CGOpenMPRuntime.cpp

Go to the documentation of this file.

//===----- CGOpenMPRuntime.cpp - Interface to OpenMP Runtimes -------------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
//
// This provides a class for OpenMP runtime code generation.
//
//===----------------------------------------------------------------------===//
 
#include "CGOpenMPRuntime.h"
#include "CGCXXABI.h"
#include "CGCleanup.h"
#include "CGRecordLayout.h"
#include "CodeGenFunction.h"
#include "TargetInfo.h"
#include "clang/AST/APValue.h"
#include "clang/AST/Attr.h"
#include "clang/AST/Decl.h"
#include "clang/AST/OpenMPClause.h"
#include "clang/AST/StmtOpenMP.h"
#include "clang/AST/StmtVisitor.h"
#include "clang/Basic/BitmaskEnum.h"
#include "clang/Basic/FileManager.h"
#include "clang/Basic/OpenMPKinds.h"
#include "clang/Basic/SourceManager.h"
#include "clang/CodeGen/ConstantInitBuilder.h"
#include "llvm/ADT/ArrayRef.h"
#include "llvm/ADT/SetOperations.h"
#include "llvm/ADT/SmallBitVector.h"
#include "llvm/ADT/StringExtras.h"
#include "llvm/Bitcode/BitcodeReader.h"
#include "llvm/IR/Constants.h"
#include "llvm/IR/DerivedTypes.h"
#include "llvm/IR/GlobalValue.h"
#include "llvm/IR/InstrTypes.h"
#include "llvm/IR/Value.h"
#include "llvm/Support/AtomicOrdering.h"
#include "llvm/Support/Format.h"
#include "llvm/Support/raw_ostream.h"
#include <cassert>
#include <numeric>
#include <optional>
 
using namespace clang;
using namespace CodeGen;
using namespace llvm::omp;
 
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.EmitBranch(CGF.ReturnBlock.getBlock());
        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.EmitLValue(&DRE).getAddress(CGF));
    }
    (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 ValueDecl *, FieldDecl *> LambdaCaptureFields;
  FieldDecl *LambdaThisCaptureField = nullptr;
  const CodeGen::CGBlockInfo *BlockInfo = nullptr;
  bool NoInheritance = false;
 
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,
                          bool NoInheritance = true)
      : CGF(CGF), NoInheritance(NoInheritance) {
    // Start emission for the construct.
    CGF.CapturedStmtInfo = new CGOpenMPInlinedRegionInfo(
        CGF.CapturedStmtInfo, CodeGen, Kind, HasCancel);
    if (NoInheritance) {
      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;
    if (NoInheritance) {
      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/main/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)
};
 
} // anonymous namespace
 
/// Describes ident structure that describes a source location.
/// All descriptions are taken from
/// https://github.com/llvm/llvm-project/blob/main/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),
};
 
/// 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()), Private);
    PrivateScope.addPrivate(cast<VarDecl>(RHSDRE->getDecl()), 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: {
      OpaqueValueExpr OVE(DRD->getLocation(), Ty, VK_LValue);
      CodeGenFunction::OpaqueValueMapping OpaqueMap(CGF, &OVE, LV);
      CGF.EmitAnyExprToMem(&OVE, Private, Ty.getQualifiers(),
                           /*IsInitializer=*/false);
      return;
    }
    }
    OpaqueValueExpr OVE(DRD->getLocation(), Ty, VK_PRValue);
    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);
  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(DestAddr.getElementType(), 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.getElementType(),
                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.getElementType(),
              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(
        SrcAddr.getElementType(), 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(
      DestAddr.getElementType(), 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, Address SharedAddr,
    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, SharedAddr);
}
 
ReductionCodeGen::ReductionCodeGen(ArrayRef<const Expr *> Shareds,
                                   ArrayRef<const Expr *> Origs,
                                   ArrayRef<const Expr *> Privates,
                                   ArrayRef<const Expr *> ReductionOps) {
  ClausesData.reserve(Shareds.size());
  SharedAddresses.reserve(Shareds.size());
  Sizes.reserve(Shareds.size());
  BaseDecls.reserve(Shareds.size());
  const auto *IOrig = Origs.begin();
  const auto *IPriv = Privates.begin();
  const auto *IRed = ReductionOps.begin();
  for (const Expr *Ref : Shareds) {
    ClausesData.emplace_back(Ref, *IOrig, *IPriv, *IRed);
    std::advance(IOrig, 1);
    std::advance(IPriv, 1);
    std::advance(IRed, 1);
  }
}
 
void ReductionCodeGen::emitSharedOrigLValue(CodeGenFunction &CGF, unsigned N) {
  assert(SharedAddresses.size() == N && OrigAddresses.size() == N &&
         "Number of generated lvalues must be exactly N.");
  LValue First = emitSharedLValue(CGF, ClausesData[N].Shared);
  LValue Second = emitSharedLValueUB(CGF, ClausesData[N].Shared);
  SharedAddresses.emplace_back(First, Second);
  if (ClausesData[N].Shared == ClausesData[N].Ref) {
    OrigAddresses.emplace_back(First, Second);
  } else {
    LValue First = emitSharedLValue(CGF, ClausesData[N].Ref);
    LValue Second = emitSharedLValueUB(CGF, ClausesData[N].Ref);
    OrigAddresses.emplace_back(First, Second);
  }
}
 
void ReductionCodeGen::emitAggregateType(CodeGenFunction &CGF, unsigned N) {
  QualType PrivateType = getPrivateType(N);
  bool AsArraySection = isa<OMPArraySectionExpr>(ClausesData[N].Ref);
  if (!PrivateType->isVariablyModifiedType()) {
    Sizes.emplace_back(
        CGF.getTypeSize(OrigAddresses[N].first.getType().getNonReferenceType()),
        nullptr);
    return;
  }
  llvm::Value *Size;
  llvm::Value *SizeInChars;
  auto *ElemType = OrigAddresses[N].first.getAddress(CGF).getElementType();
  auto *ElemSizeOf = llvm::ConstantExpr::getSizeOf(ElemType);
  if (AsArraySection) {
    Size = CGF.Builder.CreatePtrDiff(ElemType,
                                     OrigAddresses[N].second.getPointer(CGF),
                                     OrigAddresses[N].first.getPointer(CGF));
    Size = CGF.Builder.CreateNUWAdd(
        Size, llvm::ConstantInt::get(Size->getType(), /*V=*/1));
    SizeInChars = CGF.Builder.CreateNUWMul(Size, ElemSizeOf);
  } else {
    SizeInChars =
        CGF.getTypeSize(OrigAddresses[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) {
  QualType PrivateType = getPrivateType(N);
  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, Address SharedAddr,
    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);
  if (CGF.getContext().getAsArrayType(PrivateVD->getType())) {
    if (DRD && DRD->getInitializer())
      (void)DefaultInit(CGF);
    emitAggregateInitialization(CGF, N, PrivateAddr, SharedAddr, DRD);
  } else if (DRD && (DRD->getInitializer() || !PrivateVD->hasInit())) {
    (void)DefaultInit(CGF);
    QualType SharedType = SharedAddresses[N].first.getType();
    emitInitWithReductionInitializer(CGF, DRD, ClausesData[N].ReductionOp,
                                     PrivateAddr, SharedAddr, SharedType);
  } 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) {
  QualType PrivateType = getPrivateType(N);
  QualType::DestructionKind DTorKind = PrivateType.isDestructedType();
  return DTorKind != QualType::DK_none;
}
 
void ReductionCodeGen::emitCleanups(CodeGenFunction &CGF, unsigned N,
                                    Address PrivateAddr) {
  QualType PrivateType = getPrivateType(N);
  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(CGF), PtrTy);
    } else {
      LValue RefLVal = CGF.MakeAddrLValue(BaseLV.getAddress(CGF), BaseTy);
      BaseLV = CGF.EmitLoadOfReferenceLValue(RefLVal);
    }
    BaseTy = BaseTy->getPointeeType();
  }
  return CGF.MakeAddrLValue(
      CGF.Builder.CreateElementBitCast(BaseLV.getAddress(CGF),
                                       CGF.ConvertTypeForMem(ElTy)),
      BaseLV.getType(), BaseLV.getBaseInfo(),
      CGF.CGM.getTBAAInfoForSubobject(BaseLV, BaseLV.getType()));
}
 
static Address castToBase(CodeGenFunction &CGF, QualType BaseTy, QualType ElTy,
                          Address OriginalBaseAddress, 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();
  }
 
  if (Tmp.isValid()) {
    Addr = CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
        Addr, Tmp.getElementType());
    CGF.Builder.CreateStore(Addr, Tmp);
    return MostTopTmp;
  }
 
  Addr = CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
      Addr, OriginalBaseAddress.getType());
  return OriginalBaseAddress.withPointer(Addr, NotKnownNonNull);
}
 
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);
    Address SharedAddr = SharedAddresses[N].first.getAddress(CGF);
    llvm::Value *Adjustment = CGF.Builder.CreatePtrDiff(
        SharedAddr.getElementType(), BaseLValue.getPointer(CGF),
        SharedAddr.getPointer());
    llvm::Value *PrivatePointer =
        CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
            PrivateAddr.getPointer(), SharedAddr.getType());
    llvm::Value *Ptr = CGF.Builder.CreateGEP(
        SharedAddr.getElementType(), PrivatePointer, Adjustment);
    return castToBase(CGF, OrigVD->getType(),
                      SharedAddresses[N].first.getType(),
                      OriginalBaseLValue.getAddress(CGF), 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();
  if (S)
    CGF.incrementProfileCounter(S);
  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)
    : CGM(CGM), OMPBuilder(CGM.getModule()) {
  KmpCriticalNameTy = llvm::ArrayType::get(CGM.Int32Ty, /*NumElements*/ 8);
  llvm::OpenMPIRBuilderConfig Config(CGM.getLangOpts().OpenMPIsDevice, false,
                                     hasRequiresUnifiedSharedMemory(),
                                     CGM.getLangOpts().OpenMPOffloadMandatory);
  // Initialize Types used in OpenMPIRBuilder from OMPKinds.def
  OMPBuilder.initialize();
  OMPBuilder.setConfig(Config);
  loadOffloadInfoMetadata();
}
 
void CGOpenMPRuntime::clear() {
  InternalVars.clear();
  // Clean non-target variable declarations possibly used only in debug info.
  for (const auto &Data : EmittedNonTargetVariables) {
    if (!Data.getValue().pointsToAliveValue())
      continue;
    auto *GV = dyn_cast<llvm::GlobalVariable>(Data.getValue());
    if (!GV)
      continue;
    if (!GV->isDeclaration() || GV->getNumUses() > 0)
      continue;
    GV->eraseFromParent();
  }
}
 
std::string CGOpenMPRuntime::getName(ArrayRef<StringRef> Parts) const {
  return OMPBuilder.createPlatformSpecificName(Parts);
}
 
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.EmitLoadOfPointerLValue(AddrIn, PtrTy->castAs<PointerType>())
              .getAddress(CGF));
  Address AddrOut = CGF.GetAddrOfLocalVar(&OmpOutParm);
  Scope.addPrivate(
      Out, CGF.EmitLoadOfPointerLValue(AddrOut, PtrTy->castAs<PointerType>())
               .getAddress(CGF));
  (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);
}
 
namespace {
// Temporary RAII solution to perform a push/pop stack event on the OpenMP IR
// Builder if one is present.
struct PushAndPopStackRAII {
  PushAndPopStackRAII(llvm::OpenMPIRBuilder *OMPBuilder, CodeGenFunction &CGF,
                      bool HasCancel, llvm::omp::Directive Kind)
      : OMPBuilder(OMPBuilder) {
    if (!OMPBuilder)
      return;
 
    // The following callback is the crucial part of clangs cleanup process.
    //
    // NOTE:
    // Once the OpenMPIRBuilder is used to create parallel regions (and
    // similar), the cancellation destination (Dest below) is determined via
    // IP. That means if we have variables to finalize we split the block at IP,
    // use the new block (=BB) as destination to build a JumpDest (via
    // getJumpDestInCurrentScope(BB)) which then is fed to
    // EmitBranchThroughCleanup. Furthermore, there will not be the need
    // to push & pop an FinalizationInfo object.
    // The FiniCB will still be needed but at the point where the
    // OpenMPIRBuilder is asked to construct a parallel (or similar) construct.
    auto FiniCB = [&CGF](llvm::OpenMPIRBuilder::InsertPointTy IP) {
      assert(IP.getBlock()->end() == IP.getPoint() &&
             "Clang CG should cause non-terminated block!");
      CGBuilderTy::InsertPointGuard IPG(CGF.Builder);
      CGF.Builder.restoreIP(IP);
      CodeGenFunction::JumpDest Dest =
          CGF.getOMPCancelDestination(OMPD_parallel);
      CGF.EmitBranchThroughCleanup(Dest);
    };
 
    // TODO: Remove this once we emit parallel regions through the
    //       OpenMPIRBuilder as it can do this setup internally.
    llvm::OpenMPIRBuilder::FinalizationInfo FI({FiniCB, Kind, HasCancel});
    OMPBuilder->pushFinalizationCB(std::move(FI));
  }
  ~PushAndPopStackRAII() {
    if (OMPBuilder)
      OMPBuilder->popFinalizationCB();
  }
  llvm::OpenMPIRBuilder *OMPBuilder;
};
} // namespace
 
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 *OPD = dyn_cast<OMPTargetParallelDirective>(&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();
 
  // TODO: Temporarily inform the OpenMPIRBuilder, if any, about the new
  //       parallel region to make cancellation barriers work properly.
  llvm::OpenMPIRBuilder &OMPBuilder = CGM.getOpenMPRuntime().getOMPBuilder();
  PushAndPopStackRAII PSR(&OMPBuilder, CGF, HasCancel, InnermostKind);
  CGOpenMPOutlinedRegionInfo CGInfo(*CS, ThreadIDVar, CodeGen, InnermostKind,
                                    HasCancel, OutlinedHelperName);
  CodeGenFunction::CGCapturedStmtRAII CapInfoRAII(CGF, &CGInfo);
  return CGF.GenerateOpenMPCapturedStmtFunction(*CS, D.getBeginLoc());
}
 
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)};
    CGF.EmitRuntimeCall(OMPBuilder.getOrCreateRuntimeFunction(
                            CGM.getModule(), 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);
  bool HasCancel = false;
  if (const auto *TD = dyn_cast<OMPTaskDirective>(&D))
    HasCancel = TD->hasCancel();
  else if (const auto *TD = dyn_cast<OMPTaskLoopDirective>(&D))
    HasCancel = TD->hasCancel();
  else if (const auto *TD = dyn_cast<OMPMasterTaskLoopDirective>(&D))
    HasCancel = TD->hasCancel();
  else if (const auto *TD = dyn_cast<OMPParallelMasterTaskLoopDirective>(&D))
    HasCancel = TD->hasCancel();
 
  CodeGenFunction CGF(CGM, true);
  CGOpenMPTaskOutlinedRegionInfo CGInfo(*CS, ThreadIDVar, CodeGen,
                                        InnermostKind, HasCancel, Action);
  CodeGenFunction::CGCapturedStmtRAII CapInfoRAII(CGF, &CGInfo);
  llvm::Function *Res = CGF.GenerateCapturedStmtFunction(*CS);
  if (!Tied)
    NumberOfParts = Action.getNumberOfParts();
  return Res;
}
 
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();
  }
}
 
static StringRef getIdentStringFromSourceLocation(CodeGenFunction &CGF,
                                                  SourceLocation Loc,
                                                  SmallString<128> &Buffer) {
  llvm::raw_svector_ostream OS(Buffer);
  // Build debug location
  PresumedLoc PLoc = CGF.getContext().getSourceManager().getPresumedLoc(Loc);
  OS << ";" << PLoc.getFilename() << ";";
  if (const auto *FD = dyn_cast_or_null<FunctionDecl>(CGF.CurFuncDecl))
    OS << FD->getQualifiedNameAsString();
  OS << ";" << PLoc.getLine() << ";" << PLoc.getColumn() << ";;";
  return OS.str();
}
 
llvm::Value *CGOpenMPRuntime::emitUpdateLocation(CodeGenFunction &CGF,
                                                 SourceLocation Loc,
                                                 unsigned Flags, bool EmitLoc) {
  uint32_t SrcLocStrSize;
  llvm::Constant *SrcLocStr;
  if ((!EmitLoc && CGM.getCodeGenOpts().getDebugInfo() ==
                       llvm::codegenoptions::NoDebugInfo) ||
      Loc.isInvalid()) {
    SrcLocStr = OMPBuilder.getOrCreateDefaultSrcLocStr(SrcLocStrSize);
  } else {
    std::string FunctionName;
    if (const auto *FD = dyn_cast_or_null<FunctionDecl>(CGF.CurFuncDecl))
      FunctionName = FD->getQualifiedNameAsString();
    PresumedLoc PLoc = CGF.getContext().getSourceManager().getPresumedLoc(Loc);
    const char *FileName = PLoc.getFilename();
    unsigned Line = PLoc.getLine();
    unsigned Column = PLoc.getColumn();
    SrcLocStr = OMPBuilder.getOrCreateSrcLocStr(FunctionName, FileName, Line,
                                                Column, SrcLocStrSize);
  }
  unsigned Reserved2Flags = getDefaultLocationReserved2Flags();
  return OMPBuilder.getOrCreateIdent(
      SrcLocStr, SrcLocStrSize, llvm::omp::IdentFlag(Flags), Reserved2Flags);
}
 
llvm::Value *CGOpenMPRuntime::getThreadID(CodeGenFunction &CGF,
                                          SourceLocation Loc) {
  assert(CGF.CurFn && "No function in current CodeGenFunction.");
  // If the OpenMPIRBuilder is used we need to use it for all thread id calls as
  // the clang invariants used below might be broken.
  if (CGM.getLangOpts().OpenMPIRBuilder) {
    SmallString<128> Buffer;
    OMPBuilder.updateToLocation(CGF.Builder.saveIP());
    uint32_t SrcLocStrSize;
    auto *SrcLocStr = OMPBuilder.getOrCreateSrcLocStr(
        getIdentStringFromSourceLocation(CGF, Loc, Buffer), SrcLocStrSize);
    return OMPBuilder.getOrCreateThreadID(
        OMPBuilder.getOrCreateIdent(SrcLocStr, SrcLocStrSize));
  }
 
  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 (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);
      llvm::BasicBlock *TopBlock = CGF.AllocaInsertPt->getParent();
      if (!CGF.EHStack.requiresLandingPad() || !CGF.getLangOpts().Exceptions ||
          !CGF.getLangOpts().CXXExceptions ||
          CGF.Builder.GetInsertBlock() == TopBlock ||
          !isa<llvm::Instruction>(LVal.getPointer(CGF)) ||
          cast<llvm::Instruction>(LVal.getPointer(CGF))->getParent() ==
              TopBlock ||
          cast<llvm::Instruction>(LVal.getPointer(CGF))->getParent() ==
              CGF.Builder.GetInsertBlock()) {
        ThreadID = CGF.EmitLoadOfScalar(LVal, Loc);
        // If value loaded in entry block, cache it and use it everywhere in
        // function.
        if (CGF.Builder.GetInsertBlock() == TopBlock) {
          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(
      OMPBuilder.getOrCreateRuntimeFunction(CGM.getModule(),
                                            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(const auto *D : FunctionUDRMap[CGF.CurFn])
      UDRMap.erase(D);
    FunctionUDRMap.erase(CGF.CurFn);
  }
  auto I = FunctionUDMMap.find(CGF.CurFn);
  if (I != FunctionUDMMap.end()) {
    for(const auto *D : I->second)
      UDMMap.erase(D);
    FunctionUDMMap.erase(I);
  }
  LastprivateConditionalToTypes.erase(CGF.CurFn);
  FunctionToUntiedTaskStackMap.erase(CGF.CurFn);
}
 
llvm::Type *CGOpenMPRuntime::getIdentTyPointerTy() {
  return OMPBuilder.IdentPtr;
}
 
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::createForStaticInitFunction(unsigned IVSize, bool IVSigned,
                                             bool IsGPUDistribute) {
  assert((IVSize == 32 || IVSize == 64) &&
         "IV size is not compatible with the omp runtime");
  StringRef Name;
  if (IsGPUDistribute)
    Name = IVSize == 32 ? (IVSigned ? "__kmpc_distribute_static_init_4"
                                    : "__kmpc_distribute_static_init_4u")
                        : (IVSigned ? "__kmpc_distribute_static_init_8"
                                    : "__kmpc_distribute_static_init_8u");
  else
    Name = IVSize == 32 ? (IVSigned ? "__kmpc_for_static_init_4"
                                    : "__kmpc_for_static_init_4u")
                        : (IVSigned ? "__kmpc_for_static_init_8"
                                    : "__kmpc_for_static_init_8u");
 
  llvm::Type *ITy = IVSize == 32 ? CGM.Int32Ty : CGM.Int64Ty;
  auto *PtrTy = llvm::PointerType::getUnqual(ITy);
  llvm::Type *TypeParams[] = {
    getIdentTyPointerTy(),                     // loc
    CGM.Int32Ty,                               // tid
    CGM.Int32Ty,                               // schedtype
    llvm::PointerType::getUnqual(CGM.Int32Ty), // p_lastiter
    PtrTy,                                     // p_lower
    PtrTy,                                     // p_upper
    PtrTy,                                     // p_stride
    ITy,                                       // incr
    ITy                                        // chunk
  };
  auto *FnTy =
      llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg*/ false);
  return CGM.CreateRuntimeFunction(FnTy, Name);
}
 
llvm::FunctionCallee
CGOpenMPRuntime::createDispatchInitFunction(unsigned IVSize, bool IVSigned) {
  assert((IVSize == 32 || IVSize == 64) &&
         "IV size is not compatible with the omp runtime");
  StringRef Name =
      IVSize == 32
          ? (IVSigned ? "__kmpc_dispatch_init_4" : "__kmpc_dispatch_init_4u")
          : (IVSigned ? "__kmpc_dispatch_init_8" : "__kmpc_dispatch_init_8u");
  llvm::Type *ITy = IVSize == 32 ? CGM.Int32Ty : CGM.Int64Ty;
  llvm::Type *TypeParams[] = { getIdentTyPointerTy(), // loc
                               CGM.Int32Ty,           // tid
                               CGM.Int32Ty,           // schedtype
                               ITy,                   // lower
                               ITy,                   // upper
                               ITy,                   // stride
                               ITy                    // chunk
  };
  auto *FnTy =
      llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg*/ false);
  return CGM.CreateRuntimeFunction(FnTy, Name);
}
 
llvm::FunctionCallee
CGOpenMPRuntime::createDispatchFiniFunction(unsigned IVSize, bool IVSigned) {
  assert((IVSize == 32 || IVSize == 64) &&
         "IV size is not compatible with the omp runtime");
  StringRef Name =
      IVSize == 32
          ? (IVSigned ? "__kmpc_dispatch_fini_4" : "__kmpc_dispatch_fini_4u")
          : (IVSigned ? "__kmpc_dispatch_fini_8" : "__kmpc_dispatch_fini_8u");
  llvm::Type *TypeParams[] = {
      getIdentTyPointerTy(), // loc
      CGM.Int32Ty,           // tid
  };
  auto *FnTy =
      llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg=*/false);
  return CGM.CreateRuntimeFunction(FnTy, Name);
}
 
llvm::FunctionCallee
CGOpenMPRuntime::createDispatchNextFunction(unsigned IVSize, bool IVSigned) {
  assert((IVSize == 32 || IVSize == 64) &&
         "IV size is not compatible with the omp runtime");
  StringRef Name =
      IVSize == 32
          ? (IVSigned ? "__kmpc_dispatch_next_4" : "__kmpc_dispatch_next_4u")
          : (IVSigned ? "__kmpc_dispatch_next_8" : "__kmpc_dispatch_next_8u");
  llvm::Type *ITy = IVSize == 32 ? CGM.Int32Ty : CGM.Int64Ty;
  auto *PtrTy = llvm::PointerType::getUnqual(ITy);
  llvm::Type *TypeParams[] = {
    getIdentTyPointerTy(),                     // loc
    CGM.Int32Ty,                               // tid
    llvm::PointerType::getUnqual(CGM.Int32Ty), // p_lastiter
    PtrTy,                                     // p_lower
    PtrTy,                                     // p_upper
    PtrTy                                      // p_stride
  };
  auto *FnTy =
      llvm::FunctionType::get(CGM.Int32Ty, TypeParams, /*isVarArg*/ false);
  return CGM.CreateRuntimeFunction(FnTy, Name);
}
 
