diff --git a/clang/lib/Sema/SemaOpenMP.cpp b/clang/lib/Sema/SemaOpenMP.cpp
--- a/clang/lib/Sema/SemaOpenMP.cpp
+++ b/clang/lib/Sema/SemaOpenMP.cpp
@@ -70,12 +70,15 @@
     const Expr *RefExpr = nullptr;
     DeclRefExpr *PrivateCopy = nullptr;
     SourceLocation ImplicitDSALoc;
+    bool AppliedToPointee = false;
     DSAVarData() = default;
     DSAVarData(OpenMPDirectiveKind DKind, OpenMPClauseKind CKind,
                const Expr *RefExpr, DeclRefExpr *PrivateCopy,
-               SourceLocation ImplicitDSALoc, unsigned Modifier)
+               SourceLocation ImplicitDSALoc, unsigned Modifier,
+               bool AppliedToPointee)
         : DKind(DKind), CKind(CKind), Modifier(Modifier), RefExpr(RefExpr),
-          PrivateCopy(PrivateCopy), ImplicitDSALoc(ImplicitDSALoc) {}
+          PrivateCopy(PrivateCopy), ImplicitDSALoc(ImplicitDSALoc),
+          AppliedToPointee(AppliedToPointee) {}
   };
   using OperatorOffsetTy =
       llvm::SmallVector<std::pair<Expr *, OverloadedOperatorKind>, 4>;
@@ -99,6 +102,9 @@
     /// variable is marked as lastprivate(true) or not (false).
     llvm::PointerIntPair<const Expr *, 1, bool> RefExpr;
     DeclRefExpr *PrivateCopy = nullptr;
+    /// true if the attribute is applied to the pointee, not the variable
+    /// itself.
+    bool AppliedToPointee = false;
   };
   using DeclSAMapTy = llvm::SmallDenseMap<const ValueDecl *, DSAInfo, 8>;
   using UsedRefMapTy = llvm::SmallDenseMap<const ValueDecl *, const Expr *, 8>;
@@ -511,7 +517,8 @@
 
   /// Adds explicit data sharing attribute to the specified declaration.
   void addDSA(const ValueDecl *D, const Expr *E, OpenMPClauseKind A,
-              DeclRefExpr *PrivateCopy = nullptr, unsigned Modifier = 0);
+              DeclRefExpr *PrivateCopy = nullptr, unsigned Modifier = 0,
+              bool AppliedToPointee = false);
 
   /// Adds additional information for the reduction items with the reduction id
   /// represented as an operator.
@@ -563,7 +570,8 @@
   /// match specified \a CPred predicate in any directive which matches \a DPred
   /// predicate.
   const DSAVarData
-  hasDSA(ValueDecl *D, const llvm::function_ref<bool(OpenMPClauseKind)> CPred,
+  hasDSA(ValueDecl *D,
+         const llvm::function_ref<bool(OpenMPClauseKind, bool)> CPred,
          const llvm::function_ref<bool(OpenMPDirectiveKind)> DPred,
          bool FromParent) const;
   /// Checks if the specified variables has data-sharing attributes which
@@ -571,15 +579,16 @@
   /// matches \a DPred predicate.
   const DSAVarData
   hasInnermostDSA(ValueDecl *D,
-                  const llvm::function_ref<bool(OpenMPClauseKind)> CPred,
+                  const llvm::function_ref<bool(OpenMPClauseKind, bool)> CPred,
                   const llvm::function_ref<bool(OpenMPDirectiveKind)> DPred,
                   bool FromParent) const;
   /// Checks if the specified variables has explicit data-sharing
   /// attributes which match specified \a CPred predicate at the specified
   /// OpenMP region.
-  bool hasExplicitDSA(const ValueDecl *D,
-                      const llvm::function_ref<bool(OpenMPClauseKind)> CPred,
-                      unsigned Level, bool NotLastprivate = false) const;
+  bool
+  hasExplicitDSA(const ValueDecl *D,
+                 const llvm::function_ref<bool(OpenMPClauseKind, bool)> CPred,
+                 unsigned Level, bool NotLastprivate = false) const;
 
   /// Returns true if the directive at level \Level matches in the
   /// specified \a DPred predicate.
@@ -1185,6 +1194,7 @@
     DVar.CKind = Data.Attributes;
     DVar.ImplicitDSALoc = Iter->DefaultAttrLoc;
     DVar.Modifier = Data.Modifier;
+    DVar.AppliedToPointee = Data.AppliedToPointee;
     return DVar;
   }
 
