CGM.Int64Ty->getPointerTo()};
    auto *FnTy =
        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg=*/false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, "__tgt_target_data_update_nowait");
    break;
  }
  case OMPRTL__tgt_mapper_num_components: {
    // Build int64_t __tgt_mapper_num_components(void *rt_mapper_handle);
    llvm::Type *TypeParams[] = {CGM.VoidPtrTy};
    auto *FnTy =
        llvm::FunctionType::get(CGM.Int64Ty, TypeParams, /*isVarArg*/ false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, "__tgt_mapper_num_components");
    break;
  }
  case OMPRTL__tgt_push_mapper_component: {
    // Build void __tgt_push_mapper_component(void *rt_mapper_handle, void
    // *base, void *begin, int64_t size, int64_t type);
    llvm::Type *TypeParams[] = {CGM.VoidPtrTy, CGM.VoidPtrTy, CGM.VoidPtrTy,
                                CGM.Int64Ty, CGM.Int64Ty};
    auto *FnTy =
        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg*/ false);
    RTLFn = CGM.CreateRuntimeFunction(FnTy, "__tgt_push_mapper_component");
    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,