/// Obtain information that uniquely identifies a target entry. This
/// consists of the file and device IDs as well as line number associated with
/// the relevant entry source location.
static llvm::TargetRegionEntryInfo
getTargetEntryUniqueInfo(ASTContext &C, SourceLocation Loc,
                         StringRef ParentName = "") {
  SourceManager &SM = C.getSourceManager();
 
  // The loc should be always valid and have a file ID (the user cannot use
  // #pragma directives in macros)
 
  assert(Loc.isValid() && "Source location is expected to be always valid.");
 
  PresumedLoc PLoc = SM.getPresumedLoc(Loc);
  assert(PLoc.isValid() && "Source location is expected to be always valid.");
 
  llvm::sys::fs::UniqueID ID;
  if (auto EC = llvm::sys::fs::getUniqueID(PLoc.getFilename(), ID)) {
    PLoc = SM.getPresumedLoc(Loc, /*UseLineDirectives=*/false);
    assert(PLoc.isValid() && "Source location is expected to be always valid.");
    if (auto EC = llvm::sys::fs::getUniqueID(PLoc.getFilename(), ID))
      SM.getDiagnostics().Report(diag::err_cannot_open_file)
          << PLoc.getFilename() << EC.message();
  }
 
  return llvm::TargetRegionEntryInfo(ParentName, ID.getDevice(), ID.getFile(),
                                     PLoc.getLine());
}
 
Address CGOpenMPRuntime::getAddrOfDeclareTargetVar(const VarDecl *VD) {
  if (CGM.getLangOpts().OpenMPSimd)
    return Address::invalid();
  std::optional<OMPDeclareTargetDeclAttr::MapTypeTy> Res =
      OMPDeclareTargetDeclAttr::isDeclareTargetDeclaration(VD);
  if (Res && (*Res == OMPDeclareTargetDeclAttr::MT_Link ||
              ((*Res == OMPDeclareTargetDeclAttr::MT_To ||
                *Res == OMPDeclareTargetDeclAttr::MT_Enter) &&
               HasRequiresUnifiedSharedMemory))) {
    SmallString<64> PtrName;
    {
      llvm::raw_svector_ostream OS(PtrName);
      OS << CGM.getMangledName(GlobalDecl(VD));
      if (!VD->isExternallyVisible()) {
        auto EntryInfo = getTargetEntryUniqueInfo(
            CGM.getContext(), VD->getCanonicalDecl()->getBeginLoc());
        OS << llvm::format("_%x", EntryInfo.FileID);
      }
      OS << "_decl_tgt_ref_ptr";
    }
    llvm::Value *Ptr = CGM.getModule().getNamedValue(PtrName);
    QualType PtrTy = CGM.getContext().getPointerType(VD->getType());
    llvm::Type *LlvmPtrTy = CGM.getTypes().ConvertTypeForMem(PtrTy);
    if (!Ptr) {
      Ptr = OMPBuilder.getOrCreateInternalVariable(LlvmPtrTy, PtrName);
 
      auto *GV = cast<llvm::GlobalVariable>(Ptr);
      GV->setLinkage(llvm::GlobalValue::WeakAnyLinkage);
 
      if (!CGM.getLangOpts().OpenMPIsDevice)
        GV->setInitializer(CGM.GetAddrOfGlobal(VD));
      registerTargetGlobalVariable(VD, cast<llvm::Constant>(Ptr));
    }
    return Address(Ptr, LlvmPtrTy, CGM.getContext().getDeclAlign(VD));
  }
  return Address::invalid();
}
 
llvm::Constant *
CGOpenMPRuntime::getOrCreateThreadPrivateCache(const VarDecl *VD) {
  assert(!CGM.getLangOpts().OpenMPUseTLS ||
         !CGM.getContext().getTargetInfo().isTLSSupported());
  // Lookup the entry, lazily creating it if necessary.
  std::string Suffix = getName({"cache", ""});
  return OMPBuilder.getOrCreateInternalVariable(
      CGM.Int8PtrPtrTy, Twine(CGM.getMangledName(VD)).concat(Suffix).str());
}
 
Address CGOpenMPRuntime::getAddrOfThreadPrivate(CodeGenFunction &CGF,
                                                const VarDecl *VD,
                                                Address VDAddr,
                                                SourceLocation Loc) {
  if (CGM.getLangOpts().OpenMPUseTLS &&
      CGM.getContext().getTargetInfo().isTLSSupported())
    return VDAddr;
 
  llvm::Type *VarTy = VDAddr.getElementType();
  llvm::Value *Args[] = {
      emitUpdateLocation(CGF, Loc), getThreadID(CGF, Loc),
      CGF.Builder.CreatePointerCast(VDAddr.getPointer(), CGM.Int8PtrTy),
      CGM.getSize(CGM.GetTargetTypeStoreSize(VarTy)),
      getOrCreateThreadPrivateCache(VD)};
  return Address(
      CGF.EmitRuntimeCall(
          OMPBuilder.getOrCreateRuntimeFunction(
              CGM.getModule(), OMPRTL___kmpc_threadprivate_cached),
          Args),
      CGF.Int8Ty, VDAddr.getAlignment());
}
 
void CGOpenMPRuntime::emitThreadPrivateVarInit(
    CodeGenFunction &CGF, Address VDAddr, llvm::Value *Ctor,
    llvm::Value *CopyCtor, llvm::Value *Dtor, SourceLocation Loc) {
  // Call kmp_int32 __kmpc_global_thread_num(&loc) to init OpenMP runtime
  // library.
  llvm::Value *OMPLoc = emitUpdateLocation(CGF, Loc);
  CGF.EmitRuntimeCall(OMPBuilder.getOrCreateRuntimeFunction(
                          CGM.getModule(), OMPRTL___kmpc_global_thread_num),
                      OMPLoc);
  // Call __kmpc_threadprivate_register(&loc, &var, ctor, cctor/*NULL*/, dtor)
  // to register constructor/destructor for variable.
  llvm::Value *Args[] = {
      OMPLoc, CGF.Builder.CreatePointerCast(VDAddr.getPointer(), CGM.VoidPtrTy),
      Ctor, CopyCtor, Dtor};
  CGF.EmitRuntimeCall(
      OMPBuilder.getOrCreateRuntimeFunction(
          CGM.getModule(), OMPRTL___kmpc_threadprivate_register),
      Args);
}
 
llvm::Function *CGOpenMPRuntime::emitThreadPrivateVarDefinition(
    const VarDecl *VD, Address VDAddr, SourceLocation Loc,
    bool PerformInit, CodeGenFunction *CGF) {
  if (CGM.getLangOpts().OpenMPUseTLS &&
      CGM.getContext().getTargetInfo().isTLSSupported())
    return nullptr;
 
  VD = VD->getDefinition(CGM.getContext());
  if (VD && ThreadPrivateWithDefinition.insert(CGM.getMangledName(VD)).second) {
    QualType ASTTy = VD->getType();
 
    llvm::Value *Ctor = nullptr, *CopyCtor = nullptr, *Dtor = nullptr;
    const Expr *Init = VD->getAnyInitializer();
    if (CGM.getLangOpts().CPlusPlus && PerformInit) {
      // Generate function that re-emits the declaration's initializer into the
      // threadprivate copy of the variable VD
      CodeGenFunction CtorCGF(CGM);
      FunctionArgList Args;
      ImplicitParamDecl Dst(CGM.getContext(), /*DC=*/nullptr, Loc,
                            /*Id=*/nullptr, CGM.getContext().VoidPtrTy,
                            ImplicitParamDecl::Other);
      Args.push_back(&Dst);
 
      const auto &FI = CGM.getTypes().arrangeBuiltinFunctionDeclaration(
          CGM.getContext().VoidPtrTy, Args);
      llvm::FunctionType *FTy = CGM.getTypes().GetFunctionType(FI);
      std::string Name = getName({"__kmpc_global_ctor_", ""});
      llvm::Function *Fn =
          CGM.CreateGlobalInitOrCleanUpFunction(FTy, Name, FI, Loc);
      CtorCGF.StartFunction(GlobalDecl(), CGM.getContext().VoidPtrTy, Fn, FI,
                            Args, Loc, Loc);
      llvm::Value *ArgVal = CtorCGF.EmitLoadOfScalar(
          CtorCGF.GetAddrOfLocalVar(&Dst), /*Volatile=*/false,
          CGM.getContext().VoidPtrTy, Dst.getLocation());
      Address Arg(ArgVal, CtorCGF.Int8Ty, VDAddr.getAlignment());
      Arg = CtorCGF.Builder.CreateElementBitCast(
          Arg, CtorCGF.ConvertTypeForMem(ASTTy));
      CtorCGF.EmitAnyExprToMem(Init, Arg, Init->getType().getQualifiers(),
                               /*IsInitializer=*/true);
      ArgVal = CtorCGF.EmitLoadOfScalar(
          CtorCGF.GetAddrOfLocalVar(&Dst), /*Volatile=*/false,
          CGM.getContext().VoidPtrTy, Dst.getLocation());
      CtorCGF.Builder.CreateStore(ArgVal, CtorCGF.ReturnValue);
      CtorCGF.FinishFunction();
      Ctor = Fn;
    }
    if (VD->getType().isDestructedType() != QualType::DK_none) {
      // Generate function that emits destructor call for the threadprivate copy
      // of the variable VD
      CodeGenFunction DtorCGF(CGM);
      FunctionArgList Args;
      ImplicitParamDecl Dst(CGM.getContext(), /*DC=*/nullptr, Loc,
                            /*Id=*/nullptr, CGM.getContext().VoidPtrTy,
                            ImplicitParamDecl::Other);
      Args.push_back(&Dst);
 
      const auto &FI = CGM.getTypes().arrangeBuiltinFunctionDeclaration(
          CGM.getContext().VoidTy, Args);
      llvm::FunctionType *FTy = CGM.getTypes().GetFunctionType(FI);
      std::string Name = getName({"__kmpc_global_dtor_", ""});
      llvm::Function *Fn =
          CGM.CreateGlobalInitOrCleanUpFunction(FTy, Name, FI, Loc);
      auto NL = ApplyDebugLocation::CreateEmpty(DtorCGF);
      DtorCGF.StartFunction(GlobalDecl(), CGM.getContext().VoidTy, Fn, FI, Args,
                            Loc, Loc);
      // Create a scope with an artificial location for the body of this function.
      auto AL = ApplyDebugLocation::CreateArtificial(DtorCGF);
      llvm::Value *ArgVal = DtorCGF.EmitLoadOfScalar(
          DtorCGF.GetAddrOfLocalVar(&Dst),
          /*Volatile=*/false, CGM.getContext().VoidPtrTy, Dst.getLocation());
      DtorCGF.emitDestroy(
          Address(ArgVal, DtorCGF.Int8Ty, VDAddr.getAlignment()), ASTTy,
          DtorCGF.getDestroyer(ASTTy.isDestructedType()),
          DtorCGF.needsEHCleanup(ASTTy.isDestructedType()));
      DtorCGF.FinishFunction();
      Dtor = Fn;
    }
    // Do not emit init function if it is not required.
    if (!Ctor && !Dtor)
      return nullptr;
 
    llvm::Type *CopyCtorTyArgs[] = {CGM.VoidPtrTy, CGM.VoidPtrTy};
    auto *CopyCtorTy = llvm::FunctionType::get(CGM.VoidPtrTy, CopyCtorTyArgs,
                                               /*isVarArg=*/false)
                           ->getPointerTo();
    // Copying constructor for the threadprivate variable.
    // Must be NULL - reserved by runtime, but currently it requires that this
    // parameter is always NULL. Otherwise it fires assertion.
    CopyCtor = llvm::Constant::getNullValue(CopyCtorTy);
    if (Ctor == nullptr) {
      auto *CtorTy = llvm::FunctionType::get(CGM.VoidPtrTy, CGM.VoidPtrTy,
                                             /*isVarArg=*/false)
                         ->getPointerTo();
      Ctor = llvm::Constant::getNullValue(CtorTy);
    }
    if (Dtor == nullptr) {
      auto *DtorTy = llvm::FunctionType::get(CGM.VoidTy, CGM.VoidPtrTy,
                                             /*isVarArg=*/false)
                         ->getPointerTo();
      Dtor = llvm::Constant::getNullValue(DtorTy);
    }
    if (!CGF) {
      auto *InitFunctionTy =
          llvm::FunctionType::get(CGM.VoidTy, /*isVarArg*/ false);
      std::string Name = getName({"__omp_threadprivate_init_", ""});
      llvm::Function *InitFunction = CGM.CreateGlobalInitOrCleanUpFunction(
          InitFunctionTy, Name, CGM.getTypes().arrangeNullaryFunction());
      CodeGenFunction InitCGF(CGM);
      FunctionArgList ArgList;
      InitCGF.StartFunction(GlobalDecl(), CGM.getContext().VoidTy, InitFunction,
                            CGM.getTypes().arrangeNullaryFunction(), ArgList,
                            Loc, Loc);
      emitThreadPrivateVarInit(InitCGF, VDAddr, Ctor, CopyCtor, Dtor, Loc);
      InitCGF.FinishFunction();
      return InitFunction;
    }
    emitThreadPrivateVarInit(*CGF, VDAddr, Ctor, CopyCtor, Dtor, Loc);
  }
  return nullptr;
}
 
bool CGOpenMPRuntime::emitDeclareTargetVarDefinition(const VarDecl *VD,
                                                     llvm::GlobalVariable *Addr,
                                                     bool PerformInit) {
  if (CGM.getLangOpts().OMPTargetTriples.empty() &&
      !CGM.getLangOpts().OpenMPIsDevice)
    return false;
  std::optional<OMPDeclareTargetDeclAttr::MapTypeTy> Res =
      OMPDeclareTargetDeclAttr::isDeclareTargetDeclaration(VD);
  if (!Res || *Res == OMPDeclareTargetDeclAttr::MT_Link ||
      ((*Res == OMPDeclareTargetDeclAttr::MT_To ||
        *Res == OMPDeclareTargetDeclAttr::MT_Enter) &&
       HasRequiresUnifiedSharedMemory))
    return CGM.getLangOpts().OpenMPIsDevice;
  VD = VD->getDefinition(CGM.getContext());
  assert(VD && "Unknown VarDecl");
 
  if (!DeclareTargetWithDefinition.insert(CGM.getMangledName(VD)).second)
    return CGM.getLangOpts().OpenMPIsDevice;
 
  QualType ASTTy = VD->getType();
  SourceLocation Loc = VD->getCanonicalDecl()->getBeginLoc();
 
  // Produce the unique prefix to identify the new target regions. We use
  // the source location of the variable declaration which we know to not
  // conflict with any target region.
  auto EntryInfo =
      getTargetEntryUniqueInfo(CGM.getContext(), Loc, VD->getName());
  SmallString<128> Buffer, Out;
  OMPBuilder.OffloadInfoManager.getTargetRegionEntryFnName(Buffer, EntryInfo);
 
  const Expr *Init = VD->getAnyInitializer();
  if (CGM.getLangOpts().CPlusPlus && PerformInit) {
    llvm::Constant *Ctor;
    llvm::Constant *ID;
    if (CGM.getLangOpts().OpenMPIsDevice) {
      // Generate function that re-emits the declaration's initializer into
      // the threadprivate copy of the variable VD
      CodeGenFunction CtorCGF(CGM);
 
      const CGFunctionInfo &FI = CGM.getTypes().arrangeNullaryFunction();
      llvm::FunctionType *FTy = CGM.getTypes().GetFunctionType(FI);
      llvm::Function *Fn = CGM.CreateGlobalInitOrCleanUpFunction(
          FTy, Twine(Buffer, "_ctor"), FI, Loc, false,
          llvm::GlobalValue::WeakODRLinkage);
      Fn->setVisibility(llvm::GlobalValue::ProtectedVisibility);
      if (CGM.getTriple().isAMDGCN())
        Fn->setCallingConv(llvm::CallingConv::AMDGPU_KERNEL);
      auto NL = ApplyDebugLocation::CreateEmpty(CtorCGF);
      CtorCGF.StartFunction(GlobalDecl(), CGM.getContext().VoidTy, Fn, FI,
                            FunctionArgList(), Loc, Loc);
      auto AL = ApplyDebugLocation::CreateArtificial(CtorCGF);
      llvm::Constant *AddrInAS0 = Addr;
      if (Addr->getAddressSpace() != 0)
        AddrInAS0 = llvm::ConstantExpr::getAddrSpaceCast(
            Addr, llvm::PointerType::getWithSamePointeeType(
                      cast<llvm::PointerType>(Addr->getType()), 0));
      CtorCGF.EmitAnyExprToMem(Init,
                               Address(AddrInAS0, Addr->getValueType(),
                                       CGM.getContext().getDeclAlign(VD)),
                               Init->getType().getQualifiers(),
                               /*IsInitializer=*/true);
      CtorCGF.FinishFunction();
      Ctor = Fn;
      ID = llvm::ConstantExpr::getBitCast(Fn, CGM.Int8PtrTy);
    } else {
      Ctor = new llvm::GlobalVariable(
          CGM.getModule(), CGM.Int8Ty, /*isConstant=*/true,
          llvm::GlobalValue::PrivateLinkage,
          llvm::Constant::getNullValue(CGM.Int8Ty), Twine(Buffer, "_ctor"));
      ID = Ctor;
    }
 
    // Register the information for the entry associated with the constructor.
    Out.clear();
    auto CtorEntryInfo = EntryInfo;
    CtorEntryInfo.ParentName = Twine(Buffer, "_ctor").toStringRef(Out);
    OMPBuilder.OffloadInfoManager.registerTargetRegionEntryInfo(
        CtorEntryInfo, Ctor, ID,
        llvm::OffloadEntriesInfoManager::OMPTargetRegionEntryCtor);
  }
  if (VD->getType().isDestructedType() != QualType::DK_none) {
    llvm::Constant *Dtor;
    llvm::Constant *ID;
    if (CGM.getLangOpts().OpenMPIsDevice) {
      // Generate function that emits destructor call for the threadprivate
      // copy of the variable VD
      CodeGenFunction DtorCGF(CGM);
 
      const CGFunctionInfo &FI = CGM.getTypes().arrangeNullaryFunction();
      llvm::FunctionType *FTy = CGM.getTypes().GetFunctionType(FI);
      llvm::Function *Fn = CGM.CreateGlobalInitOrCleanUpFunction(
          FTy, Twine(Buffer, "_dtor"), FI, Loc, false,
          llvm::GlobalValue::WeakODRLinkage);
      Fn->setVisibility(llvm::GlobalValue::ProtectedVisibility);
      if (CGM.getTriple().isAMDGCN())
        Fn->setCallingConv(llvm::CallingConv::AMDGPU_KERNEL);
      auto NL = ApplyDebugLocation::CreateEmpty(DtorCGF);
      DtorCGF.StartFunction(GlobalDecl(), CGM.getContext().VoidTy, Fn, FI,
                            FunctionArgList(), Loc, Loc);
      // Create a scope with an artificial location for the body of this
      // function.
      auto AL = ApplyDebugLocation::CreateArtificial(DtorCGF);
      llvm::Constant *AddrInAS0 = Addr;
      if (Addr->getAddressSpace() != 0)
        AddrInAS0 = llvm::ConstantExpr::getAddrSpaceCast(
            Addr, llvm::PointerType::getWithSamePointeeType(
                      cast<llvm::PointerType>(Addr->getType()), 0));
      DtorCGF.emitDestroy(Address(AddrInAS0, Addr->getValueType(),
                                  CGM.getContext().getDeclAlign(VD)),
                          ASTTy, DtorCGF.getDestroyer(ASTTy.isDestructedType()),
                          DtorCGF.needsEHCleanup(ASTTy.isDestructedType()));
      DtorCGF.FinishFunction();
      Dtor = Fn;
      ID = llvm::ConstantExpr::getBitCast(Fn, CGM.Int8PtrTy);
    } else {
      Dtor = new llvm::GlobalVariable(
          CGM.getModule(), CGM.Int8Ty, /*isConstant=*/true,
          llvm::GlobalValue::PrivateLinkage,
          llvm::Constant::getNullValue(CGM.Int8Ty), Twine(Buffer, "_dtor"));
      ID = Dtor;
    }
    // Register the information for the entry associated with the destructor.
    Out.clear();
    auto DtorEntryInfo = EntryInfo;
    DtorEntryInfo.ParentName = Twine(Buffer, "_dtor").toStringRef(Out);
    OMPBuilder.OffloadInfoManager.registerTargetRegionEntryInfo(
        DtorEntryInfo, Dtor, ID,
        llvm::OffloadEntriesInfoManager::OMPTargetRegionEntryDtor);
  }
  return CGM.getLangOpts().OpenMPIsDevice;
}
 