@@ -1341,7 +1351,8 @@
 }
 
 void DSAStackTy::addDSA(const ValueDecl *D, const Expr *E, OpenMPClauseKind A,
-                        DeclRefExpr *PrivateCopy, unsigned Modifier) {
+                        DeclRefExpr *PrivateCopy, unsigned Modifier,
+                        bool AppliedToPointee) {
   D = getCanonicalDecl(D);
   if (A == OMPC_threadprivate) {
     DSAInfo &Data = Threadprivates[D];
@@ -1365,12 +1376,14 @@
     Data.Attributes = A;
     Data.RefExpr.setPointerAndInt(E, IsLastprivate);
     Data.PrivateCopy = PrivateCopy;
+    Data.AppliedToPointee = AppliedToPointee;
     if (PrivateCopy) {
       DSAInfo &Data = getTopOfStack().SharingMap[PrivateCopy->getDecl()];
       Data.Modifier = Modifier;
       Data.Attributes = A;
       Data.RefExpr.setPointerAndInt(PrivateCopy, IsLastprivate);
       Data.PrivateCopy = nullptr;
+      Data.AppliedToPointee = AppliedToPointee;
     }
   }
 }
@@ -1480,7 +1493,8 @@
                                        "set.");
     TaskgroupDescriptor = I->TaskgroupReductionRef;
     return DSAVarData(I->Directive, OMPC_reduction, Data.RefExpr.getPointer(),
-                      Data.PrivateCopy, I->DefaultAttrLoc, OMPC_REDUCTION_task);
+                      Data.PrivateCopy, I->DefaultAttrLoc, OMPC_REDUCTION_task,
+                      /*AppliedToPointee=*/false);
   }
   return DSAVarData();
 }
@@ -1506,7 +1520,8 @@
                                        "set.");
     TaskgroupDescriptor = I->TaskgroupReductionRef;
     return DSAVarData(I->Directive, OMPC_reduction, Data.RefExpr.getPointer(),
-                      Data.PrivateCopy, I->DefaultAttrLoc, OMPC_REDUCTION_task);
+                      Data.PrivateCopy, I->DefaultAttrLoc, OMPC_REDUCTION_task,
+                      /*AppliedToPointee=*/false);
   }
   return DSAVarData();
 }
@@ -1675,6 +1690,7 @@
         DVar.ImplicitDSALoc = I->DefaultAttrLoc;
         DVar.DKind = I->Directive;
         DVar.Modifier = Data.Modifier;
+        DVar.AppliedToPointee = Data.AppliedToPointee;
         return DVar;
       }
     }
@@ -1696,7 +1712,7 @@
       // listed in a firstprivate clause, even if they are static data members.
       DSAVarData DVarTemp = hasInnermostDSA(
           D,
-          [](OpenMPClauseKind C) {
+          [](OpenMPClauseKind C, bool) {
             return C == OMPC_firstprivate || C == OMPC_shared;
           },
           MatchesAlways, FromParent);
@@ -1725,6 +1741,7 @@
     DVar.ImplicitDSALoc = I->DefaultAttrLoc;
     DVar.DKind = I->Directive;
     DVar.Modifier = Data.Modifier;
+    DVar.AppliedToPointee = Data.AppliedToPointee;
   }
 
   return DVar;
@@ -1755,7 +1772,7 @@
 
 const DSAStackTy::DSAVarData
 DSAStackTy::hasDSA(ValueDecl *D,
-                   const llvm::function_ref<bool(OpenMPClauseKind)> CPred,
+                   const llvm::function_ref<bool(OpenMPClauseKind, bool)> CPred,
                    const llvm::function_ref<bool(OpenMPDirectiveKind)> DPred,
                    bool FromParent) const {
   if (isStackEmpty())
@@ -1771,14 +1788,14 @@
       continue;
     const_iterator NewI = I;
     DSAVarData DVar = getDSA(NewI, D);
-    if (I == NewI && CPred(DVar.CKind))
+    if (I == NewI && CPred(DVar.CKind, DVar.AppliedToPointee))
       return DVar;
   }
   return {};
 }
 