Address CGOpenMPRuntime::getAddrOfArtificialThreadPrivate(CodeGenFunction &CGF,
                                                          QualType VarType,
                                                          StringRef Name) {
  std::string Suffix = getName({"artificial", ""});
  llvm::Type *VarLVType = CGF.ConvertTypeForMem(VarType);
  llvm::GlobalVariable *GAddr = OMPBuilder.getOrCreateInternalVariable(
      VarLVType, Twine(Name).concat(Suffix).str());
  if (CGM.getLangOpts().OpenMP && CGM.getLangOpts().OpenMPUseTLS &&
      CGM.getTarget().isTLSSupported()) {
    GAddr->setThreadLocal(/*Val=*/true);
    return Address(GAddr, GAddr->getValueType(),
                   CGM.getContext().getTypeAlignInChars(VarType));
  }
  std::string CacheSuffix = getName({"cache", ""});
  llvm::Value *Args[] = {
      emitUpdateLocation(CGF, SourceLocation()),
      getThreadID(CGF, SourceLocation()),
      CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(GAddr, CGM.VoidPtrTy),
      CGF.Builder.CreateIntCast(CGF.getTypeSize(VarType), CGM.SizeTy,
                                /*isSigned=*/false),
      OMPBuilder.getOrCreateInternalVariable(
          CGM.VoidPtrPtrTy,
          Twine(Name).concat(Suffix).concat(CacheSuffix).str())};
  return Address(
      CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
          CGF.EmitRuntimeCall(
              OMPBuilder.getOrCreateRuntimeFunction(
                  CGM.getModule(), OMPRTL___kmpc_threadprivate_cached),
              Args),
          VarLVType->getPointerTo(/*AddrSpace=*/0)),
      VarLVType, CGM.getContext().getTypeAlignInChars(VarType));
}
 
void CGOpenMPRuntime::emitIfClause(CodeGenFunction &CGF, const Expr *Cond,
                                   const RegionCodeGenTy &ThenGen,
                                   const RegionCodeGenTy &ElseGen) {
  CodeGenFunction::LexicalScope ConditionScope(CGF, Cond->getSourceRange());
 
  // If the condition constant folds and can be elided, try to avoid emitting
  // the condition and the dead arm of the if/else.
  bool CondConstant;
  if (CGF.ConstantFoldsToSimpleInteger(Cond, CondConstant)) {
    if (CondConstant)
      ThenGen(CGF);
    else
      ElseGen(CGF);
    return;
  }
 
  // Otherwise, the condition did not fold, or we couldn't elide it.  Just
  // emit the conditional branch.
  llvm::BasicBlock *ThenBlock = CGF.createBasicBlock("omp_if.then");
  llvm::BasicBlock *ElseBlock = CGF.createBasicBlock("omp_if.else");
  llvm::BasicBlock *ContBlock = CGF.createBasicBlock("omp_if.end");
  CGF.EmitBranchOnBoolExpr(Cond, ThenBlock, ElseBlock, /*TrueCount=*/0);
 
  // Emit the 'then' code.
  CGF.EmitBlock(ThenBlock);
  ThenGen(CGF);
  CGF.EmitBranch(ContBlock);
  // Emit the 'else' code if present.
  // There is no need to emit line number for unconditional branch.
  (void)ApplyDebugLocation::CreateEmpty(CGF);
  CGF.EmitBlock(ElseBlock);
  ElseGen(CGF);
  // There is no need to emit line number for unconditional branch.
  (void)ApplyDebugLocation::CreateEmpty(CGF);
  CGF.EmitBranch(ContBlock);
  // Emit the continuation block for code after the if.
  CGF.EmitBlock(ContBlock, /*IsFinished=*/true);
}
 
void CGOpenMPRuntime::emitParallelCall(CodeGenFunction &CGF, SourceLocation Loc,
                                       llvm::Function *OutlinedFn,
                                       ArrayRef<llvm::Value *> CapturedVars,
                                       const Expr *IfCond,
                                       llvm::Value *NumThreads) {
  if (!CGF.HaveInsertPoint())
    return;
  llvm::Value *RTLoc = emitUpdateLocation(CGF, Loc);
  auto &M = CGM.getModule();
  auto &&ThenGen = [&M, OutlinedFn, CapturedVars, RTLoc,
                    this](CodeGenFunction &CGF, PrePostActionTy &) {
    // Build call __kmpc_fork_call(loc, n, microtask, var1, .., varn);
    CGOpenMPRuntime &RT = CGF.CGM.getOpenMPRuntime();
    llvm::Value *Args[] = {
        RTLoc,
        CGF.Builder.getInt32(CapturedVars.size()), // Number of captured vars
        CGF.Builder.CreateBitCast(OutlinedFn, RT.getKmpc_MicroPointerTy())};
    llvm::SmallVector<llvm::Value *, 16> RealArgs;
    RealArgs.append(std::begin(Args), std::end(Args));
    RealArgs.append(CapturedVars.begin(), CapturedVars.end());
 
    llvm::FunctionCallee RTLFn =
        OMPBuilder.getOrCreateRuntimeFunction(M, OMPRTL___kmpc_fork_call);
    CGF.EmitRuntimeCall(RTLFn, RealArgs);
  };
  auto &&ElseGen = [&M, OutlinedFn, CapturedVars, RTLoc, Loc,
                    this](CodeGenFunction &CGF, PrePostActionTy &) {
    CGOpenMPRuntime &RT = CGF.CGM.getOpenMPRuntime();
    llvm::Value *ThreadID = RT.getThreadID(CGF, Loc);
    // Build calls:
    // __kmpc_serialized_parallel(&Loc, GTid);
    llvm::Value *Args[] = {RTLoc, ThreadID};
    CGF.EmitRuntimeCall(OMPBuilder.getOrCreateRuntimeFunction(
                            M, OMPRTL___kmpc_serialized_parallel),
                        Args);
 
    // OutlinedFn(&GTid, &zero_bound, CapturedStruct);
    Address ThreadIDAddr = RT.emitThreadIDAddress(CGF, Loc);
    Address ZeroAddrBound =
        CGF.CreateDefaultAlignTempAlloca(CGF.Int32Ty,
                                         /*Name=*/".bound.zero.addr");
    CGF.Builder.CreateStore(CGF.Builder.getInt32(/*C*/ 0), ZeroAddrBound);
    llvm::SmallVector<llvm::Value *, 16> OutlinedFnArgs;
    // ThreadId for serialized parallels is 0.
    OutlinedFnArgs.push_back(ThreadIDAddr.getPointer());
    OutlinedFnArgs.push_back(ZeroAddrBound.getPointer());
    OutlinedFnArgs.append(CapturedVars.begin(), CapturedVars.end());
 
    // Ensure we do not inline the function. This is trivially true for the ones
    // passed to __kmpc_fork_call but the ones called in serialized regions
    // could be inlined. This is not a perfect but it is closer to the invariant
    // we want, namely, every data environment starts with a new function.
    // TODO: We should pass the if condition to the runtime function and do the
    //       handling there. Much cleaner code.
    OutlinedFn->removeFnAttr(llvm::Attribute::AlwaysInline);
    OutlinedFn->addFnAttr(llvm::Attribute::NoInline);
    RT.emitOutlinedFunctionCall(CGF, Loc, OutlinedFn, OutlinedFnArgs);
 
    // __kmpc_end_serialized_parallel(&Loc, GTid);
    llvm::Value *EndArgs[] = {RT.emitUpdateLocation(CGF, Loc), ThreadID};
    CGF.EmitRuntimeCall(OMPBuilder.getOrCreateRuntimeFunction(
                            M, OMPRTL___kmpc_end_serialized_parallel),
                        EndArgs);
  };
  if (IfCond) {
    emitIfClause(CGF, IfCond, ThenGen, ElseGen);
  } else {
    RegionCodeGenTy ThenRCG(ThenGen);
    ThenRCG(CGF);
  }
}
 
// If we're inside an (outlined) parallel region, use the region info's
// thread-ID variable (it is passed in a first argument of the outlined function
// as "kmp_int32 *gtid"). Otherwise, if we're not inside parallel region, but in
// regular serial code region, get thread ID by calling kmp_int32
// kmpc_global_thread_num(ident_t *loc), stash this thread ID in a temporary and
// return the address of that temp.
Address CGOpenMPRuntime::emitThreadIDAddress(CodeGenFunction &CGF,
                                             SourceLocation Loc) {
  if (auto *OMPRegionInfo =
          dyn_cast_or_null<CGOpenMPRegionInfo>(CGF.CapturedStmtInfo))
    if (OMPRegionInfo->getThreadIDVariable())
      return OMPRegionInfo->getThreadIDVariableLValue(CGF).getAddress(CGF);
 
  llvm::Value *ThreadID = getThreadID(CGF, Loc);
  QualType Int32Ty =
      CGF.getContext().getIntTypeForBitwidth(/*DestWidth*/ 32, /*Signed*/ true);
  Address ThreadIDTemp = CGF.CreateMemTemp(Int32Ty, /*Name*/ ".threadid_temp.");
  CGF.EmitStoreOfScalar(ThreadID,
                        CGF.MakeAddrLValue(ThreadIDTemp, Int32Ty));
 
  return ThreadIDTemp;
}
 
llvm::Value *CGOpenMPRuntime::getCriticalRegionLock(StringRef CriticalName) {
  std::string Prefix = Twine("gomp_critical_user_", CriticalName).str();
  std::string Name = getName({Prefix, "var"});
  return OMPBuilder.getOrCreateInternalVariable(KmpCriticalNameTy, Name);
}
 
namespace {
/// Common pre(post)-action for different OpenMP constructs.
class CommonActionTy final : public PrePostActionTy {
  llvm::FunctionCallee EnterCallee;
  ArrayRef<llvm::Value *> EnterArgs;
  llvm::FunctionCallee ExitCallee;
  ArrayRef<llvm::Value *> ExitArgs;
  bool Conditional;
  llvm::BasicBlock *ContBlock = nullptr;
 
public:
  CommonActionTy(llvm::FunctionCallee EnterCallee,
                 ArrayRef<llvm::Value *> EnterArgs,
                 llvm::FunctionCallee ExitCallee,
                 ArrayRef<llvm::Value *> ExitArgs, bool Conditional = false)
      : EnterCallee(EnterCallee), EnterArgs(EnterArgs), ExitCallee(ExitCallee),
        ExitArgs(ExitArgs), Conditional(Conditional) {}
  void Enter(CodeGenFunction &CGF) override {
    llvm::Value *EnterRes = CGF.EmitRuntimeCall(EnterCallee, EnterArgs);
    if (Conditional) {
      llvm::Value *CallBool = CGF.Builder.CreateIsNotNull(EnterRes);
      auto *ThenBlock = CGF.createBasicBlock("omp_if.then");
      ContBlock = CGF.createBasicBlock("omp_if.end");
      // Generate the branch (If-stmt)
      CGF.Builder.CreateCondBr(CallBool, ThenBlock, ContBlock);
      CGF.EmitBlock(ThenBlock);
    }
  }
  void Done(CodeGenFunction &CGF) {
    // Emit the rest of blocks/branches
    CGF.EmitBranch(ContBlock);
    CGF.EmitBlock(ContBlock, true);
  }
  void Exit(CodeGenFunction &CGF) override {
    CGF.EmitRuntimeCall(ExitCallee, ExitArgs);
  }
};
} // anonymous namespace
 
void CGOpenMPRuntime::emitCriticalRegion(CodeGenFunction &CGF,
                                         StringRef CriticalName,
                                         const RegionCodeGenTy &CriticalOpGen,
                                         SourceLocation Loc, const Expr *Hint) {
  // __kmpc_critical[_with_hint](ident_t *, gtid, Lock[, hint]);
  // CriticalOpGen();
  // __kmpc_end_critical(ident_t *, gtid, Lock);
  // Prepare arguments and build a call to __kmpc_critical
  if (!CGF.HaveInsertPoint())
    return;
  llvm::Value *Args[] = {emitUpdateLocation(CGF, Loc), getThreadID(CGF, Loc),
                         getCriticalRegionLock(CriticalName)};
  llvm::SmallVector<llvm::Value *, 4> EnterArgs(std::begin(Args),
                                                std::end(Args));
  if (Hint) {
    EnterArgs.push_back(CGF.Builder.CreateIntCast(
        CGF.EmitScalarExpr(Hint), CGM.Int32Ty, /*isSigned=*/false));
  }
  CommonActionTy Action(
      OMPBuilder.getOrCreateRuntimeFunction(
          CGM.getModule(),
          Hint ? OMPRTL___kmpc_critical_with_hint : OMPRTL___kmpc_critical),
      EnterArgs,
      OMPBuilder.getOrCreateRuntimeFunction(CGM.getModule(),
                                            OMPRTL___kmpc_end_critical),
      Args);
  CriticalOpGen.setAction(Action);
  emitInlinedDirective(CGF, OMPD_critical, CriticalOpGen);
}
 
void CGOpenMPRuntime::emitMasterRegion(CodeGenFunction &CGF,
                                       const RegionCodeGenTy &MasterOpGen,
                                       SourceLocation Loc) {
  if (!CGF.HaveInsertPoint())
    return;
  // if(__kmpc_master(ident_t *, gtid)) {
  //   MasterOpGen();
  //   __kmpc_end_master(ident_t *, gtid);
  // }
  // Prepare arguments and build a call to __kmpc_master
  llvm::Value *Args[] = {emitUpdateLocation(CGF, Loc), getThreadID(CGF, Loc)};
  CommonActionTy Action(OMPBuilder.getOrCreateRuntimeFunction(
                            CGM.getModule(), OMPRTL___kmpc_master),
                        Args,
                        OMPBuilder.getOrCreateRuntimeFunction(
                            CGM.getModule(), OMPRTL___kmpc_end_master),
                        Args,
                        /*Conditional=*/true);
  MasterOpGen.setAction(Action);
  emitInlinedDirective(CGF, OMPD_master, MasterOpGen);
  Action.Done(CGF);
}
 
void CGOpenMPRuntime::emitMaskedRegion(CodeGenFunction &CGF,
                                       const RegionCodeGenTy &MaskedOpGen,
                                       SourceLocation Loc, const Expr *Filter) {
  if (!CGF.HaveInsertPoint())
    return;
  // if(__kmpc_masked(ident_t *, gtid, filter)) {
  //   MaskedOpGen();
  //   __kmpc_end_masked(iden_t *, gtid);
  // }
  // Prepare arguments and build a call to __kmpc_masked
  llvm::Value *FilterVal = Filter
                               ? CGF.EmitScalarExpr(Filter, CGF.Int32Ty)
                               : llvm::ConstantInt::get(CGM.Int32Ty, /*V=*/0);
  llvm::Value *Args[] = {emitUpdateLocation(CGF, Loc), getThreadID(CGF, Loc),
                         FilterVal};
  llvm::Value *ArgsEnd[] = {emitUpdateLocation(CGF, Loc),
                            getThreadID(CGF, Loc)};
  CommonActionTy Action(OMPBuilder.getOrCreateRuntimeFunction(
                            CGM.getModule(), OMPRTL___kmpc_masked),
                        Args,
                        OMPBuilder.getOrCreateRuntimeFunction(
                            CGM.getModule(), OMPRTL___kmpc_end_masked),
                        ArgsEnd,
                        /*Conditional=*/true);
  MaskedOpGen.setAction(Action);
  emitInlinedDirective(CGF, OMPD_masked, MaskedOpGen);
  Action.Done(CGF);
}
 
void CGOpenMPRuntime::emitTaskyieldCall(CodeGenFunction &CGF,
                                        SourceLocation Loc) {
  if (!CGF.HaveInsertPoint())
    return;
  if (CGF.CGM.getLangOpts().OpenMPIRBuilder) {
    OMPBuilder.createTaskyield(CGF.Builder);
  } else {
    // Build call __kmpc_omp_taskyield(loc, thread_id, 0);
    llvm::Value *Args[] = {
        emitUpdateLocation(CGF, Loc), getThreadID(CGF, Loc),
        llvm::ConstantInt::get(CGM.IntTy, /*V=*/0, /*isSigned=*/true)};
    CGF.EmitRuntimeCall(OMPBuilder.getOrCreateRuntimeFunction(
                            CGM.getModule(), OMPRTL___kmpc_omp_taskyield),
                        Args);
  }
 
  if (auto *Region = dyn_cast_or_null<CGOpenMPRegionInfo>(CGF.CapturedStmtInfo))
    Region->emitUntiedSwitch(CGF);
}
 
void CGOpenMPRuntime::emitTaskgroupRegion(CodeGenFunction &CGF,
                                          const RegionCodeGenTy &TaskgroupOpGen,
                                          SourceLocation Loc) {
  if (!CGF.HaveInsertPoint())
    return;
  // __kmpc_taskgroup(ident_t *, gtid);
  // TaskgroupOpGen();
  // __kmpc_end_taskgroup(ident_t *, gtid);
  // Prepare arguments and build a call to __kmpc_taskgroup
  llvm::Value *Args[] = {emitUpdateLocation(CGF, Loc), getThreadID(CGF, Loc)};
  CommonActionTy Action(OMPBuilder.getOrCreateRuntimeFunction(
                            CGM.getModule(), OMPRTL___kmpc_taskgroup),
                        Args,
                        OMPBuilder.getOrCreateRuntimeFunction(
                            CGM.getModule(), OMPRTL___kmpc_end_taskgroup),
                        Args);
  TaskgroupOpGen.setAction(Action);
  emitInlinedDirective(CGF, OMPD_taskgroup, TaskgroupOpGen);
}
 
/// Given an array of pointers to variables, project the address of a
/// given variable.
static Address emitAddrOfVarFromArray(CodeGenFunction &CGF, Address Array,
                                      unsigned Index, const VarDecl *Var) {
  // Pull out the pointer to the variable.
  Address PtrAddr = CGF.Builder.CreateConstArrayGEP(Array, Index);
  llvm::Value *Ptr = CGF.Builder.CreateLoad(PtrAddr);
 
  llvm::Type *ElemTy = CGF.ConvertTypeForMem(Var->getType());
  return Address(
      CGF.Builder.CreateBitCast(
          Ptr, ElemTy->getPointerTo(Ptr->getType()->getPointerAddressSpace())),
      ElemTy, CGF.getContext().getDeclAlign(Var));
}
 
static llvm::Value *emitCopyprivateCopyFunction(
    CodeGenModule &CGM, llvm::Type *ArgsElemType,
    ArrayRef<const Expr *> CopyprivateVars, ArrayRef<const Expr *> DestExprs,
    ArrayRef<const Expr *> SrcExprs, ArrayRef<const Expr *> AssignmentOps,
    SourceLocation Loc) {
  ASTContext &C = CGM.getContext();
  // void copy_func(void *LHSArg, void *RHSArg);
  FunctionArgList Args;
  ImplicitParamDecl LHSArg(C, /*DC=*/nullptr, Loc, /*Id=*/nullptr, C.VoidPtrTy,
                           ImplicitParamDecl::Other);
  ImplicitParamDecl RHSArg(C, /*DC=*/nullptr, Loc, /*Id=*/nullptr, C.VoidPtrTy,
                           ImplicitParamDecl::Other);
  Args.push_back(&LHSArg);
  Args.push_back(&RHSArg);
  const auto &CGFI =
      CGM.getTypes().arrangeBuiltinFunctionDeclaration(C.VoidTy, Args);
  std::string Name =
      CGM.getOpenMPRuntime().getName({"omp", "copyprivate", "copy_func"});
  auto *Fn = llvm::Function::Create(CGM.getTypes().GetFunctionType(CGFI),
                                    llvm::GlobalValue::InternalLinkage, Name,
                                    &CGM.getModule());
  CGM.SetInternalFunctionAttributes(GlobalDecl(), Fn, CGFI);
  Fn->setDoesNotRecurse();
  CodeGenFunction CGF(CGM);
  CGF.StartFunction(GlobalDecl(), C.VoidTy, Fn, CGFI, Args, Loc, Loc);
  // Dest = (void*[n])(LHSArg);
  // Src = (void*[n])(RHSArg);
  Address LHS(CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
                  CGF.Builder.CreateLoad(CGF.GetAddrOfLocalVar(&LHSArg)),
                  ArgsElemType->getPointerTo()),
              ArgsElemType, CGF.getPointerAlign());
  Address RHS(CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
                  CGF.Builder.CreateLoad(CGF.GetAddrOfLocalVar(&RHSArg)),
                  ArgsElemType->getPointerTo()),
              ArgsElemType, CGF.getPointerAlign());
  // *(Type0*)Dst[0] = *(Type0*)Src[0];
  // *(Type1*)Dst[1] = *(Type1*)Src[1];
  // ...
  // *(Typen*)Dst[n] = *(Typen*)Src[n];
  for (unsigned I = 0, E = AssignmentOps.size(); I < E; ++I) {
    const auto *DestVar =
        cast<VarDecl>(cast<DeclRefExpr>(DestExprs[I])->getDecl());
    Address DestAddr = emitAddrOfVarFromArray(CGF, LHS, I, DestVar);
 
    const auto *SrcVar =
        cast<VarDecl>(cast<DeclRefExpr>(SrcExprs[I])->getDecl());
    Address SrcAddr = emitAddrOfVarFromArray(CGF, RHS, I, SrcVar);
 
    const auto *VD = cast<DeclRefExpr>(CopyprivateVars[I])->getDecl();
    QualType Type = VD->getType();
    CGF.EmitOMPCopy(Type, DestAddr, SrcAddr, DestVar, SrcVar, AssignmentOps[I]);
  }
  CGF.FinishFunction();
  return Fn;
}
 
void CGOpenMPRuntime::emitSingleRegion(CodeGenFunction &CGF,
                                       const RegionCodeGenTy &SingleOpGen,
                                       SourceLocation Loc,
                                       ArrayRef<const Expr *> CopyprivateVars,
                                       ArrayRef<const Expr *> SrcExprs,
                                       ArrayRef<const Expr *> DstExprs,
                                       ArrayRef<const Expr *> AssignmentOps) {
  if (!CGF.HaveInsertPoint())
    return;
  assert(CopyprivateVars.size() == SrcExprs.size() &&
         CopyprivateVars.size() == DstExprs.size() &&
         CopyprivateVars.size() == AssignmentOps.size());
  ASTContext &C = CGM.getContext();
  // int32 did_it = 0;
  // if(__kmpc_single(ident_t *, gtid)) {
  //   SingleOpGen();
  //   __kmpc_end_single(ident_t *, gtid);
  //   did_it = 1;
  // }
  // call __kmpc_copyprivate(ident_t *, gtid, <buf_size>, <copyprivate list>,
  // <copy_func>, did_it);
 
  Address DidIt = Address::invalid();
  if (!CopyprivateVars.empty()) {
    // int32 did_it = 0;
    QualType KmpInt32Ty =
        C.getIntTypeForBitwidth(/*DestWidth=*/32, /*Signed=*/1);
    DidIt = CGF.CreateMemTemp(KmpInt32Ty, ".omp.copyprivate.did_it");
    CGF.Builder.CreateStore(CGF.Builder.getInt32(0), DidIt);
  }
  // Prepare arguments and build a call to __kmpc_single
  llvm::Value *Args[] = {emitUpdateLocation(CGF, Loc), getThreadID(CGF, Loc)};
  CommonActionTy Action(OMPBuilder.getOrCreateRuntimeFunction(
                            CGM.getModule(), OMPRTL___kmpc_single),
                        Args,
                        OMPBuilder.getOrCreateRuntimeFunction(
                            CGM.getModule(), OMPRTL___kmpc_end_single),
                        Args,
                        /*Conditional=*/true);
  SingleOpGen.setAction(Action);
  emitInlinedDirective(CGF, OMPD_single, SingleOpGen);
  if (DidIt.isValid()) {
    // did_it = 1;
    CGF.Builder.CreateStore(CGF.Builder.getInt32(1), DidIt);
  }
  Action.Done(CGF);
  // call __kmpc_copyprivate(ident_t *, gtid, <buf_size>, <copyprivate list>,
  // <copy_func>, did_it);
  if (DidIt.isValid()) {
    llvm::APInt ArraySize(/*unsigned int numBits=*/32, CopyprivateVars.size());
    QualType CopyprivateArrayTy = C.getConstantArrayType(
        C.VoidPtrTy, ArraySize, nullptr, ArrayType::Normal,
        /*IndexTypeQuals=*/0);
    // Create a list of all private variables for copyprivate.
    Address CopyprivateList =
        CGF.CreateMemTemp(CopyprivateArrayTy, ".omp.copyprivate.cpr_list");
    for (unsigned I = 0, E = CopyprivateVars.size(); I < E; ++I) {
      Address Elem = CGF.Builder.CreateConstArrayGEP(CopyprivateList, I);
      CGF.Builder.CreateStore(
          CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
              CGF.EmitLValue(CopyprivateVars[I]).getPointer(CGF),
              CGF.VoidPtrTy),
          Elem);
    }
    // Build function that copies private values from single region to all other
    // threads in the corresponding parallel region.
    llvm::Value *CpyFn = emitCopyprivateCopyFunction(
        CGM, CGF.ConvertTypeForMem(CopyprivateArrayTy), CopyprivateVars,
        SrcExprs, DstExprs, AssignmentOps, Loc);
    llvm::Value *BufSize = CGF.getTypeSize(CopyprivateArrayTy);
    Address CL = CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
        CopyprivateList, CGF.VoidPtrTy, CGF.Int8Ty);
    llvm::Value *DidItVal = CGF.Builder.CreateLoad(DidIt);
    llvm::Value *Args[] = {
        emitUpdateLocation(CGF, Loc), // ident_t *<loc>
        getThreadID(CGF, Loc),        // i32 <gtid>
        BufSize,                      // size_t <buf_size>
        CL.getPointer(),              // void *<copyprivate list>
        CpyFn,                        // void (*) (void *, void *) <copy_func>
        DidItVal                      // i32 did_it
    };
    CGF.EmitRuntimeCall(OMPBuilder.getOrCreateRuntimeFunction(
                            CGM.getModule(), OMPRTL___kmpc_copyprivate),
                        Args);
  }
}
 
void CGOpenMPRuntime::emitOrderedRegion(CodeGenFunction &CGF,
                                        const RegionCodeGenTy &OrderedOpGen,
                                        SourceLocation Loc, bool IsThreads) {
  if (!CGF.HaveInsertPoint())
    return;
  // __kmpc_ordered(ident_t *, gtid);
  // OrderedOpGen();
  // __kmpc_end_ordered(ident_t *, gtid);
  // Prepare arguments and build a call to __kmpc_ordered
  if (IsThreads) {
    llvm::Value *Args[] = {emitUpdateLocation(CGF, Loc), getThreadID(CGF, Loc)};
    CommonActionTy Action(OMPBuilder.getOrCreateRuntimeFunction(
                              CGM.getModule(), OMPRTL___kmpc_ordered),
                          Args,
                          OMPBuilder.getOrCreateRuntimeFunction(
                              CGM.getModule(), OMPRTL___kmpc_end_ordered),
                          Args);
    OrderedOpGen.setAction(Action);
    emitInlinedDirective(CGF, OMPD_ordered, OrderedOpGen);
    return;
  }
  emitInlinedDirective(CGF, OMPD_ordered, OrderedOpGen);
}
 
unsigned CGOpenMPRuntime::getDefaultFlagsForBarriers(OpenMPDirectiveKind Kind) {
  unsigned Flags;
  if (Kind == OMPD_for)
    Flags = OMP_IDENT_BARRIER_IMPL_FOR;
  else if (Kind == OMPD_sections)
    Flags = OMP_IDENT_BARRIER_IMPL_SECTIONS;
  else if (Kind == OMPD_single)
    Flags = OMP_IDENT_BARRIER_IMPL_SINGLE;
  else if (Kind == OMPD_barrier)
    Flags = OMP_IDENT_BARRIER_EXPL;
  else
    Flags = OMP_IDENT_BARRIER_IMPL;
  return Flags;
}
 
void CGOpenMPRuntime::getDefaultScheduleAndChunk(
    CodeGenFunction &CGF, const OMPLoopDirective &S,
    OpenMPScheduleClauseKind &ScheduleKind, const Expr *&ChunkExpr) const {
  // Check if the loop directive is actually a doacross loop directive. In this
  // case choose static, 1 schedule.
  if (llvm::any_of(
          S.getClausesOfKind<OMPOrderedClause>(),
          [](const OMPOrderedClause *C) { return C->getNumForLoops(); })) {
    ScheduleKind = OMPC_SCHEDULE_static;
    // Chunk size is 1 in this case.
    llvm::APInt ChunkSize(32, 1);
    ChunkExpr = IntegerLiteral::Create(
        CGF.getContext(), ChunkSize,
        CGF.getContext().getIntTypeForBitwidth(32, /*Signed=*/0),
        SourceLocation());
  }
}
 
void CGOpenMPRuntime::emitBarrierCall(CodeGenFunction &CGF, SourceLocation Loc,
                                      OpenMPDirectiveKind Kind, bool EmitChecks,
                                      bool ForceSimpleCall) {
  // Check if we should use the OMPBuilder
  auto *OMPRegionInfo =
      dyn_cast_or_null<CGOpenMPRegionInfo>(CGF.CapturedStmtInfo);
  if (CGF.CGM.getLangOpts().OpenMPIRBuilder) {
    CGF.Builder.restoreIP(OMPBuilder.createBarrier(
        CGF.Builder, Kind, ForceSimpleCall, EmitChecks));
    return;
  }
 
  if (!CGF.HaveInsertPoint())
    return;
  // Build call __kmpc_cancel_barrier(loc, thread_id);
  // Build call __kmpc_barrier(loc, thread_id);
  unsigned Flags = getDefaultFlagsForBarriers(Kind);
  // Build call __kmpc_cancel_barrier(loc, thread_id) or __kmpc_barrier(loc,
  // thread_id);
  llvm::Value *Args[] = {emitUpdateLocation(CGF, Loc, Flags),
                         getThreadID(CGF, Loc)};
  if (OMPRegionInfo) {
    if (!ForceSimpleCall && OMPRegionInfo->hasCancel()) {
      llvm::Value *Result = CGF.EmitRuntimeCall(
          OMPBuilder.getOrCreateRuntimeFunction(CGM.getModule(),
                                                OMPRTL___kmpc_cancel_barrier),
          Args);
      if (EmitChecks) {
        // if (__kmpc_cancel_barrier()) {
        //   exit from construct;
        // }
        llvm::BasicBlock *ExitBB = CGF.createBasicBlock(".cancel.exit");
        llvm::BasicBlock *ContBB = CGF.createBasicBlock(".cancel.continue");
        llvm::Value *Cmp = CGF.Builder.CreateIsNotNull(Result);
        CGF.Builder.CreateCondBr(Cmp, ExitBB, ContBB);
        CGF.EmitBlock(ExitBB);
        //   exit from construct;
        CodeGenFunction::JumpDest CancelDestination =
            CGF.getOMPCancelDestination(OMPRegionInfo->getDirectiveKind());
        CGF.EmitBranchThroughCleanup(CancelDestination);
        CGF.EmitBlock(ContBB, /*IsFinished=*/true);
      }
      return;
    }
  }
  CGF.EmitRuntimeCall(OMPBuilder.getOrCreateRuntimeFunction(
                          CGM.getModule(), OMPRTL___kmpc_barrier),
                      Args);
}
 
void CGOpenMPRuntime::emitErrorCall(CodeGenFunction &CGF, SourceLocation Loc,
                                    Expr *ME, bool IsFatal) {
  llvm::Value *MVL =
      ME ? CGF.EmitStringLiteralLValue(cast<StringLiteral>(ME)).getPointer(CGF)
         : llvm::ConstantPointerNull::get(CGF.VoidPtrTy);
  // Build call void __kmpc_error(ident_t *loc, int severity, const char
  // *message)
  llvm::Value *Args[] = {
      emitUpdateLocation(CGF, Loc, /*Flags=*/0, /*GenLoc=*/true),
      llvm::ConstantInt::get(CGM.Int32Ty, IsFatal ? 2 : 1),
      CGF.Builder.CreatePointerCast(MVL, CGM.Int8PtrTy)};
  CGF.EmitRuntimeCall(OMPBuilder.getOrCreateRuntimeFunction(
                          CGM.getModule(), OMPRTL___kmpc_error),
                      Args);
}
 
/// Map the OpenMP loop schedule to the runtime enumeration.
static OpenMPSchedType getRuntimeSchedule(OpenMPScheduleClauseKind ScheduleKind,
                                          bool Chunked, bool Ordered) {
  switch (ScheduleKind) {
  case OMPC_SCHEDULE_static:
    return Chunked ? (Ordered ? OMP_ord_static_chunked : OMP_sch_static_chunked)
                   : (Ordered ? OMP_ord_static : OMP_sch_static);
  case OMPC_SCHEDULE_dynamic:
    return Ordered ? OMP_ord_dynamic_chunked : OMP_sch_dynamic_chunked;
  case OMPC_SCHEDULE_guided:
    return Ordered ? OMP_ord_guided_chunked : OMP_sch_guided_chunked;
  case OMPC_SCHEDULE_runtime:
    return Ordered ? OMP_ord_runtime : OMP_sch_runtime;
  case OMPC_SCHEDULE_auto:
    return Ordered ? OMP_ord_auto : OMP_sch_auto;
  case OMPC_SCHEDULE_unknown:
    assert(!Chunked && "chunk was specified but schedule kind not known");
    return Ordered ? OMP_ord_static : OMP_sch_static;
  }
  llvm_unreachable("Unexpected runtime schedule");
}
 
/// Map the OpenMP distribute schedule to the runtime enumeration.
static OpenMPSchedType
getRuntimeSchedule(OpenMPDistScheduleClauseKind ScheduleKind, bool Chunked) {
  // only static is allowed for dist_schedule
  return Chunked ? OMP_dist_sch_static_chunked : OMP_dist_sch_static;
}
 
bool CGOpenMPRuntime::isStaticNonchunked(OpenMPScheduleClauseKind ScheduleKind,
                                         bool Chunked) const {
  OpenMPSchedType Schedule =
      getRuntimeSchedule(ScheduleKind, Chunked, /*Ordered=*/false);
  return Schedule == OMP_sch_static;
}
 
bool CGOpenMPRuntime::isStaticNonchunked(
    OpenMPDistScheduleClauseKind ScheduleKind, bool Chunked) const {
  OpenMPSchedType Schedule = getRuntimeSchedule(ScheduleKind, Chunked);
  return Schedule == OMP_dist_sch_static;
}
 
bool CGOpenMPRuntime::isStaticChunked(OpenMPScheduleClauseKind ScheduleKind,
                                      bool Chunked) const {
  OpenMPSchedType Schedule =
      getRuntimeSchedule(ScheduleKind, Chunked, /*Ordered=*/false);
  return Schedule == OMP_sch_static_chunked;
}
 
bool CGOpenMPRuntime::isStaticChunked(
    OpenMPDistScheduleClauseKind ScheduleKind, bool Chunked) const {
  OpenMPSchedType Schedule = getRuntimeSchedule(ScheduleKind, Chunked);
  return Schedule == OMP_dist_sch_static_chunked;
}
 
bool CGOpenMPRuntime::isDynamic(OpenMPScheduleClauseKind ScheduleKind) const {
  OpenMPSchedType Schedule =
      getRuntimeSchedule(ScheduleKind, /*Chunked=*/false, /*Ordered=*/false);
  assert(Schedule != OMP_sch_static_chunked && "cannot be chunked here");
  return Schedule != OMP_sch_static;
}
 
static int addMonoNonMonoModifier(CodeGenModule &CGM, OpenMPSchedType Schedule,
                                  OpenMPScheduleClauseModifier M1,
                                  OpenMPScheduleClauseModifier M2) {
  int Modifier = 0;
  switch (M1) {
  case OMPC_SCHEDULE_MODIFIER_monotonic:
    Modifier = OMP_sch_modifier_monotonic;
    break;
  case OMPC_SCHEDULE_MODIFIER_nonmonotonic:
    Modifier = OMP_sch_modifier_nonmonotonic;
    break;
  case OMPC_SCHEDULE_MODIFIER_simd:
    if (Schedule == OMP_sch_static_chunked)
      Schedule = OMP_sch_static_balanced_chunked;
    break;
  case OMPC_SCHEDULE_MODIFIER_last:
  case OMPC_SCHEDULE_MODIFIER_unknown:
    break;
  }
  switch (M2) {
  case OMPC_SCHEDULE_MODIFIER_monotonic:
    Modifier = OMP_sch_modifier_monotonic;
    break;
  case OMPC_SCHEDULE_MODIFIER_nonmonotonic:
    Modifier = OMP_sch_modifier_nonmonotonic;
    break;
  case OMPC_SCHEDULE_MODIFIER_simd:
    if (Schedule == OMP_sch_static_chunked)
      Schedule = OMP_sch_static_balanced_chunked;
    break;
  case OMPC_SCHEDULE_MODIFIER_last:
  case OMPC_SCHEDULE_MODIFIER_unknown:
    break;
  }
  // OpenMP 5.0, 2.9.2 Worksharing-Loop Construct, Desription.
  // If the static schedule kind is specified or if the ordered clause is
  // specified, and if the nonmonotonic modifier is not specified, the effect is
  // as if the monotonic modifier is specified. Otherwise, unless the monotonic
  // modifier is specified, the effect is as if the nonmonotonic modifier is
  // specified.
  if (CGM.getLangOpts().OpenMP >= 50 && Modifier == 0) {
    if (!(Schedule == OMP_sch_static_chunked || Schedule == OMP_sch_static ||
          Schedule == OMP_sch_static_balanced_chunked ||
          Schedule == OMP_ord_static_chunked || Schedule == OMP_ord_static ||
          Schedule == OMP_dist_sch_static_chunked ||
          Schedule == OMP_dist_sch_static))
      Modifier = OMP_sch_modifier_nonmonotonic;
  }
  return Schedule | Modifier;
}
 
void CGOpenMPRuntime::emitForDispatchInit(
    CodeGenFunction &CGF, SourceLocation Loc,
    const OpenMPScheduleTy &ScheduleKind, unsigned IVSize, bool IVSigned,
    bool Ordered, const DispatchRTInput &DispatchValues) {
  if (!CGF.HaveInsertPoint())
    return;
  OpenMPSchedType Schedule = getRuntimeSchedule(
      ScheduleKind.Schedule, DispatchValues.Chunk != nullptr, Ordered);
  assert(Ordered ||
         (Schedule != OMP_sch_static && Schedule != OMP_sch_static_chunked &&
          Schedule != OMP_ord_static && Schedule != OMP_ord_static_chunked &&
          Schedule != OMP_sch_static_balanced_chunked));
  // Call __kmpc_dispatch_init(
  //          ident_t *loc, kmp_int32 tid, kmp_int32 schedule,
  //          kmp_int[32|64] lower, kmp_int[32|64] upper,
  //          kmp_int[32|64] stride, kmp_int[32|64] chunk);
 
  // If the Chunk was not specified in the clause - use default value 1.
  llvm::Value *Chunk = DispatchValues.Chunk ? DispatchValues.Chunk
                                            : CGF.Builder.getIntN(IVSize, 1);
  llvm::Value *Args[] = {
      emitUpdateLocation(CGF, Loc),
      getThreadID(CGF, Loc),
      CGF.Builder.getInt32(addMonoNonMonoModifier(
          CGM, Schedule, ScheduleKind.M1, ScheduleKind.M2)), // Schedule type
      DispatchValues.LB,                                     // Lower
      DispatchValues.UB,                                     // Upper
      CGF.Builder.getIntN(IVSize, 1),                        // Stride
      Chunk                                                  // Chunk
  };
  CGF.EmitRuntimeCall(createDispatchInitFunction(IVSize, IVSigned), Args);
}
 
static void emitForStaticInitCall(
    CodeGenFunction &CGF, llvm::Value *UpdateLocation, llvm::Value *ThreadId,
    llvm::FunctionCallee ForStaticInitFunction, OpenMPSchedType Schedule,
    OpenMPScheduleClauseModifier M1, OpenMPScheduleClauseModifier M2,
    const CGOpenMPRuntime::StaticRTInput &Values) {
  if (!CGF.HaveInsertPoint())
    return;
 
  assert(!Values.Ordered);
  assert(Schedule == OMP_sch_static || Schedule == OMP_sch_static_chunked ||
         Schedule == OMP_sch_static_balanced_chunked ||
         Schedule == OMP_ord_static || Schedule == OMP_ord_static_chunked ||
         Schedule == OMP_dist_sch_static ||
         Schedule == OMP_dist_sch_static_chunked);
 
  // Call __kmpc_for_static_init(
  //          ident_t *loc, kmp_int32 tid, kmp_int32 schedtype,
  //          kmp_int32 *p_lastiter, kmp_int[32|64] *p_lower,
  //          kmp_int[32|64] *p_upper, kmp_int[32|64] *p_stride,
  //          kmp_int[32|64] incr, kmp_int[32|64] chunk);
  llvm::Value *Chunk = Values.Chunk;
  if (Chunk == nullptr) {
    assert((Schedule == OMP_sch_static || Schedule == OMP_ord_static ||
            Schedule == OMP_dist_sch_static) &&
           "expected static non-chunked schedule");
    // If the Chunk was not specified in the clause - use default value 1.
    Chunk = CGF.Builder.getIntN(Values.IVSize, 1);
  } else {
    assert((Schedule == OMP_sch_static_chunked ||
            Schedule == OMP_sch_static_balanced_chunked ||
            Schedule == OMP_ord_static_chunked ||
            Schedule == OMP_dist_sch_static_chunked) &&
           "expected static chunked schedule");
  }
  llvm::Value *Args[] = {
      UpdateLocation,
      ThreadId,
      CGF.Builder.getInt32(addMonoNonMonoModifier(CGF.CGM, Schedule, M1,
                                                  M2)), // Schedule type
      Values.IL.getPointer(),                           // &isLastIter
      Values.LB.getPointer(),                           // &LB
      Values.UB.getPointer(),                           // &UB
      Values.ST.getPointer(),                           // &Stride
      CGF.Builder.getIntN(Values.IVSize, 1),            // Incr
      Chunk                                             // Chunk
  };
  CGF.EmitRuntimeCall(ForStaticInitFunction, Args);
}
 
void CGOpenMPRuntime::emitForStaticInit(CodeGenFunction &CGF,
                                        SourceLocation Loc,
                                        OpenMPDirectiveKind DKind,
                                        const OpenMPScheduleTy &ScheduleKind,
                                        const StaticRTInput &Values) {
  OpenMPSchedType ScheduleNum = getRuntimeSchedule(
      ScheduleKind.Schedule, Values.Chunk != nullptr, Values.Ordered);
  assert(isOpenMPWorksharingDirective(DKind) &&
         "Expected loop-based or sections-based directive.");
  llvm::Value *UpdatedLocation = emitUpdateLocation(CGF, Loc,
                                             isOpenMPLoopDirective(DKind)
                                                 ? OMP_IDENT_WORK_LOOP
                                                 : OMP_IDENT_WORK_SECTIONS);
  llvm::Value *ThreadId = getThreadID(CGF, Loc);
  llvm::FunctionCallee StaticInitFunction =
      createForStaticInitFunction(Values.IVSize, Values.IVSigned, false);
  auto DL = ApplyDebugLocation::CreateDefaultArtificial(CGF, Loc);
  emitForStaticInitCall(CGF, UpdatedLocation, ThreadId, StaticInitFunction,
                        ScheduleNum, ScheduleKind.M1, ScheduleKind.M2, Values);
}
 
void CGOpenMPRuntime::emitDistributeStaticInit(
    CodeGenFunction &CGF, SourceLocation Loc,
    OpenMPDistScheduleClauseKind SchedKind,
    const CGOpenMPRuntime::StaticRTInput &Values) {
  OpenMPSchedType ScheduleNum =
      getRuntimeSchedule(SchedKind, Values.Chunk != nullptr);
  llvm::Value *UpdatedLocation =
      emitUpdateLocation(CGF, Loc, OMP_IDENT_WORK_DISTRIBUTE);
  llvm::Value *ThreadId = getThreadID(CGF, Loc);
  llvm::FunctionCallee StaticInitFunction;
  bool isGPUDistribute =
      CGM.getLangOpts().OpenMPIsDevice &&
      (CGM.getTriple().isAMDGCN() || CGM.getTriple().isNVPTX());
  StaticInitFunction = createForStaticInitFunction(
      Values.IVSize, Values.IVSigned, isGPUDistribute);
 
  emitForStaticInitCall(CGF, UpdatedLocation, ThreadId, StaticInitFunction,
                        ScheduleNum, OMPC_SCHEDULE_MODIFIER_unknown,
                        OMPC_SCHEDULE_MODIFIER_unknown, Values);
}
 
void CGOpenMPRuntime::emitForStaticFinish(CodeGenFunction &CGF,
                                          SourceLocation Loc,
                                          OpenMPDirectiveKind DKind) {
  if (!CGF.HaveInsertPoint())
    return;
  // Call __kmpc_for_static_fini(ident_t *loc, kmp_int32 tid);
  llvm::Value *Args[] = {
      emitUpdateLocation(CGF, Loc,
                         isOpenMPDistributeDirective(DKind)
                             ? OMP_IDENT_WORK_DISTRIBUTE
                             : isOpenMPLoopDirective(DKind)
                                   ? OMP_IDENT_WORK_LOOP
                                   : OMP_IDENT_WORK_SECTIONS),
      getThreadID(CGF, Loc)};
  auto DL = ApplyDebugLocation::CreateDefaultArtificial(CGF, Loc);
  if (isOpenMPDistributeDirective(DKind) && CGM.getLangOpts().OpenMPIsDevice &&
      (CGM.getTriple().isAMDGCN() || CGM.getTriple().isNVPTX()))
    CGF.EmitRuntimeCall(
        OMPBuilder.getOrCreateRuntimeFunction(
            CGM.getModule(), OMPRTL___kmpc_distribute_static_fini),
        Args);
  else
    CGF.EmitRuntimeCall(OMPBuilder.getOrCreateRuntimeFunction(
                            CGM.getModule(), OMPRTL___kmpc_for_static_fini),
                        Args);
}
 
void CGOpenMPRuntime::emitForOrderedIterationEnd(CodeGenFunction &CGF,
                                                 SourceLocation Loc,
                                                 unsigned IVSize,
                                                 bool IVSigned) {
  if (!CGF.HaveInsertPoint())
    return;
  // Call __kmpc_for_dynamic_fini_(4|8)[u](ident_t *loc, kmp_int32 tid);
  llvm::Value *Args[] = {emitUpdateLocation(CGF, Loc), getThreadID(CGF, Loc)};
  CGF.EmitRuntimeCall(createDispatchFiniFunction(IVSize, IVSigned), Args);
}
 
llvm::Value *CGOpenMPRuntime::emitForNext(CodeGenFunction &CGF,
                                          SourceLocation Loc, unsigned IVSize,
                                          bool IVSigned, Address IL,
                                          Address LB, Address UB,
                                          Address ST) {
  // Call __kmpc_dispatch_next(
  //          ident_t *loc, kmp_int32 tid, kmp_int32 *p_lastiter,
  //          kmp_int[32|64] *p_lower, kmp_int[32|64] *p_upper,
  //          kmp_int[32|64] *p_stride);
  llvm::Value *Args[] = {
      emitUpdateLocation(CGF, Loc),
      getThreadID(CGF, Loc),
      IL.getPointer(), // &isLastIter
      LB.getPointer(), // &Lower
      UB.getPointer(), // &Upper
      ST.getPointer()  // &Stride
  };
  llvm::Value *Call =
      CGF.EmitRuntimeCall(createDispatchNextFunction(IVSize, IVSigned), Args);
  return CGF.EmitScalarConversion(
      Call, CGF.getContext().getIntTypeForBitwidth(32, /*Signed=*/1),
      CGF.getContext().BoolTy, Loc);
}
 
void CGOpenMPRuntime::emitNumThreadsClause(CodeGenFunction &CGF,
                                           llvm::Value *NumThreads,
                                           SourceLocation Loc) {
  if (!CGF.HaveInsertPoint())
    return;
  // Build call __kmpc_push_num_threads(&loc, global_tid, num_threads)
  llvm::Value *Args[] = {
      emitUpdateLocation(CGF, Loc), getThreadID(CGF, Loc),
      CGF.Builder.CreateIntCast(NumThreads, CGF.Int32Ty, /*isSigned*/ true)};
  CGF.EmitRuntimeCall(OMPBuilder.getOrCreateRuntimeFunction(
                          CGM.getModule(), OMPRTL___kmpc_push_num_threads),
                      Args);
}
 
void CGOpenMPRuntime::emitProcBindClause(CodeGenFunction &CGF,
                                         ProcBindKind ProcBind,
                                         SourceLocation Loc) {
  if (!CGF.HaveInsertPoint())
    return;
  assert(ProcBind != OMP_PROC_BIND_unknown && "Unsupported proc_bind value.");
  // Build call __kmpc_push_proc_bind(&loc, global_tid, proc_bind)
  llvm::Value *Args[] = {
      emitUpdateLocation(CGF, Loc), getThreadID(CGF, Loc),
      llvm::ConstantInt::get(CGM.IntTy, unsigned(ProcBind), /*isSigned=*/true)};
  CGF.EmitRuntimeCall(OMPBuilder.getOrCreateRuntimeFunction(
                          CGM.getModule(), OMPRTL___kmpc_push_proc_bind),
                      Args);
}
 
void CGOpenMPRuntime::emitFlush(CodeGenFunction &CGF, ArrayRef<const Expr *>,
                                SourceLocation Loc, llvm::AtomicOrdering AO) {
  if (CGF.CGM.getLangOpts().OpenMPIRBuilder) {
    OMPBuilder.createFlush(CGF.Builder);
  } else {
    if (!CGF.HaveInsertPoint())
      return;
    // Build call void __kmpc_flush(ident_t *loc)
    CGF.EmitRuntimeCall(OMPBuilder.getOrCreateRuntimeFunction(
                            CGM.getModule(), OMPRTL___kmpc_flush),
                        emitUpdateLocation(CGF, Loc));
  }
}
 
namespace {
/// Indexes of fields for type kmp_task_t.
enum KmpTaskTFields {
  /// List of shared variables.
  KmpTaskTShareds,
  /// Task routine.
  KmpTaskTRoutine,
  /// Partition id for the untied tasks.
  KmpTaskTPartId,
  /// Function with call of destructors for private variables.
  Data1,
  /// Task priority.
  Data2,
  /// (Taskloops only) Lower bound.
  KmpTaskTLowerBound,
  /// (Taskloops only) Upper bound.
  KmpTaskTUpperBound,
  /// (Taskloops only) Stride.
  KmpTaskTStride,
  /// (Taskloops only) Is last iteration flag.
  KmpTaskTLastIter,
  /// (Taskloops only) Reduction data.
  KmpTaskTReductions,
};
} // anonymous namespace
 
void CGOpenMPRuntime::createOffloadEntriesAndInfoMetadata() {
  // If we are in simd mode or there are no entries, we don't need to do
  // anything.
  if (CGM.getLangOpts().OpenMPSimd || OMPBuilder.OffloadInfoManager.empty())
    return;
 
  llvm::OpenMPIRBuilder::EmitMetadataErrorReportFunctionTy &&ErrorReportFn =
      [this](llvm::OpenMPIRBuilder::EmitMetadataErrorKind Kind,
             const llvm::TargetRegionEntryInfo &EntryInfo) -> void {
    SourceLocation Loc;
    if (Kind != llvm::OpenMPIRBuilder::EMIT_MD_GLOBAL_VAR_LINK_ERROR) {
      for (auto I = CGM.getContext().getSourceManager().fileinfo_begin(),
                E = CGM.getContext().getSourceManager().fileinfo_end();
           I != E; ++I) {
        if (I->getFirst()->getUniqueID().getDevice() == EntryInfo.DeviceID &&
            I->getFirst()->getUniqueID().getFile() == EntryInfo.FileID) {
          Loc = CGM.getContext().getSourceManager().translateFileLineCol(
              I->getFirst(), EntryInfo.Line, 1);
          break;
        }
      }
    }
    switch (Kind) {
    case llvm::OpenMPIRBuilder::EMIT_MD_TARGET_REGION_ERROR: {
      unsigned DiagID = CGM.getDiags().getCustomDiagID(
          DiagnosticsEngine::Error, "Offloading entry for target region in "
                                    "%0 is incorrect: either the "
                                    "address or the ID is invalid.");
      CGM.getDiags().Report(Loc, DiagID) << EntryInfo.ParentName;
    } break;
    case llvm::OpenMPIRBuilder::EMIT_MD_DECLARE_TARGET_ERROR: {
      unsigned DiagID = CGM.getDiags().getCustomDiagID(
          DiagnosticsEngine::Error, "Offloading entry for declare target "
                                    "variable %0 is incorrect: the "
                                    "address is invalid.");
      CGM.getDiags().Report(Loc, DiagID) << EntryInfo.ParentName;
    } break;
    case llvm::OpenMPIRBuilder::EMIT_MD_GLOBAL_VAR_LINK_ERROR: {
      unsigned DiagID = CGM.getDiags().getCustomDiagID(
          DiagnosticsEngine::Error,
          "Offloading entry for declare target variable is incorrect: the "
          "address is invalid.");
      CGM.getDiags().Report(DiagID);
    } break;
    }
  };
 
  OMPBuilder.createOffloadEntriesAndInfoMetadata(ErrorReportFn);
}
 
/// Loads all the offload entries information from the host IR
/// metadata.
void CGOpenMPRuntime::loadOffloadInfoMetadata() {
  // If we are in target mode, load the metadata from the host IR. This code has
  // to match the metadaata creation in createOffloadEntriesAndInfoMetadata().
 
  if (!CGM.getLangOpts().OpenMPIsDevice)
    return;
 
  if (CGM.getLangOpts().OMPHostIRFile.empty())
    return;
 
  auto Buf = llvm::MemoryBuffer::getFile(CGM.getLangOpts().OMPHostIRFile);
  if (auto EC = Buf.getError()) {
    CGM.getDiags().Report(diag::err_cannot_open_file)
        << CGM.getLangOpts().OMPHostIRFile << EC.message();
    return;
  }
 
  llvm::LLVMContext C;
  auto ME = expectedToErrorOrAndEmitErrors(
      C, llvm::parseBitcodeFile(Buf.get()->getMemBufferRef(), C));
 
  if (auto EC = ME.getError()) {
    unsigned DiagID = CGM.getDiags().getCustomDiagID(
        DiagnosticsEngine::Error, "Unable to parse host IR file '%0':'%1'");
    CGM.getDiags().Report(DiagID)
        << CGM.getLangOpts().OMPHostIRFile << EC.message();
    return;
  }
 
  OMPBuilder.loadOffloadInfoMetadata(*ME.get());
}
 
void CGOpenMPRuntime::emitKmpRoutineEntryT(QualType KmpInt32Ty) {
  if (!KmpRoutineEntryPtrTy) {
    // Build typedef kmp_int32 (* kmp_routine_entry_t)(kmp_int32, void *); type.
    ASTContext &C = CGM.getContext();
    QualType KmpRoutineEntryTyArgs[] = {KmpInt32Ty, C.VoidPtrTy};
    FunctionProtoType::ExtProtoInfo EPI;
    KmpRoutineEntryPtrQTy = C.getPointerType(
        C.getFunctionType(KmpInt32Ty, KmpRoutineEntryTyArgs, EPI));
    KmpRoutineEntryPtrTy = CGM.getTypes().ConvertType(KmpRoutineEntryPtrQTy);
  }
}
 
namespace {
struct PrivateHelpersTy {
  PrivateHelpersTy(const Expr *OriginalRef, const VarDecl *Original,
                   const VarDecl *PrivateCopy, const VarDecl *PrivateElemInit)
      : OriginalRef(OriginalRef), Original(Original), PrivateCopy(PrivateCopy),
        PrivateElemInit(PrivateElemInit) {}
  PrivateHelpersTy(const VarDecl *Original) : Original(Original) {}
  const Expr *OriginalRef = nullptr;
  const VarDecl *Original = nullptr;
  const VarDecl *PrivateCopy = nullptr;
  const VarDecl *PrivateElemInit = nullptr;
  bool isLocalPrivate() const {
    return !OriginalRef && !PrivateCopy && !PrivateElemInit;
  }
};
typedef std::pair<CharUnits /*Align*/, PrivateHelpersTy> PrivateDataTy;
} // anonymous namespace
 
static bool isAllocatableDecl(const VarDecl *VD) {
  const VarDecl *CVD = VD->getCanonicalDecl();
  if (!CVD->hasAttr<OMPAllocateDeclAttr>())
    return false;
  const auto *AA = CVD->getAttr<OMPAllocateDeclAttr>();
  // Use the default allocation.
  return !(AA->getAllocatorType() == OMPAllocateDeclAttr::OMPDefaultMemAlloc &&
           !AA->getAllocator());
}
 
static RecordDecl *
createPrivatesRecordDecl(CodeGenModule &CGM, ArrayRef<PrivateDataTy> Privates) {
  if (!Privates.empty()) {
    ASTContext &C = CGM.getContext();
    // Build struct .kmp_privates_t. {
    //         /*  private vars  */
    //       };
    RecordDecl *RD = C.buildImplicitRecord(".kmp_privates.t");
    RD->startDefinition();
    for (const auto &Pair : Privates) {
      const VarDecl *VD = Pair.second.Original;
      QualType Type = VD->getType().getNonReferenceType();
      // If the private variable is a local variable with lvalue ref type,
      // allocate the pointer instead of the pointee type.
      if (Pair.second.isLocalPrivate()) {
        if (VD->getType()->isLValueReferenceType())
          Type = C.getPointerType(Type);
        if (isAllocatableDecl(VD))
          Type = C.getPointerType(Type);
      }
      FieldDecl *FD = addFieldToRecordDecl(C, RD, Type);
      if (VD->hasAttrs()) {
        for (specific_attr_iterator<AlignedAttr> I(VD->getAttrs().begin()),
             E(VD->getAttrs().end());
             I != E; ++I)
          FD->addAttr(*I);
      }
    }
    RD->completeDefinition();
    return RD;
  }
  return nullptr;
}
 
static RecordDecl *
createKmpTaskTRecordDecl(CodeGenModule &CGM, OpenMPDirectiveKind Kind,
                         QualType KmpInt32Ty,
                         QualType KmpRoutineEntryPointerQTy) {
  ASTContext &C = CGM.getContext();
  // Build struct kmp_task_t {
  //         void *              shareds;
  //         kmp_routine_entry_t routine;
  //         kmp_int32           part_id;
  //         kmp_cmplrdata_t data1;
  //         kmp_cmplrdata_t data2;
  // For taskloops additional fields:
  //         kmp_uint64          lb;
  //         kmp_uint64          ub;
  //         kmp_int64           st;
  //         kmp_int32           liter;
  //         void *              reductions;
  //       };
  RecordDecl *UD = C.buildImplicitRecord("kmp_cmplrdata_t", TTK_Union);
  UD->startDefinition();
  addFieldToRecordDecl(C, UD, KmpInt32Ty);
  addFieldToRecordDecl(C, UD, KmpRoutineEntryPointerQTy);
  UD->completeDefinition();
  QualType KmpCmplrdataTy = C.getRecordType(UD);
  RecordDecl *RD = C.buildImplicitRecord("kmp_task_t");
  RD->startDefinition();
  addFieldToRecordDecl(C, RD, C.VoidPtrTy);
  addFieldToRecordDecl(C, RD, KmpRoutineEntryPointerQTy);
  addFieldToRecordDecl(C, RD, KmpInt32Ty);
  addFieldToRecordDecl(C, RD, KmpCmplrdataTy);
  addFieldToRecordDecl(C, RD, KmpCmplrdataTy);
  if (isOpenMPTaskLoopDirective(Kind)) {
    QualType KmpUInt64Ty =
        CGM.getContext().getIntTypeForBitwidth(/*DestWidth=*/64, /*Signed=*/0);
    QualType KmpInt64Ty =
        CGM.getContext().getIntTypeForBitwidth(/*DestWidth=*/64, /*Signed=*/1);
    addFieldToRecordDecl(C, RD, KmpUInt64Ty);
    addFieldToRecordDecl(C, RD, KmpUInt64Ty);
    addFieldToRecordDecl(C, RD, KmpInt64Ty);
    addFieldToRecordDecl(C, RD, KmpInt32Ty);
    addFieldToRecordDecl(C, RD, C.VoidPtrTy);
  }
  RD->completeDefinition();
  return RD;
}
 
static RecordDecl *
createKmpTaskTWithPrivatesRecordDecl(CodeGenModule &CGM, QualType KmpTaskTQTy,
                                     ArrayRef<PrivateDataTy> Privates) {
  ASTContext &C = CGM.getContext();
  // Build struct kmp_task_t_with_privates {
  //         kmp_task_t task_data;
  //         .kmp_privates_t. privates;
  //       };
  RecordDecl *RD = C.buildImplicitRecord("kmp_task_t_with_privates");
  RD->startDefinition();
  addFieldToRecordDecl(C, RD, KmpTaskTQTy);
  if (const RecordDecl *PrivateRD = createPrivatesRecordDecl(CGM, Privates))
    addFieldToRecordDecl(C, RD, C.getRecordType(PrivateRD));
  RD->completeDefinition();
  return RD;
}
 
/// Emit a proxy function which accepts kmp_task_t as the second
/// argument.
/// \code
/// kmp_int32 .omp_task_entry.(kmp_int32 gtid, kmp_task_t *tt) {
///   TaskFunction(gtid, tt->part_id, &tt->privates, task_privates_map, tt,
///   For taskloops:
///   tt->task_data.lb, tt->task_data.ub, tt->task_data.st, tt->task_data.liter,
///   tt->reductions, tt->shareds);
///   return 0;
/// }
/// \endcode
static llvm::Function *
emitProxyTaskFunction(CodeGenModule &CGM, SourceLocation Loc,
                      OpenMPDirectiveKind Kind, QualType KmpInt32Ty,
                      QualType KmpTaskTWithPrivatesPtrQTy,
                      QualType KmpTaskTWithPrivatesQTy, QualType KmpTaskTQTy,
                      QualType SharedsPtrTy, llvm::Function *TaskFunction,
                      llvm::Value *TaskPrivatesMap) {
  ASTContext &C = CGM.getContext();
  FunctionArgList Args;
  ImplicitParamDecl GtidArg(C, /*DC=*/nullptr, Loc, /*Id=*/nullptr, KmpInt32Ty,
                            ImplicitParamDecl::Other);
  ImplicitParamDecl TaskTypeArg(C, /*DC=*/nullptr, Loc, /*Id=*/nullptr,
                                KmpTaskTWithPrivatesPtrQTy.withRestrict(),
                                ImplicitParamDecl::Other);
  Args.push_back(&GtidArg);
  Args.push_back(&TaskTypeArg);
  const auto &TaskEntryFnInfo =
      CGM.getTypes().arrangeBuiltinFunctionDeclaration(KmpInt32Ty, Args);
  llvm::FunctionType *TaskEntryTy =
      CGM.getTypes().GetFunctionType(TaskEntryFnInfo);
  std::string Name = CGM.getOpenMPRuntime().getName({"omp_task_entry", ""});
  auto *TaskEntry = llvm::Function::Create(
      TaskEntryTy, llvm::GlobalValue::InternalLinkage, Name, &CGM.getModule());
  CGM.SetInternalFunctionAttributes(GlobalDecl(), TaskEntry, TaskEntryFnInfo);
  TaskEntry->setDoesNotRecurse();
  CodeGenFunction CGF(CGM);
  CGF.StartFunction(GlobalDecl(), KmpInt32Ty, TaskEntry, TaskEntryFnInfo, Args,
                    Loc, Loc);
 
  // TaskFunction(gtid, tt->task_data.part_id, &tt->privates, task_privates_map,
  // tt,
  // For taskloops:
  // tt->task_data.lb, tt->task_data.ub, tt->task_data.st, tt->task_data.liter,
  // tt->task_data.shareds);
  llvm::Value *GtidParam = CGF.EmitLoadOfScalar(
      CGF.GetAddrOfLocalVar(&GtidArg), /*Volatile=*/false, KmpInt32Ty, Loc);
  LValue TDBase = CGF.EmitLoadOfPointerLValue(
      CGF.GetAddrOfLocalVar(&TaskTypeArg),
      KmpTaskTWithPrivatesPtrQTy->castAs<PointerType>());
  const auto *KmpTaskTWithPrivatesQTyRD =
      cast<RecordDecl>(KmpTaskTWithPrivatesQTy->getAsTagDecl());
  LValue Base =
      CGF.EmitLValueForField(TDBase, *KmpTaskTWithPrivatesQTyRD->field_begin());
  const auto *KmpTaskTQTyRD = cast<RecordDecl>(KmpTaskTQTy->getAsTagDecl());
  auto PartIdFI = std::next(KmpTaskTQTyRD->field_begin(), KmpTaskTPartId);
  LValue PartIdLVal = CGF.EmitLValueForField(Base, *PartIdFI);
  llvm::Value *PartidParam = PartIdLVal.getPointer(CGF);
 
  auto SharedsFI = std::next(KmpTaskTQTyRD->field_begin(), KmpTaskTShareds);
  LValue SharedsLVal = CGF.EmitLValueForField(Base, *SharedsFI);
  llvm::Value *SharedsParam = CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
      CGF.EmitLoadOfScalar(SharedsLVal, Loc),
      CGF.ConvertTypeForMem(SharedsPtrTy));
 
  auto PrivatesFI = std::next(KmpTaskTWithPrivatesQTyRD->field_begin(), 1);
  llvm::Value *PrivatesParam;
  if (PrivatesFI != KmpTaskTWithPrivatesQTyRD->field_end()) {
    LValue PrivatesLVal = CGF.EmitLValueForField(TDBase, *PrivatesFI);
    PrivatesParam = CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
        PrivatesLVal.getPointer(CGF), CGF.VoidPtrTy);
  } else {
    PrivatesParam = llvm::ConstantPointerNull::get(CGF.VoidPtrTy);
  }
 
  llvm::Value *CommonArgs[] = {
      GtidParam, PartidParam, PrivatesParam, TaskPrivatesMap,
      CGF.Builder
          .CreatePointerBitCastOrAddrSpaceCast(TDBase.getAddress(CGF),
                                               CGF.VoidPtrTy, CGF.Int8Ty)
          .getPointer()};
  SmallVector<llvm::Value *, 16> CallArgs(std::begin(CommonArgs),
                                          std::end(CommonArgs));
  if (isOpenMPTaskLoopDirective(Kind)) {
    auto LBFI = std::next(KmpTaskTQTyRD->field_begin(), KmpTaskTLowerBound);
    LValue LBLVal = CGF.EmitLValueForField(Base, *LBFI);
    llvm::Value *LBParam = CGF.EmitLoadOfScalar(LBLVal, Loc);
    auto UBFI = std::next(KmpTaskTQTyRD->field_begin(), KmpTaskTUpperBound);
    LValue UBLVal = CGF.EmitLValueForField(Base, *UBFI);
    llvm::Value *UBParam = CGF.EmitLoadOfScalar(UBLVal, Loc);
    auto StFI = std::next(KmpTaskTQTyRD->field_begin(), KmpTaskTStride);
    LValue StLVal = CGF.EmitLValueForField(Base, *StFI);
    llvm::Value *StParam = CGF.EmitLoadOfScalar(StLVal, Loc);
    auto LIFI = std::next(KmpTaskTQTyRD->field_begin(), KmpTaskTLastIter);
    LValue LILVal = CGF.EmitLValueForField(Base, *LIFI);
    llvm::Value *LIParam = CGF.EmitLoadOfScalar(LILVal, Loc);
    auto RFI = std::next(KmpTaskTQTyRD->field_begin(), KmpTaskTReductions);
    LValue RLVal = CGF.EmitLValueForField(Base, *RFI);
    llvm::Value *RParam = CGF.EmitLoadOfScalar(RLVal, Loc);
    CallArgs.push_back(LBParam);
    CallArgs.push_back(UBParam);
    CallArgs.push_back(StParam);
    CallArgs.push_back(LIParam);
    CallArgs.push_back(RParam);
  }
  CallArgs.push_back(SharedsParam);
 
  CGM.getOpenMPRuntime().emitOutlinedFunctionCall(CGF, Loc, TaskFunction,
                                                  CallArgs);
  CGF.EmitStoreThroughLValue(RValue::get(CGF.Builder.getInt32(/*C=*/0)),
                             CGF.MakeAddrLValue(CGF.ReturnValue, KmpInt32Ty));
  CGF.FinishFunction();
  return TaskEntry;
}
 
static llvm::Value *emitDestructorsFunction(CodeGenModule &CGM,
                                            SourceLocation Loc,
                                            QualType KmpInt32Ty,
                                            QualType KmpTaskTWithPrivatesPtrQTy,
                                            QualType KmpTaskTWithPrivatesQTy) {
  ASTContext &C = CGM.getContext();
  FunctionArgList Args;
  ImplicitParamDecl GtidArg(C, /*DC=*/nullptr, Loc, /*Id=*/nullptr, KmpInt32Ty,
                            ImplicitParamDecl::Other);
  ImplicitParamDecl TaskTypeArg(C, /*DC=*/nullptr, Loc, /*Id=*/nullptr,
                                KmpTaskTWithPrivatesPtrQTy.withRestrict(),
                                ImplicitParamDecl::Other);
  Args.push_back(&GtidArg);
  Args.push_back(&TaskTypeArg);
  const auto &DestructorFnInfo =
      CGM.getTypes().arrangeBuiltinFunctionDeclaration(KmpInt32Ty, Args);
  llvm::FunctionType *DestructorFnTy =
      CGM.getTypes().GetFunctionType(DestructorFnInfo);
  std::string Name =
      CGM.getOpenMPRuntime().getName({"omp_task_destructor", ""});
  auto *DestructorFn =
      llvm::Function::Create(DestructorFnTy, llvm::GlobalValue::InternalLinkage,
                             Name, &CGM.getModule());
  CGM.SetInternalFunctionAttributes(GlobalDecl(), DestructorFn,
                                    DestructorFnInfo);
  DestructorFn->setDoesNotRecurse();
  CodeGenFunction CGF(CGM);
  CGF.StartFunction(GlobalDecl(), KmpInt32Ty, DestructorFn, DestructorFnInfo,
                    Args, Loc, Loc);
 
  LValue Base = CGF.EmitLoadOfPointerLValue(
      CGF.GetAddrOfLocalVar(&TaskTypeArg),
      KmpTaskTWithPrivatesPtrQTy->castAs<PointerType>());
  const auto *KmpTaskTWithPrivatesQTyRD =
      cast<RecordDecl>(KmpTaskTWithPrivatesQTy->getAsTagDecl());
  auto FI = std::next(KmpTaskTWithPrivatesQTyRD->field_begin());
  Base = CGF.EmitLValueForField(Base, *FI);
  for (const auto *Field :
       cast<RecordDecl>(FI->getType()->getAsTagDecl())->fields()) {
    if (QualType::DestructionKind DtorKind =
            Field->getType().isDestructedType()) {
      LValue FieldLValue = CGF.EmitLValueForField(Base, Field);
      CGF.pushDestroy(DtorKind, FieldLValue.getAddress(CGF), Field->getType());
    }
  }
  CGF.FinishFunction();
  return DestructorFn;
}
 
/// Emit a privates mapping function for correct handling of private and
/// firstprivate variables.
/// \code
/// void .omp_task_privates_map.(const .privates. *noalias privs, <ty1>
/// **noalias priv1,...,  <tyn> **noalias privn) {
///   *priv1 = &.privates.priv1;
///   ...;
///   *privn = &.privates.privn;
/// }
/// \endcode
static llvm::Value *
emitTaskPrivateMappingFunction(CodeGenModule &CGM, SourceLocation Loc,
                               const OMPTaskDataTy &Data, QualType PrivatesQTy,
                               ArrayRef<PrivateDataTy> Privates) {
  ASTContext &C = CGM.getContext();
  FunctionArgList Args;
  ImplicitParamDecl TaskPrivatesArg(
      C, /*DC=*/nullptr, Loc, /*Id=*/nullptr,
      C.getPointerType(PrivatesQTy).withConst().withRestrict(),
      ImplicitParamDecl::Other);
  Args.push_back(&TaskPrivatesArg);
  llvm::DenseMap<CanonicalDeclPtr<const VarDecl>, unsigned> PrivateVarsPos;
  unsigned Counter = 1;
  for (const Expr *E : Data.PrivateVars) {
    Args.push_back(ImplicitParamDecl::Create(
        C, /*DC=*/nullptr, Loc, /*Id=*/nullptr,
        C.getPointerType(C.getPointerType(E->getType()))
            .withConst()
            .withRestrict(),
        ImplicitParamDecl::Other));
    const auto *VD = cast<VarDecl>(cast<DeclRefExpr>(E)->getDecl());
    PrivateVarsPos[VD] = Counter;
    ++Counter;
  }
  for (const Expr *E : Data.FirstprivateVars) {
    Args.push_back(ImplicitParamDecl::Create(
        C, /*DC=*/nullptr, Loc, /*Id=*/nullptr,
        C.getPointerType(C.getPointerType(E->getType()))
            .withConst()
            .withRestrict(),
        ImplicitParamDecl::Other));
    const auto *VD = cast<VarDecl>(cast<DeclRefExpr>(E)->getDecl());
    PrivateVarsPos[VD] = Counter;
    ++Counter;
  }
  for (const Expr *E : Data.LastprivateVars) {
    Args.push_back(ImplicitParamDecl::Create(
        C, /*DC=*/nullptr, Loc, /*Id=*/nullptr,
        C.getPointerType(C.getPointerType(E->getType()))
            .withConst()
            .withRestrict(),
        ImplicitParamDecl::Other));
    const auto *VD = cast<VarDecl>(cast<DeclRefExpr>(E)->getDecl());
    PrivateVarsPos[VD] = Counter;
    ++Counter;
  }
  for (const VarDecl *VD : Data.PrivateLocals) {
    QualType Ty = VD->getType().getNonReferenceType();
    if (VD->getType()->isLValueReferenceType())
      Ty = C.getPointerType(Ty);
    if (isAllocatableDecl(VD))
      Ty = C.getPointerType(Ty);
    Args.push_back(ImplicitParamDecl::Create(
        C, /*DC=*/nullptr, Loc, /*Id=*/nullptr,
        C.getPointerType(C.getPointerType(Ty)).withConst().withRestrict(),
        ImplicitParamDecl::Other));
    PrivateVarsPos[VD] = Counter;
    ++Counter;
  }
  const auto &TaskPrivatesMapFnInfo =
      CGM.getTypes().arrangeBuiltinFunctionDeclaration(C.VoidTy, Args);
  llvm::FunctionType *TaskPrivatesMapTy =
      CGM.getTypes().GetFunctionType(TaskPrivatesMapFnInfo);
  std::string Name =
      CGM.getOpenMPRuntime().getName({"omp_task_privates_map", ""});
  auto *TaskPrivatesMap = llvm::Function::Create(
      TaskPrivatesMapTy, llvm::GlobalValue::InternalLinkage, Name,
      &CGM.getModule());
  CGM.SetInternalFunctionAttributes(GlobalDecl(), TaskPrivatesMap,
                                    TaskPrivatesMapFnInfo);
  if (CGM.getLangOpts().Optimize) {
    TaskPrivatesMap->removeFnAttr(llvm::Attribute::NoInline);
    TaskPrivatesMap->removeFnAttr(llvm::Attribute::OptimizeNone);
    TaskPrivatesMap->addFnAttr(llvm::Attribute::AlwaysInline);
  }
  CodeGenFunction CGF(CGM);
  CGF.StartFunction(GlobalDecl(), C.VoidTy, TaskPrivatesMap,
                    TaskPrivatesMapFnInfo, Args, Loc, Loc);
 
  // *privi = &.privates.privi;
  LValue Base = CGF.EmitLoadOfPointerLValue(
      CGF.GetAddrOfLocalVar(&TaskPrivatesArg),
      TaskPrivatesArg.getType()->castAs<PointerType>());
  const auto *PrivatesQTyRD = cast<RecordDecl>(PrivatesQTy->getAsTagDecl());
  Counter = 0;
  for (const FieldDecl *Field : PrivatesQTyRD->fields()) {
    LValue FieldLVal = CGF.EmitLValueForField(Base, Field);
    const VarDecl *VD = Args[PrivateVarsPos[Privates[Counter].second.Original]];
    LValue RefLVal =
        CGF.MakeAddrLValue(CGF.GetAddrOfLocalVar(VD), VD->getType());
    LValue RefLoadLVal = CGF.EmitLoadOfPointerLValue(
        RefLVal.getAddress(CGF), RefLVal.getType()->castAs<PointerType>());
    CGF.EmitStoreOfScalar(FieldLVal.getPointer(CGF), RefLoadLVal);
    ++Counter;
  }
  CGF.FinishFunction();
  return TaskPrivatesMap;
}
 
/// Emit initialization for private variables in task-based directives.
static void emitPrivatesInit(CodeGenFunction &CGF,
                             const OMPExecutableDirective &D,
                             Address KmpTaskSharedsPtr, LValue TDBase,
                             const RecordDecl *KmpTaskTWithPrivatesQTyRD,
                             QualType SharedsTy, QualType SharedsPtrTy,
                             const OMPTaskDataTy &Data,
                             ArrayRef<PrivateDataTy> Privates, bool ForDup) {
  ASTContext &C = CGF.getContext();
  auto FI = std::next(KmpTaskTWithPrivatesQTyRD->field_begin());
  LValue PrivatesBase = CGF.EmitLValueForField(TDBase, *FI);
  OpenMPDirectiveKind Kind = isOpenMPTaskLoopDirective(D.getDirectiveKind())
                                 ? OMPD_taskloop
                                 : OMPD_task;
  const CapturedStmt &CS = *D.getCapturedStmt(Kind);
  CodeGenFunction::CGCapturedStmtInfo CapturesInfo(CS);
  LValue SrcBase;
  bool IsTargetTask =
      isOpenMPTargetDataManagementDirective(D.getDirectiveKind()) ||
      isOpenMPTargetExecutionDirective(D.getDirectiveKind());
  // For target-based directives skip 4 firstprivate arrays BasePointersArray,
  // PointersArray, SizesArray, and MappersArray. The original variables for
  // these arrays are not captured and we get their addresses explicitly.
  if ((!IsTargetTask && !Data.FirstprivateVars.empty() && ForDup) ||
      (IsTargetTask && KmpTaskSharedsPtr.isValid())) {
    SrcBase = CGF.MakeAddrLValue(
        CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
            KmpTaskSharedsPtr, CGF.ConvertTypeForMem(SharedsPtrTy),
            CGF.ConvertTypeForMem(SharedsTy)),
        SharedsTy);
  }
  FI = cast<RecordDecl>(FI->getType()->getAsTagDecl())->field_begin();
  for (const PrivateDataTy &Pair : Privates) {
    // Do not initialize private locals.
    if (Pair.second.isLocalPrivate()) {
      ++FI;
      continue;
    }
    const VarDecl *VD = Pair.second.PrivateCopy;
    const Expr *Init = VD->getAnyInitializer();
    if (Init && (!ForDup || (isa<CXXConstructExpr>(Init) &&
                             !CGF.isTrivialInitializer(Init)))) {
      LValue PrivateLValue = CGF.EmitLValueForField(PrivatesBase, *FI);
      if (const VarDecl *Elem = Pair.second.PrivateElemInit) {
        const VarDecl *OriginalVD = Pair.second.Original;
        // Check if the variable is the target-based BasePointersArray,
        // PointersArray, SizesArray, or MappersArray.
        LValue SharedRefLValue;
        QualType Type = PrivateLValue.getType();
        const FieldDecl *SharedField = CapturesInfo.lookup(OriginalVD);
        if (IsTargetTask && !SharedField) {
          assert(isa<ImplicitParamDecl>(OriginalVD) &&
                 isa<CapturedDecl>(OriginalVD->getDeclContext()) &&
                 cast<CapturedDecl>(OriginalVD->getDeclContext())
                         ->getNumParams() == 0 &&
                 isa<TranslationUnitDecl>(
                     cast<CapturedDecl>(OriginalVD->getDeclContext())
                         ->getDeclContext()) &&
                 "Expected artificial target data variable.");
          SharedRefLValue =
              CGF.MakeAddrLValue(CGF.GetAddrOfLocalVar(OriginalVD), Type);
        } else if (ForDup) {
          SharedRefLValue = CGF.EmitLValueForField(SrcBase, SharedField);
          SharedRefLValue = CGF.MakeAddrLValue(
              SharedRefLValue.getAddress(CGF).withAlignment(
                  C.getDeclAlign(OriginalVD)),
              SharedRefLValue.getType(), LValueBaseInfo(AlignmentSource::Decl),
              SharedRefLValue.getTBAAInfo());
        } else if (CGF.LambdaCaptureFields.count(
                       Pair.second.Original->getCanonicalDecl()) > 0 ||
                   isa_and_nonnull<BlockDecl>(CGF.CurCodeDecl)) {
          SharedRefLValue = CGF.EmitLValue(Pair.second.OriginalRef);
        } else {
          // Processing for implicitly captured variables.
          InlinedOpenMPRegionRAII Region(
              CGF, [](CodeGenFunction &, PrePostActionTy &) {}, OMPD_unknown,
              /*HasCancel=*/false, /*NoInheritance=*/true);
          SharedRefLValue = CGF.EmitLValue(Pair.second.OriginalRef);
        }
        if (Type->isArrayType()) {
          // Initialize firstprivate array.
          if (!isa<CXXConstructExpr>(Init) || CGF.isTrivialInitializer(Init)) {
            // Perform simple memcpy.
            CGF.EmitAggregateAssign(PrivateLValue, SharedRefLValue, Type);
          } else {
            // Initialize firstprivate array using element-by-element
            // initialization.
            CGF.EmitOMPAggregateAssign(
                PrivateLValue.getAddress(CGF), SharedRefLValue.getAddress(CGF),
                Type,
                [&CGF, Elem, Init, &CapturesInfo](Address DestElement,
                                                  Address SrcElement) {
                  // Clean up any temporaries needed by the initialization.
                  CodeGenFunction::OMPPrivateScope InitScope(CGF);
                  InitScope.addPrivate(Elem, SrcElement);
                  (void)InitScope.Privatize();
                  // Emit initialization for single element.
                  CodeGenFunction::CGCapturedStmtRAII CapInfoRAII(
                      CGF, &CapturesInfo);
                  CGF.EmitAnyExprToMem(Init, DestElement,
                                       Init->getType().getQualifiers(),
                                       /*IsInitializer=*/false);
                });
          }
        } else {
          CodeGenFunction::OMPPrivateScope InitScope(CGF);
          InitScope.addPrivate(Elem, SharedRefLValue.getAddress(CGF));
          (void)InitScope.Privatize();
          CodeGenFunction::CGCapturedStmtRAII CapInfoRAII(CGF, &CapturesInfo);
          CGF.EmitExprAsInit(Init, VD, PrivateLValue,
                             /*capturedByInit=*/false);
        }
      } else {
        CGF.EmitExprAsInit(Init, VD, PrivateLValue, /*capturedByInit=*/false);
      }
    }
    ++FI;
  }
}
 
/// Check if duplication function is required for taskloops.
static bool checkInitIsRequired(CodeGenFunction &CGF,
                                ArrayRef<PrivateDataTy> Privates) {
  bool InitRequired = false;
  for (const PrivateDataTy &Pair : Privates) {
    if (Pair.second.isLocalPrivate())
      continue;
    const VarDecl *VD = Pair.second.PrivateCopy;
    const Expr *Init = VD->getAnyInitializer();
    InitRequired = InitRequired || (isa_and_nonnull<CXXConstructExpr>(Init) &&
                                    !CGF.isTrivialInitializer(Init));
    if (InitRequired)
      break;
  }
  return InitRequired;
}
 
 
/// Emit task_dup function (for initialization of
/// private/firstprivate/lastprivate vars and last_iter flag)
/// \code
/// void __task_dup_entry(kmp_task_t *task_dst, const kmp_task_t *task_src, int
/// lastpriv) {
/// // setup lastprivate flag
///    task_dst->last = lastpriv;
/// // could be constructor calls here...
/// }
/// \endcode
static llvm::Value *
emitTaskDupFunction(CodeGenModule &CGM, SourceLocation Loc,
                    const OMPExecutableDirective &D,
                    QualType KmpTaskTWithPrivatesPtrQTy,
                    const RecordDecl *KmpTaskTWithPrivatesQTyRD,
                    const RecordDecl *KmpTaskTQTyRD, QualType SharedsTy,
                    QualType SharedsPtrTy, const OMPTaskDataTy &Data,
                    ArrayRef<PrivateDataTy> Privates, bool WithLastIter) {
  ASTContext &C = CGM.getContext();
  FunctionArgList Args;
  ImplicitParamDecl DstArg(C, /*DC=*/nullptr, Loc, /*Id=*/nullptr,
                           KmpTaskTWithPrivatesPtrQTy,
                           ImplicitParamDecl::Other);
  ImplicitParamDecl SrcArg(C, /*DC=*/nullptr, Loc, /*Id=*/nullptr,
                           KmpTaskTWithPrivatesPtrQTy,
                           ImplicitParamDecl::Other);
  ImplicitParamDecl LastprivArg(C, /*DC=*/nullptr, Loc, /*Id=*/nullptr, C.IntTy,
                                ImplicitParamDecl::Other);
  Args.push_back(&DstArg);
  Args.push_back(&SrcArg);
  Args.push_back(&LastprivArg);
  const auto &TaskDupFnInfo =
      CGM.getTypes().arrangeBuiltinFunctionDeclaration(C.VoidTy, Args);
  llvm::FunctionType *TaskDupTy = CGM.getTypes().GetFunctionType(TaskDupFnInfo);
  std::string Name = CGM.getOpenMPRuntime().getName({"omp_task_dup", ""});
  auto *TaskDup = llvm::Function::Create(
      TaskDupTy, llvm::GlobalValue::InternalLinkage, Name, &CGM.getModule());
  CGM.SetInternalFunctionAttributes(GlobalDecl(), TaskDup, TaskDupFnInfo);
  TaskDup->setDoesNotRecurse();
  CodeGenFunction CGF(CGM);
  CGF.StartFunction(GlobalDecl(), C.VoidTy, TaskDup, TaskDupFnInfo, Args, Loc,
                    Loc);
 
  LValue TDBase = CGF.EmitLoadOfPointerLValue(
      CGF.GetAddrOfLocalVar(&DstArg),
      KmpTaskTWithPrivatesPtrQTy->castAs<PointerType>());
  // task_dst->liter = lastpriv;
  if (WithLastIter) {
    auto LIFI = std::next(KmpTaskTQTyRD->field_begin(), KmpTaskTLastIter);
    LValue Base = CGF.EmitLValueForField(
        TDBase, *KmpTaskTWithPrivatesQTyRD->field_begin());
    LValue LILVal = CGF.EmitLValueForField(Base, *LIFI);
    llvm::Value *Lastpriv = CGF.EmitLoadOfScalar(
        CGF.GetAddrOfLocalVar(&LastprivArg), /*Volatile=*/false, C.IntTy, Loc);
    CGF.EmitStoreOfScalar(Lastpriv, LILVal);
  }
 
  // Emit initial values for private copies (if any).
  assert(!Privates.empty());
  Address KmpTaskSharedsPtr = Address::invalid();
  if (!Data.FirstprivateVars.empty()) {
    LValue TDBase = CGF.EmitLoadOfPointerLValue(
        CGF.GetAddrOfLocalVar(&SrcArg),
        KmpTaskTWithPrivatesPtrQTy->castAs<PointerType>());
    LValue Base = CGF.EmitLValueForField(
        TDBase, *KmpTaskTWithPrivatesQTyRD->field_begin());
    KmpTaskSharedsPtr = Address(
        CGF.EmitLoadOfScalar(CGF.EmitLValueForField(
                                 Base, *std::next(KmpTaskTQTyRD->field_begin(),
                                                  KmpTaskTShareds)),
                             Loc),
        CGF.Int8Ty, CGM.getNaturalTypeAlignment(SharedsTy));
  }
  emitPrivatesInit(CGF, D, KmpTaskSharedsPtr, TDBase, KmpTaskTWithPrivatesQTyRD,
                   SharedsTy, SharedsPtrTy, Data, Privates, /*ForDup=*/true);
  CGF.FinishFunction();
  return TaskDup;
}
 
/// Checks if destructor function is required to be generated.
/// \return true if cleanups are required, false otherwise.
static bool
checkDestructorsRequired(const RecordDecl *KmpTaskTWithPrivatesQTyRD,
                         ArrayRef<PrivateDataTy> Privates) {
  for (const PrivateDataTy &P : Privates) {
    if (P.second.isLocalPrivate())
      continue;
    QualType Ty = P.second.Original->getType().getNonReferenceType();
    if (Ty.isDestructedType())
      return true;
  }
  return false;
}
 
namespace {
/// Loop generator for OpenMP iterator expression.
class OMPIteratorGeneratorScope final
    : public CodeGenFunction::OMPPrivateScope {
  CodeGenFunction &CGF;
  const OMPIteratorExpr *E = nullptr;
  SmallVector<CodeGenFunction::JumpDest, 4> ContDests;
  SmallVector<CodeGenFunction::JumpDest, 4> ExitDests;
  OMPIteratorGeneratorScope() = delete;
  OMPIteratorGeneratorScope(OMPIteratorGeneratorScope &) = delete;
 
public:
  OMPIteratorGeneratorScope(CodeGenFunction &CGF, const OMPIteratorExpr *E)
      : CodeGenFunction::OMPPrivateScope(CGF), CGF(CGF), E(E) {
    if (!E)
      return;
    SmallVector<llvm::Value *, 4> Uppers;
    for (unsigned I = 0, End = E->numOfIterators(); I < End; ++I) {
      Uppers.push_back(CGF.EmitScalarExpr(E->getHelper(I).Upper));
      const auto *VD = cast<VarDecl>(E->getIteratorDecl(I));
      addPrivate(VD, CGF.CreateMemTemp(VD->getType(), VD->getName()));
      const OMPIteratorHelperData &HelperData = E->getHelper(I);
      addPrivate(
          HelperData.CounterVD,
          CGF.CreateMemTemp(HelperData.CounterVD->getType(), "counter.addr"));
    }
    Privatize();
 
    for (unsigned I = 0, End = E->numOfIterators(); I < End; ++I) {
      const OMPIteratorHelperData &HelperData = E->getHelper(I);
      LValue CLVal =
          CGF.MakeAddrLValue(CGF.GetAddrOfLocalVar(HelperData.CounterVD),
                             HelperData.CounterVD->getType());
      // Counter = 0;
      CGF.EmitStoreOfScalar(
          llvm::ConstantInt::get(CLVal.getAddress(CGF).getElementType(), 0),
          CLVal);
      CodeGenFunction::JumpDest &ContDest =
          ContDests.emplace_back(CGF.getJumpDestInCurrentScope("iter.cont"));
      CodeGenFunction::JumpDest &ExitDest =
          ExitDests.emplace_back(CGF.getJumpDestInCurrentScope("iter.exit"));
      // N = <number-of_iterations>;
      llvm::Value *N = Uppers[I];
      // cont:
      // if (Counter < N) goto body; else goto exit;
      CGF.EmitBlock(ContDest.getBlock());
      auto *CVal =
          CGF.EmitLoadOfScalar(CLVal, HelperData.CounterVD->getLocation());
      llvm::Value *Cmp =
          HelperData.CounterVD->getType()->isSignedIntegerOrEnumerationType()
              ? CGF.Builder.CreateICmpSLT(CVal, N)
              : CGF.Builder.CreateICmpULT(CVal, N);
      llvm::BasicBlock *BodyBB = CGF.createBasicBlock("iter.body");
      CGF.Builder.CreateCondBr(Cmp, BodyBB, ExitDest.getBlock());
      // body:
      CGF.EmitBlock(BodyBB);
      // Iteri = Begini + Counter * Stepi;
      CGF.EmitIgnoredExpr(HelperData.Update);
    }
  }
  ~OMPIteratorGeneratorScope() {
    if (!E)
      return;
    for (unsigned I = E->numOfIterators(); I > 0; --I) {
      // Counter = Counter + 1;
      const OMPIteratorHelperData &HelperData = E->getHelper(I - 1);
      CGF.EmitIgnoredExpr(HelperData.CounterUpdate);
      // goto cont;
      CGF.EmitBranchThroughCleanup(ContDests[I - 1]);
      // exit:
      CGF.EmitBlock(ExitDests[I - 1].getBlock(), /*IsFinished=*/I == 1);
    }
  }
};
} // namespace
 
static std::pair<llvm::Value *, llvm::Value *>
getPointerAndSize(CodeGenFunction &CGF, const Expr *E) {
  const auto *OASE = dyn_cast<OMPArrayShapingExpr>(E);
  llvm::Value *Addr;
  if (OASE) {
    const Expr *Base = OASE->getBase();
    Addr = CGF.EmitScalarExpr(Base);
  } else {
    Addr = CGF.EmitLValue(E).getPointer(CGF);
  }
  llvm::Value *SizeVal;
  QualType Ty = E->getType();
  if (OASE) {
    SizeVal = CGF.getTypeSize(OASE->getBase()->getType()->getPointeeType());
    for (const Expr *SE : OASE->getDimensions()) {
      llvm::Value *Sz = CGF.EmitScalarExpr(SE);
      Sz = CGF.EmitScalarConversion(
          Sz, SE->getType(), CGF.getContext().getSizeType(), SE->getExprLoc());
      SizeVal = CGF.Builder.CreateNUWMul(SizeVal, Sz);
    }
  } else if (const auto *ASE =
                 dyn_cast<OMPArraySectionExpr>(E->IgnoreParenImpCasts())) {
    LValue UpAddrLVal =
        CGF.EmitOMPArraySectionExpr(ASE, /*IsLowerBound=*/false);
    Address UpAddrAddress = UpAddrLVal.getAddress(CGF);
    llvm::Value *UpAddr = CGF.Builder.CreateConstGEP1_32(
        UpAddrAddress.getElementType(), UpAddrAddress.getPointer(), /*Idx0=*/1);
    llvm::Value *LowIntPtr = CGF.Builder.CreatePtrToInt(Addr, CGF.SizeTy);
    llvm::Value *UpIntPtr = CGF.Builder.CreatePtrToInt(UpAddr, CGF.SizeTy);
    SizeVal = CGF.Builder.CreateNUWSub(UpIntPtr, LowIntPtr);
  } else {
    SizeVal = CGF.getTypeSize(Ty);
  }
  return std::make_pair(Addr, SizeVal);
}
 
/// Builds kmp_depend_info, if it is not built yet, and builds flags type.
static void getKmpAffinityType(ASTContext &C, QualType &KmpTaskAffinityInfoTy) {
  QualType FlagsTy = C.getIntTypeForBitwidth(32, /*Signed=*/false);
  if (KmpTaskAffinityInfoTy.isNull()) {
    RecordDecl *KmpAffinityInfoRD =
        C.buildImplicitRecord("kmp_task_affinity_info_t");
    KmpAffinityInfoRD->startDefinition();
    addFieldToRecordDecl(C, KmpAffinityInfoRD, C.getIntPtrType());
    addFieldToRecordDecl(C, KmpAffinityInfoRD, C.getSizeType());
    addFieldToRecordDecl(C, KmpAffinityInfoRD, FlagsTy);
    KmpAffinityInfoRD->completeDefinition();
    KmpTaskAffinityInfoTy = C.getRecordType(KmpAffinityInfoRD);
  }
}
 
CGOpenMPRuntime::TaskResultTy
CGOpenMPRuntime::emitTaskInit(CodeGenFunction &CGF, SourceLocation Loc,
                              const OMPExecutableDirective &D,
                              llvm::Function *TaskFunction, QualType SharedsTy,
                              Address Shareds, const OMPTaskDataTy &Data) {
  ASTContext &C = CGM.getContext();
  llvm::SmallVector<PrivateDataTy, 4> Privates;
  // Aggregate privates and sort them by the alignment.
  const auto *I = Data.PrivateCopies.begin();
  for (const Expr *E : Data.PrivateVars) {
    const auto *VD = cast<VarDecl>(cast<DeclRefExpr>(E)->getDecl());
    Privates.emplace_back(
        C.getDeclAlign(VD),
        PrivateHelpersTy(E, VD, cast<VarDecl>(cast<DeclRefExpr>(*I)->getDecl()),
                         /*PrivateElemInit=*/nullptr));
    ++I;
  }
  I = Data.FirstprivateCopies.begin();
  const auto *IElemInitRef = Data.FirstprivateInits.begin();
  for (const Expr *E : Data.FirstprivateVars) {
    const auto *VD = cast<VarDecl>(cast<DeclRefExpr>(E)->getDecl());
    Privates.emplace_back(
        C.getDeclAlign(VD),
        PrivateHelpersTy(
            E, VD, cast<VarDecl>(cast<DeclRefExpr>(*I)->getDecl()),
            cast<VarDecl>(cast<DeclRefExpr>(*IElemInitRef)->getDecl())));
    ++I;
    ++IElemInitRef;
  }
  I = Data.LastprivateCopies.begin();
  for (const Expr *E : Data.LastprivateVars) {
    const auto *VD = cast<VarDecl>(cast<DeclRefExpr>(E)->getDecl());
    Privates.emplace_back(
        C.getDeclAlign(VD),
        PrivateHelpersTy(E, VD, cast<VarDecl>(cast<DeclRefExpr>(*I)->getDecl()),
                         /*PrivateElemInit=*/nullptr));
    ++I;
  }
  for (const VarDecl *VD : Data.PrivateLocals) {
    if (isAllocatableDecl(VD))
      Privates.emplace_back(CGM.getPointerAlign(), PrivateHelpersTy(VD));
    else
      Privates.emplace_back(C.getDeclAlign(VD), PrivateHelpersTy(VD));
  }
  llvm::stable_sort(Privates,
                    [](const PrivateDataTy &L, const PrivateDataTy &R) {
                      return L.first > R.first;
                    });
  QualType KmpInt32Ty = C.getIntTypeForBitwidth(/*DestWidth=*/32, /*Signed=*/1);
  // Build type kmp_routine_entry_t (if not built yet).
  emitKmpRoutineEntryT(KmpInt32Ty);
  // Build type kmp_task_t (if not built yet).
  if (isOpenMPTaskLoopDirective(D.getDirectiveKind())) {
    if (SavedKmpTaskloopTQTy.isNull()) {
      SavedKmpTaskloopTQTy = C.getRecordType(createKmpTaskTRecordDecl(
          CGM, D.getDirectiveKind(), KmpInt32Ty, KmpRoutineEntryPtrQTy));
    }
    KmpTaskTQTy = SavedKmpTaskloopTQTy;
  } else {
    assert((D.getDirectiveKind() == OMPD_task ||
            isOpenMPTargetExecutionDirective(D.getDirectiveKind()) ||
            isOpenMPTargetDataManagementDirective(D.getDirectiveKind())) &&
           "Expected taskloop, task or target directive");
    if (SavedKmpTaskTQTy.isNull()) {
      SavedKmpTaskTQTy = C.getRecordType(createKmpTaskTRecordDecl(
          CGM, D.getDirectiveKind(), KmpInt32Ty, KmpRoutineEntryPtrQTy));
    }
    KmpTaskTQTy = SavedKmpTaskTQTy;
  }
  const auto *KmpTaskTQTyRD = cast<RecordDecl>(KmpTaskTQTy->getAsTagDecl());
  // Build particular struct kmp_task_t for the given task.
  const RecordDecl *KmpTaskTWithPrivatesQTyRD =
      createKmpTaskTWithPrivatesRecordDecl(CGM, KmpTaskTQTy, Privates);
  QualType KmpTaskTWithPrivatesQTy = C.getRecordType(KmpTaskTWithPrivatesQTyRD);
  QualType KmpTaskTWithPrivatesPtrQTy =
      C.getPointerType(KmpTaskTWithPrivatesQTy);
  llvm::Type *KmpTaskTWithPrivatesTy = CGF.ConvertType(KmpTaskTWithPrivatesQTy);
  llvm::Type *KmpTaskTWithPrivatesPtrTy =
      KmpTaskTWithPrivatesTy->getPointerTo();
  llvm::Value *KmpTaskTWithPrivatesTySize =
      CGF.getTypeSize(KmpTaskTWithPrivatesQTy);
  QualType SharedsPtrTy = C.getPointerType(SharedsTy);
 
  // Emit initial values for private copies (if any).
  llvm::Value *TaskPrivatesMap = nullptr;
  llvm::Type *TaskPrivatesMapTy =
      std::next(TaskFunction->arg_begin(), 3)->getType();
  if (!Privates.empty()) {
    auto FI = std::next(KmpTaskTWithPrivatesQTyRD->field_begin());
    TaskPrivatesMap =
        emitTaskPrivateMappingFunction(CGM, Loc, Data, FI->getType(), Privates);
    TaskPrivatesMap = CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
        TaskPrivatesMap, TaskPrivatesMapTy);
  } else {
    TaskPrivatesMap = llvm::ConstantPointerNull::get(
        cast<llvm::PointerType>(TaskPrivatesMapTy));
  }
  // Build a proxy function kmp_int32 .omp_task_entry.(kmp_int32 gtid,
  // kmp_task_t *tt);
  llvm::Function *TaskEntry = emitProxyTaskFunction(
      CGM, Loc, D.getDirectiveKind(), KmpInt32Ty, KmpTaskTWithPrivatesPtrQTy,
      KmpTaskTWithPrivatesQTy, KmpTaskTQTy, SharedsPtrTy, TaskFunction,
      TaskPrivatesMap);
 
  // Build call 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);
  // Task flags. Format is taken from
  // https://github.com/llvm/llvm-project/blob/main/openmp/runtime/src/kmp.h,
  // description of kmp_tasking_flags struct.
  enum {
    TiedFlag = 0x1,
    FinalFlag = 0x2,
    DestructorsFlag = 0x8,
    PriorityFlag = 0x20,
    DetachableFlag = 0x40,
  };
  unsigned Flags = Data.Tied ? TiedFlag : 0;
  bool NeedsCleanup = false;
  if (!Privates.empty()) {
    NeedsCleanup =
        checkDestructorsRequired(KmpTaskTWithPrivatesQTyRD, Privates);
    if (NeedsCleanup)
      Flags = Flags | DestructorsFlag;
  }
  if (Data.Priority.getInt())
    Flags = Flags | PriorityFlag;
  if (D.hasClausesOfKind<OMPDetachClause>())
    Flags = Flags | DetachableFlag;
  llvm::Value *TaskFlags =
      Data.Final.getPointer()
          ? CGF.Builder.CreateSelect(Data.Final.getPointer(),
                                     CGF.Builder.getInt32(FinalFlag),
                                     CGF.Builder.getInt32(/*C=*/0))
          : CGF.Builder.getInt32(Data.Final.getInt() ? FinalFlag : 0);
  TaskFlags = CGF.Builder.CreateOr(TaskFlags, CGF.Builder.getInt32(Flags));
  llvm::Value *SharedsSize = CGM.getSize(C.getTypeSizeInChars(SharedsTy));
  SmallVector<llvm::Value *, 8> AllocArgs = {emitUpdateLocation(CGF, Loc),
      getThreadID(CGF, Loc), TaskFlags, KmpTaskTWithPrivatesTySize,
      SharedsSize, CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
          TaskEntry, KmpRoutineEntryPtrTy)};
  llvm::Value *NewTask;
  if (D.hasClausesOfKind<OMPNowaitClause>()) {
    // Check if we have any device clause associated with the directive.
    const Expr *Device = nullptr;
    if (auto *C = D.getSingleClause<OMPDeviceClause>())
      Device = C->getDevice();
    // Emit device ID if any otherwise use default value.
    llvm::Value *DeviceID;
    if (Device)
      DeviceID = CGF.Builder.CreateIntCast(CGF.EmitScalarExpr(Device),
                                           CGF.Int64Ty, /*isSigned=*/true);
    else
      DeviceID = CGF.Builder.getInt64(OMP_DEVICEID_UNDEF);
    AllocArgs.push_back(DeviceID);
    NewTask = CGF.EmitRuntimeCall(
        OMPBuilder.getOrCreateRuntimeFunction(
            CGM.getModule(), OMPRTL___kmpc_omp_target_task_alloc),
        AllocArgs);
  } else {
    NewTask =
        CGF.EmitRuntimeCall(OMPBuilder.getOrCreateRuntimeFunction(
                                CGM.getModule(), OMPRTL___kmpc_omp_task_alloc),
                            AllocArgs);
  }
  // Emit detach clause initialization.
  // evt = (typeof(evt))__kmpc_task_allow_completion_event(loc, tid,
  // task_descriptor);
  if (const auto *DC = D.getSingleClause<OMPDetachClause>()) {
    const Expr *Evt = DC->getEventHandler()->IgnoreParenImpCasts();
    LValue EvtLVal = CGF.EmitLValue(Evt);
 
    // Build kmp_event_t *__kmpc_task_allow_completion_event(ident_t *loc_ref,
    // int gtid, kmp_task_t *task);
    llvm::Value *Loc = emitUpdateLocation(CGF, DC->getBeginLoc());
    llvm::Value *Tid = getThreadID(CGF, DC->getBeginLoc());
    Tid = CGF.Builder.CreateIntCast(Tid, CGF.IntTy, /*isSigned=*/false);
    llvm::Value *EvtVal = CGF.EmitRuntimeCall(
        OMPBuilder.getOrCreateRuntimeFunction(
            CGM.getModule(), OMPRTL___kmpc_task_allow_completion_event),
        {Loc, Tid, NewTask});
    EvtVal = CGF.EmitScalarConversion(EvtVal, C.VoidPtrTy, Evt->getType(),
                                      Evt->getExprLoc());
    CGF.EmitStoreOfScalar(EvtVal, EvtLVal);
  }
  // Process affinity clauses.
  if (D.hasClausesOfKind<OMPAffinityClause>()) {
    // Process list of affinity data.
    ASTContext &C = CGM.getContext();
    Address AffinitiesArray = Address::invalid();
    // Calculate number of elements to form the array of affinity data.
    llvm::Value *NumOfElements = nullptr;
    unsigned NumAffinities = 0;
    for (const auto *C : D.getClausesOfKind<OMPAffinityClause>()) {
      if (const Expr *Modifier = C->getModifier()) {
        const auto *IE = cast<OMPIteratorExpr>(Modifier->IgnoreParenImpCasts());
        for (unsigned I = 0, E = IE->numOfIterators(); I < E; ++I) {
          llvm::Value *Sz = CGF.EmitScalarExpr(IE->getHelper(I).Upper);
          Sz = CGF.Builder.CreateIntCast(Sz, CGF.SizeTy, /*isSigned=*/false);
          NumOfElements =
              NumOfElements ? CGF.Builder.CreateNUWMul(NumOfElements, Sz) : Sz;
        }
      } else {
        NumAffinities += C->varlist_size();
      }
    }
    getKmpAffinityType(CGM.getContext(), KmpTaskAffinityInfoTy);
    // Fields ids in kmp_task_affinity_info record.
    enum RTLAffinityInfoFieldsTy { BaseAddr, Len, Flags };
 
    QualType KmpTaskAffinityInfoArrayTy;
    if (NumOfElements) {
      NumOfElements = CGF.Builder.CreateNUWAdd(
          llvm::ConstantInt::get(CGF.SizeTy, NumAffinities), NumOfElements);
      auto *OVE = new (C) OpaqueValueExpr(
          Loc,
          C.getIntTypeForBitwidth(C.getTypeSize(C.getSizeType()), /*Signed=*/0),
          VK_PRValue);
      CodeGenFunction::OpaqueValueMapping OpaqueMap(CGF, OVE,
                                                    RValue::get(NumOfElements));
      KmpTaskAffinityInfoArrayTy =
          C.getVariableArrayType(KmpTaskAffinityInfoTy, OVE, ArrayType::Normal,
                                 /*IndexTypeQuals=*/0, SourceRange(Loc, Loc));
      // Properly emit variable-sized array.
      auto *PD = ImplicitParamDecl::Create(C, KmpTaskAffinityInfoArrayTy,
                                           ImplicitParamDecl::Other);
      CGF.EmitVarDecl(*PD);
      AffinitiesArray = CGF.GetAddrOfLocalVar(PD);
      NumOfElements = CGF.Builder.CreateIntCast(NumOfElements, CGF.Int32Ty,
                                                /*isSigned=*/false);
    } else {
      KmpTaskAffinityInfoArrayTy = C.getConstantArrayType(
          KmpTaskAffinityInfoTy,
          llvm::APInt(C.getTypeSize(C.getSizeType()), NumAffinities), nullptr,
          ArrayType::Normal, /*IndexTypeQuals=*/0);
      AffinitiesArray =
          CGF.CreateMemTemp(KmpTaskAffinityInfoArrayTy, ".affs.arr.addr");
      AffinitiesArray = CGF.Builder.CreateConstArrayGEP(AffinitiesArray, 0);
      NumOfElements = llvm::ConstantInt::get(CGM.Int32Ty, NumAffinities,
                                             /*isSigned=*/false);
    }
 
    const auto *KmpAffinityInfoRD = KmpTaskAffinityInfoTy->getAsRecordDecl();
    // Fill array by elements without iterators.
    unsigned Pos = 0;
    bool HasIterator = false;
    for (const auto *C : D.getClausesOfKind<OMPAffinityClause>()) {
      if (C->getModifier()) {
        HasIterator = true;
        continue;
      }
      for (const Expr *E : C->varlists()) {
        llvm::Value *Addr;
        llvm::Value *Size;
        std::tie(Addr, Size) = getPointerAndSize(CGF, E);
        LValue Base =
            CGF.MakeAddrLValue(CGF.Builder.CreateConstGEP(AffinitiesArray, Pos),
                               KmpTaskAffinityInfoTy);
        // affs[i].base_addr = &<Affinities[i].second>;
        LValue BaseAddrLVal = CGF.EmitLValueForField(
            Base, *std::next(KmpAffinityInfoRD->field_begin(), BaseAddr));
        CGF.EmitStoreOfScalar(CGF.Builder.CreatePtrToInt(Addr, CGF.IntPtrTy),
                              BaseAddrLVal);
        // affs[i].len = sizeof(<Affinities[i].second>);
        LValue LenLVal = CGF.EmitLValueForField(
            Base, *std::next(KmpAffinityInfoRD->field_begin(), Len));
        CGF.EmitStoreOfScalar(Size, LenLVal);
        ++Pos;
      }
    }
    LValue PosLVal;
    if (HasIterator) {
      PosLVal = CGF.MakeAddrLValue(
          CGF.CreateMemTemp(C.getSizeType(), "affs.counter.addr"),
          C.getSizeType());
      CGF.EmitStoreOfScalar(llvm::ConstantInt::get(CGF.SizeTy, Pos), PosLVal);
    }
    // Process elements with iterators.
    for (const auto *C : D.getClausesOfKind<OMPAffinityClause>()) {
      const Expr *Modifier = C->getModifier();
      if (!Modifier)
        continue;
      OMPIteratorGeneratorScope IteratorScope(
          CGF, cast_or_null<OMPIteratorExpr>(Modifier->IgnoreParenImpCasts()));
      for (const Expr *E : C->varlists()) {
        llvm::Value *Addr;
        llvm::Value *Size;
        std::tie(Addr, Size) = getPointerAndSize(CGF, E);
        llvm::Value *Idx = CGF.EmitLoadOfScalar(PosLVal, E->getExprLoc());
        LValue Base = CGF.MakeAddrLValue(
            CGF.Builder.CreateGEP(AffinitiesArray, Idx), KmpTaskAffinityInfoTy);
        // affs[i].base_addr = &<Affinities[i].second>;
        LValue BaseAddrLVal = CGF.EmitLValueForField(
            Base, *std::next(KmpAffinityInfoRD->field_begin(), BaseAddr));
        CGF.EmitStoreOfScalar(CGF.Builder.CreatePtrToInt(Addr, CGF.IntPtrTy),
                              BaseAddrLVal);
        // affs[i].len = sizeof(<Affinities[i].second>);
        LValue LenLVal = CGF.EmitLValueForField(
            Base, *std::next(KmpAffinityInfoRD->field_begin(), Len));
        CGF.EmitStoreOfScalar(Size, LenLVal);
        Idx = CGF.Builder.CreateNUWAdd(
            Idx, llvm::ConstantInt::get(Idx->getType(), 1));
        CGF.EmitStoreOfScalar(Idx, PosLVal);
      }
    }
    // Call to kmp_int32 __kmpc_omp_reg_task_with_affinity(ident_t *loc_ref,
    // kmp_int32 gtid, kmp_task_t *new_task, kmp_int32
    // naffins, kmp_task_affinity_info_t *affin_list);
    llvm::Value *LocRef = emitUpdateLocation(CGF, Loc);
    llvm::Value *GTid = getThreadID(CGF, Loc);
    llvm::Value *AffinListPtr = CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
        AffinitiesArray.getPointer(), CGM.VoidPtrTy);
    // FIXME: Emit the function and ignore its result for now unless the
    // runtime function is properly implemented.
    (void)CGF.EmitRuntimeCall(
        OMPBuilder.getOrCreateRuntimeFunction(
            CGM.getModule(), OMPRTL___kmpc_omp_reg_task_with_affinity),
        {LocRef, GTid, NewTask, NumOfElements, AffinListPtr});
  }
  llvm::Value *NewTaskNewTaskTTy =
      CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
          NewTask, KmpTaskTWithPrivatesPtrTy);
  LValue Base = CGF.MakeNaturalAlignAddrLValue(NewTaskNewTaskTTy,
                                               KmpTaskTWithPrivatesQTy);
  LValue TDBase =
      CGF.EmitLValueForField(Base, *KmpTaskTWithPrivatesQTyRD->field_begin());
  // Fill the data in the resulting kmp_task_t record.
  // Copy shareds if there are any.
  Address KmpTaskSharedsPtr = Address::invalid();
  if (!SharedsTy->getAsStructureType()->getDecl()->field_empty()) {
    KmpTaskSharedsPtr = Address(
        CGF.EmitLoadOfScalar(
            CGF.EmitLValueForField(
                TDBase,
                *std::next(KmpTaskTQTyRD->field_begin(), KmpTaskTShareds)),
            Loc),
        CGF.Int8Ty, CGM.getNaturalTypeAlignment(SharedsTy));
    LValue Dest = CGF.MakeAddrLValue(KmpTaskSharedsPtr, SharedsTy);
    LValue Src = CGF.MakeAddrLValue(Shareds, SharedsTy);
    CGF.EmitAggregateCopy(Dest, Src, SharedsTy, AggValueSlot::DoesNotOverlap);
  }
  // Emit initial values for private copies (if any).
  TaskResultTy Result;
  if (!Privates.empty()) {
    emitPrivatesInit(CGF, D, KmpTaskSharedsPtr, Base, KmpTaskTWithPrivatesQTyRD,
                     SharedsTy, SharedsPtrTy, Data, Privates,
                     /*ForDup=*/false);
    if (isOpenMPTaskLoopDirective(D.getDirectiveKind()) &&
        (!Data.LastprivateVars.empty() || checkInitIsRequired(CGF, Privates))) {
      Result.TaskDupFn = emitTaskDupFunction(
          CGM, Loc, D, KmpTaskTWithPrivatesPtrQTy, KmpTaskTWithPrivatesQTyRD,
          KmpTaskTQTyRD, SharedsTy, SharedsPtrTy, Data, Privates,
          /*WithLastIter=*/!Data.LastprivateVars.empty());
    }
  }
  // Fields of union "kmp_cmplrdata_t" for destructors and priority.
  enum { Priority = 0, Destructors = 1 };
  // Provide pointer to function with destructors for privates.
  auto FI = std::next(KmpTaskTQTyRD->field_begin(), Data1);
  const RecordDecl *KmpCmplrdataUD =
      (*FI)->getType()->getAsUnionType()->getDecl();
  if (NeedsCleanup) {
    llvm::Value *DestructorFn = emitDestructorsFunction(
        CGM, Loc, KmpInt32Ty, KmpTaskTWithPrivatesPtrQTy,
        KmpTaskTWithPrivatesQTy);
    LValue Data1LV = CGF.EmitLValueForField(TDBase, *FI);
    LValue DestructorsLV = CGF.EmitLValueForField(
        Data1LV, *std::next(KmpCmplrdataUD->field_begin(), Destructors));
    CGF.EmitStoreOfScalar(CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
                              DestructorFn, KmpRoutineEntryPtrTy),
                          DestructorsLV);
  }
  // Set priority.
  if (Data.Priority.getInt()) {
    LValue Data2LV = CGF.EmitLValueForField(
        TDBase, *std::next(KmpTaskTQTyRD->field_begin(), Data2));
    LValue PriorityLV = CGF.EmitLValueForField(
        Data2LV, *std::next(KmpCmplrdataUD->field_begin(), Priority));
    CGF.EmitStoreOfScalar(Data.Priority.getPointer(), PriorityLV);
  }
  Result.NewTask = NewTask;
  Result.TaskEntry = TaskEntry;
  Result.NewTaskNewTaskTTy = NewTaskNewTaskTTy;
  Result.TDBase = TDBase;
  Result.KmpTaskTQTyRD = KmpTaskTQTyRD;
  return Result;
}
 
/// Translates internal dependency kind into the runtime kind.
static RTLDependenceKindTy translateDependencyKind(OpenMPDependClauseKind K) {
  RTLDependenceKindTy DepKind;
  switch (K) {
  case OMPC_DEPEND_in:
    DepKind = RTLDependenceKindTy::DepIn;
    break;
  // Out and InOut dependencies must use the same code.
  case OMPC_DEPEND_out:
  case OMPC_DEPEND_inout:
    DepKind = RTLDependenceKindTy::DepInOut;
    break;
  case OMPC_DEPEND_mutexinoutset:
    DepKind = RTLDependenceKindTy::DepMutexInOutSet;
    break;
  case OMPC_DEPEND_inoutset:
    DepKind = RTLDependenceKindTy::DepInOutSet;
    break;
  case OMPC_DEPEND_outallmemory:
    DepKind = RTLDependenceKindTy::DepOmpAllMem;
    break;
  case OMPC_DEPEND_source:
  case OMPC_DEPEND_sink:
  case OMPC_DEPEND_depobj:
  case OMPC_DEPEND_inoutallmemory:
  case OMPC_DEPEND_unknown:
    llvm_unreachable("Unknown task dependence type");
  }
  return DepKind;
}
 
/// Builds kmp_depend_info, if it is not built yet, and builds flags type.
static void getDependTypes(ASTContext &C, QualType &KmpDependInfoTy,
                           QualType &FlagsTy) {
  FlagsTy = C.getIntTypeForBitwidth(C.getTypeSize(C.BoolTy), /*Signed=*/false);
  if (KmpDependInfoTy.isNull()) {
    RecordDecl *KmpDependInfoRD = C.buildImplicitRecord("kmp_depend_info");
    KmpDependInfoRD->startDefinition();
    addFieldToRecordDecl(C, KmpDependInfoRD, C.getIntPtrType());
    addFieldToRecordDecl(C, KmpDependInfoRD, C.getSizeType());
    addFieldToRecordDecl(C, KmpDependInfoRD, FlagsTy);
    KmpDependInfoRD->completeDefinition();
    KmpDependInfoTy = C.getRecordType(KmpDependInfoRD);
  }
}
 
std::pair<llvm::Value *, LValue>
CGOpenMPRuntime::getDepobjElements(CodeGenFunction &CGF, LValue DepobjLVal,
                                   SourceLocation Loc) {
  ASTContext &C = CGM.getContext();
  QualType FlagsTy;
  getDependTypes(C, KmpDependInfoTy, FlagsTy);
  RecordDecl *KmpDependInfoRD =
      cast<RecordDecl>(KmpDependInfoTy->getAsTagDecl());
  QualType KmpDependInfoPtrTy = C.getPointerType(KmpDependInfoTy);
  LValue Base = CGF.EmitLoadOfPointerLValue(
      CGF.Builder.CreateElementBitCast(
          DepobjLVal.getAddress(CGF),
          CGF.ConvertTypeForMem(KmpDependInfoPtrTy)),
      KmpDependInfoPtrTy->castAs<PointerType>());
  Address DepObjAddr = CGF.Builder.CreateGEP(
      Base.getAddress(CGF),
      llvm::ConstantInt::get(CGF.IntPtrTy, -1, /*isSigned=*/true));
  LValue NumDepsBase = CGF.MakeAddrLValue(
      DepObjAddr, KmpDependInfoTy, Base.getBaseInfo(), Base.getTBAAInfo());
  // NumDeps = deps[i].base_addr;
  LValue BaseAddrLVal = CGF.EmitLValueForField(
      NumDepsBase,
      *std::next(KmpDependInfoRD->field_begin(),
                 static_cast<unsigned int>(RTLDependInfoFields::BaseAddr)));
  llvm::Value *NumDeps = CGF.EmitLoadOfScalar(BaseAddrLVal, Loc);
  return std::make_pair(NumDeps, Base);
}
 
static void emitDependData(CodeGenFunction &CGF, QualType &KmpDependInfoTy,
                           llvm::PointerUnion<unsigned *, LValue *> Pos,
                           const OMPTaskDataTy::DependData &Data,
                           Address DependenciesArray) {
  CodeGenModule &CGM = CGF.CGM;
  ASTContext &C = CGM.getContext();
  QualType FlagsTy;
  getDependTypes(C, KmpDependInfoTy, FlagsTy);
  RecordDecl *KmpDependInfoRD =
      cast<RecordDecl>(KmpDependInfoTy->getAsTagDecl());
  llvm::Type *LLVMFlagsTy = CGF.ConvertTypeForMem(FlagsTy);
 
  OMPIteratorGeneratorScope IteratorScope(
      CGF, cast_or_null<OMPIteratorExpr>(
               Data.IteratorExpr ? Data.IteratorExpr->IgnoreParenImpCasts()
                                 : nullptr));
  for (const Expr *E : Data.DepExprs) {
    llvm::Value *Addr;
    llvm::Value *Size;
 
    // The expression will be a nullptr in the 'omp_all_memory' case.
    if (E) {
      std::tie(Addr, Size) = getPointerAndSize(CGF, E);
      Addr = CGF.Builder.CreatePtrToInt(Addr, CGF.IntPtrTy);
    } else {
      Addr = llvm::ConstantInt::get(CGF.IntPtrTy, 0);
      Size = llvm::ConstantInt::get(CGF.SizeTy, 0);
    }
    LValue Base;
    if (unsigned *P = Pos.dyn_cast<unsigned *>()) {
      Base = CGF.MakeAddrLValue(
          CGF.Builder.CreateConstGEP(DependenciesArray, *P), KmpDependInfoTy);
    } else {
      assert(E && "Expected a non-null expression");
      LValue &PosLVal = *Pos.get<LValue *>();
      llvm::Value *Idx = CGF.EmitLoadOfScalar(PosLVal, E->getExprLoc());
      Base = CGF.MakeAddrLValue(
          CGF.Builder.CreateGEP(DependenciesArray, Idx), KmpDependInfoTy);
    }
    // deps[i].base_addr = &<Dependencies[i].second>;
    LValue BaseAddrLVal = CGF.EmitLValueForField(
        Base,
        *std::next(KmpDependInfoRD->field_begin(),
                   static_cast<unsigned int>(RTLDependInfoFields::BaseAddr)));
    CGF.EmitStoreOfScalar(Addr, BaseAddrLVal);
    // deps[i].len = sizeof(<Dependencies[i].second>);
    LValue LenLVal = CGF.EmitLValueForField(
        Base, *std::next(KmpDependInfoRD->field_begin(),
                         static_cast<unsigned int>(RTLDependInfoFields::Len)));
    CGF.EmitStoreOfScalar(Size, LenLVal);
    // deps[i].flags = <Dependencies[i].first>;
    RTLDependenceKindTy DepKind = translateDependencyKind(Data.DepKind);
    LValue FlagsLVal = CGF.EmitLValueForField(
        Base,
        *std::next(KmpDependInfoRD->field_begin(),
                   static_cast<unsigned int>(RTLDependInfoFields::Flags)));
    CGF.EmitStoreOfScalar(
        llvm::ConstantInt::get(LLVMFlagsTy, static_cast<unsigned int>(DepKind)),
        FlagsLVal);
    if (unsigned *P = Pos.dyn_cast<unsigned *>()) {
      ++(*P);
    } else {
      LValue &PosLVal = *Pos.get<LValue *>();
      llvm::Value *Idx = CGF.EmitLoadOfScalar(PosLVal, E->getExprLoc());
      Idx = CGF.Builder.CreateNUWAdd(Idx,
                                     llvm::ConstantInt::get(Idx->getType(), 1));
      CGF.EmitStoreOfScalar(Idx, PosLVal);
    }
  }
}
 
SmallVector<llvm::Value *, 4> CGOpenMPRuntime::emitDepobjElementsSizes(
    CodeGenFunction &CGF, QualType &KmpDependInfoTy,
    const OMPTaskDataTy::DependData &Data) {
  assert(Data.DepKind == OMPC_DEPEND_depobj &&
         "Expected depobj dependency kind.");
  SmallVector<llvm::Value *, 4> Sizes;
  SmallVector<LValue, 4> SizeLVals;
  ASTContext &C = CGF.getContext();
  {
    OMPIteratorGeneratorScope IteratorScope(
        CGF, cast_or_null<OMPIteratorExpr>(
                 Data.IteratorExpr ? Data.IteratorExpr->IgnoreParenImpCasts()
                                   : nullptr));
    for (const Expr *E : Data.DepExprs) {
      llvm::Value *NumDeps;
      LValue Base;
      LValue DepobjLVal = CGF.EmitLValue(E->IgnoreParenImpCasts());
      std::tie(NumDeps, Base) =
          getDepobjElements(CGF, DepobjLVal, E->getExprLoc());
      LValue NumLVal = CGF.MakeAddrLValue(
          CGF.CreateMemTemp(C.getUIntPtrType(), "depobj.size.addr"),
          C.getUIntPtrType());
      CGF.Builder.CreateStore(llvm::ConstantInt::get(CGF.IntPtrTy, 0),
                              NumLVal.getAddress(CGF));
      llvm::Value *PrevVal = CGF.EmitLoadOfScalar(NumLVal, E->getExprLoc());
      llvm::Value *Add = CGF.Builder.CreateNUWAdd(PrevVal, NumDeps);
      CGF.EmitStoreOfScalar(Add, NumLVal);
      SizeLVals.push_back(NumLVal);
    }
  }
  for (unsigned I = 0, E = SizeLVals.size(); I < E; ++I) {
    llvm::Value *Size =
        CGF.EmitLoadOfScalar(SizeLVals[I], Data.DepExprs[I]->getExprLoc());
    Sizes.push_back(Size);
  }
  return Sizes;
}
 
void CGOpenMPRuntime::emitDepobjElements(CodeGenFunction &CGF,
                                         QualType &KmpDependInfoTy,
                                         LValue PosLVal,
                                         const OMPTaskDataTy::DependData &Data,
                                         Address DependenciesArray) {
  assert(Data.DepKind == OMPC_DEPEND_depobj &&
         "Expected depobj dependency kind.");
  llvm::Value *ElSize = CGF.getTypeSize(KmpDependInfoTy);
  {
    OMPIteratorGeneratorScope IteratorScope(
        CGF, cast_or_null<OMPIteratorExpr>(
                 Data.IteratorExpr ? Data.IteratorExpr->IgnoreParenImpCasts()
                                   : nullptr));
    for (unsigned I = 0, End = Data.DepExprs.size(); I < End; ++I) {
      const Expr *E = Data.DepExprs[I];
      llvm::Value *NumDeps;
      LValue Base;
      LValue DepobjLVal = CGF.EmitLValue(E->IgnoreParenImpCasts());
      std::tie(NumDeps, Base) =
          getDepobjElements(CGF, DepobjLVal, E->getExprLoc());
 
      // memcopy dependency data.
      llvm::Value *Size = CGF.Builder.CreateNUWMul(
          ElSize,
          CGF.Builder.CreateIntCast(NumDeps, CGF.SizeTy, /*isSigned=*/false));
      llvm::Value *Pos = CGF.EmitLoadOfScalar(PosLVal, E->getExprLoc());
      Address DepAddr = CGF.Builder.CreateGEP(DependenciesArray, Pos);
      CGF.Builder.CreateMemCpy(DepAddr, Base.getAddress(CGF), Size);
 
      // Increase pos.
      // pos += size;
      llvm::Value *Add = CGF.Builder.CreateNUWAdd(Pos, NumDeps);
      CGF.EmitStoreOfScalar(Add, PosLVal);
    }
  }
}
 
std::pair<llvm::Value *, Address> CGOpenMPRuntime::emitDependClause(
    CodeGenFunction &CGF, ArrayRef<OMPTaskDataTy::DependData> Dependencies,
    SourceLocation Loc) {
  if (llvm::all_of(Dependencies, [](const OMPTaskDataTy::DependData &D) {
        return D.DepExprs.empty();
      }))
    return std::make_pair(nullptr, Address::invalid());
  // Process list of dependencies.
  ASTContext &C = CGM.getContext();
  Address DependenciesArray = Address::invalid();
  llvm::Value *NumOfElements = nullptr;
  unsigned NumDependencies = std::accumulate(
      Dependencies.begin(), Dependencies.end(), 0,
      [](unsigned V, const OMPTaskDataTy::DependData &D) {
        return D.DepKind == OMPC_DEPEND_depobj
                   ? V
                   : (V + (D.IteratorExpr ? 0 : D.DepExprs.size()));
      });
  QualType FlagsTy;
  getDependTypes(C, KmpDependInfoTy, FlagsTy);
  bool HasDepobjDeps = false;
  bool HasRegularWithIterators = false;
  llvm::Value *NumOfDepobjElements = llvm::ConstantInt::get(CGF.IntPtrTy, 0);
  llvm::Value *NumOfRegularWithIterators =
      llvm::ConstantInt::get(CGF.IntPtrTy, 0);
  // Calculate number of depobj dependencies and regular deps with the
  // iterators.
  for (const OMPTaskDataTy::DependData &D : Dependencies) {
    if (D.DepKind == OMPC_DEPEND_depobj) {
      SmallVector<llvm::Value *, 4> Sizes =
          emitDepobjElementsSizes(CGF, KmpDependInfoTy, D);
      for (llvm::Value *Size : Sizes) {
        NumOfDepobjElements =
            CGF.Builder.CreateNUWAdd(NumOfDepobjElements, Size);
      }
      HasDepobjDeps = true;
      continue;
    }
    // Include number of iterations, if any.
 
    if (const auto *IE = cast_or_null<OMPIteratorExpr>(D.IteratorExpr)) {
      for (unsigned I = 0, E = IE->numOfIterators(); I < E; ++I) {
        llvm::Value *Sz = CGF.EmitScalarExpr(IE->getHelper(I).Upper);
        Sz = CGF.Builder.CreateIntCast(Sz, CGF.IntPtrTy, /*isSigned=*/false);
        llvm::Value *NumClauseDeps = CGF.Builder.CreateNUWMul(
            Sz, llvm::ConstantInt::get(CGF.IntPtrTy, D.DepExprs.size()));
        NumOfRegularWithIterators =
            CGF.Builder.CreateNUWAdd(NumOfRegularWithIterators, NumClauseDeps);
      }
      HasRegularWithIterators = true;
      continue;
    }
  }
 
  QualType KmpDependInfoArrayTy;
  if (HasDepobjDeps || HasRegularWithIterators) {
    NumOfElements = llvm::ConstantInt::get(CGM.IntPtrTy, NumDependencies,
                                           /*isSigned=*/false);
    if (HasDepobjDeps) {
      NumOfElements =
          CGF.Builder.CreateNUWAdd(NumOfDepobjElements, NumOfElements);
    }
    if (HasRegularWithIterators) {
      NumOfElements =
          CGF.Builder.CreateNUWAdd(NumOfRegularWithIterators, NumOfElements);
    }
    auto *OVE = new (C) OpaqueValueExpr(
        Loc, C.getIntTypeForBitwidth(/*DestWidth=*/64, /*Signed=*/0),
        VK_PRValue);
    CodeGenFunction::OpaqueValueMapping OpaqueMap(CGF, OVE,
                                                  RValue::get(NumOfElements));
    KmpDependInfoArrayTy =
        C.getVariableArrayType(KmpDependInfoTy, OVE, ArrayType::Normal,
                               /*IndexTypeQuals=*/0, SourceRange(Loc, Loc));
    // CGF.EmitVariablyModifiedType(KmpDependInfoArrayTy);
    // Properly emit variable-sized array.
    auto *PD = ImplicitParamDecl::Create(C, KmpDependInfoArrayTy,
                                         ImplicitParamDecl::Other);
    CGF.EmitVarDecl(*PD);
    DependenciesArray = CGF.GetAddrOfLocalVar(PD);
    NumOfElements = CGF.Builder.CreateIntCast(NumOfElements, CGF.Int32Ty,
                                              /*isSigned=*/false);
  } else {
    KmpDependInfoArrayTy = C.getConstantArrayType(
        KmpDependInfoTy, llvm::APInt(/*numBits=*/64, NumDependencies), nullptr,
        ArrayType::Normal, /*IndexTypeQuals=*/0);
    DependenciesArray =
        CGF.CreateMemTemp(KmpDependInfoArrayTy, ".dep.arr.addr");
    DependenciesArray = CGF.Builder.CreateConstArrayGEP(DependenciesArray, 0);
    NumOfElements = llvm::ConstantInt::get(CGM.Int32Ty, NumDependencies,
                                           /*isSigned=*/false);
  }
  unsigned Pos = 0;
  for (unsigned I = 0, End = Dependencies.size(); I < End; ++I) {
    if (Dependencies[I].DepKind == OMPC_DEPEND_depobj ||
        Dependencies[I].IteratorExpr)
      continue;
    emitDependData(CGF, KmpDependInfoTy, &Pos, Dependencies[I],
                   DependenciesArray);
  }
  // Copy regular dependencies with iterators.
  LValue PosLVal = CGF.MakeAddrLValue(
      CGF.CreateMemTemp(C.getSizeType(), "dep.counter.addr"), C.getSizeType());
  CGF.EmitStoreOfScalar(llvm::ConstantInt::get(CGF.SizeTy, Pos), PosLVal);
  for (unsigned I = 0, End = Dependencies.size(); I < End; ++I) {
    if (Dependencies[I].DepKind == OMPC_DEPEND_depobj ||
        !Dependencies[I].IteratorExpr)
      continue;
    emitDependData(CGF, KmpDependInfoTy, &PosLVal, Dependencies[I],
                   DependenciesArray);
  }
  // Copy final depobj arrays without iterators.
  if (HasDepobjDeps) {
    for (unsigned I = 0, End = Dependencies.size(); I < End; ++I) {
      if (Dependencies[I].DepKind != OMPC_DEPEND_depobj)
        continue;
      emitDepobjElements(CGF, KmpDependInfoTy, PosLVal, Dependencies[I],
                         DependenciesArray);
    }
  }
  DependenciesArray = CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
      DependenciesArray, CGF.VoidPtrTy, CGF.Int8Ty);
  return std::make_pair(NumOfElements, DependenciesArray);
}