 const DSAStackTy::DSAVarData DSAStackTy::hasInnermostDSA(
-    ValueDecl *D, const llvm::function_ref<bool(OpenMPClauseKind)> CPred,
+    ValueDecl *D, const llvm::function_ref<bool(OpenMPClauseKind, bool)> CPred,
     const llvm::function_ref<bool(OpenMPDirectiveKind)> DPred,
     bool FromParent) const {
   if (isStackEmpty())
@@ -1792,26 +1809,28 @@
     return {};
   const_iterator NewI = StartI;
   DSAVarData DVar = getDSA(NewI, D);
-  return (NewI == StartI && CPred(DVar.CKind)) ? DVar : DSAVarData();
+  return (NewI == StartI && CPred(DVar.CKind, DVar.AppliedToPointee))
+             ? DVar
+             : DSAVarData();
 }
 
 bool DSAStackTy::hasExplicitDSA(
-    const ValueDecl *D, const llvm::function_ref<bool(OpenMPClauseKind)> CPred,
+    const ValueDecl *D,
+    const llvm::function_ref<bool(OpenMPClauseKind, bool)> CPred,
     unsigned Level, bool NotLastprivate) const {
   if (getStackSize() <= Level)
     return false;
   D = getCanonicalDecl(D);
   const SharingMapTy &StackElem = getStackElemAtLevel(Level);
   auto I = StackElem.SharingMap.find(D);
-  if (I != StackElem.SharingMap.end() &&
-      I->getSecond().RefExpr.getPointer() &&
-      CPred(I->getSecond().Attributes) &&
+  if (I != StackElem.SharingMap.end() && I->getSecond().RefExpr.getPointer() &&
+      CPred(I->getSecond().Attributes, I->getSecond().AppliedToPointee) &&
       (!NotLastprivate || !I->getSecond().RefExpr.getInt()))
     return true;
   // Check predetermined rules for the loop control variables.
   auto LI = StackElem.LCVMap.find(D);
   if (LI != StackElem.LCVMap.end())
-    return CPred(OMPC_private);
+    return CPred(OMPC_private, /*AppliedToPointee=*/false);
   return false;
 }
 
@@ -2057,14 +2076,17 @@
       // By default, all the data that has a scalar type is mapped by copy
       // (except for reduction variables).
       // Defaultmap scalar is mutual exclusive to defaultmap pointer
-      IsByRef =
-          (DSAStack->isForceCaptureByReferenceInTargetExecutable() &&
-           !Ty->isAnyPointerType()) ||
-          !Ty->isScalarType() ||
-          DSAStack->isDefaultmapCapturedByRef(
-              Level, getVariableCategoryFromDecl(LangOpts, D)) ||
-          DSAStack->hasExplicitDSA(
-              D, [](OpenMPClauseKind K) { return K == OMPC_reduction; }, Level);
+      IsByRef = (DSAStack->isForceCaptureByReferenceInTargetExecutable() &&
+                 !Ty->isAnyPointerType()) ||
+                !Ty->isScalarType() ||
+                DSAStack->isDefaultmapCapturedByRef(
+                    Level, getVariableCategoryFromDecl(LangOpts, D)) ||
+                DSAStack->hasExplicitDSA(
+                    D,
+                    [](OpenMPClauseKind K, bool AppliedToPointee) {
+                      return K == OMPC_reduction && !AppliedToPointee;
+                    },
+                    Level);
     }
   }
 
@@ -2075,8 +2097,9 @@
               OMPD_target) ||
          !(DSAStack->hasExplicitDSA(
                D,
-               [](OpenMPClauseKind K) -> bool {
-                 return K == OMPC_firstprivate;
+               [](OpenMPClauseKind K, bool AppliedToPointee) -> bool {
+                 return K == OMPC_firstprivate ||
+                        (K == OMPC_reduction && AppliedToPointee);
                },
                Level, /*NotLastprivate=*/true) ||
            DSAStack->isUsesAllocatorsDecl(Level, D))) &&
@@ -2088,7 +2111,8 @@
         // copy
         !(DSAStack->getDefaultDSA() == DSA_firstprivate &&
           !DSAStack->hasExplicitDSA(
-              D, [](OpenMPClauseKind K) { return K != OMPC_unknown; }, Level) &&
+              D, [](OpenMPClauseKind K, bool) { return K != OMPC_unknown; },
+              Level) &&
           !DSAStack->isLoopControlVariable(D, Level).first);
   }
 
@@ -2151,7 +2175,8 @@
           !OMPDeclareTargetDeclAttr::isDeclareTargetDeclaration(VD))
         checkDeclIsAllowedInOpenMPTarget(nullptr, VD);
       return nullptr;
-    } else if (isInOpenMPTargetExecutionDirective()) {
+    }
+    if (isInOpenMPTargetExecutionDirective()) {
       // If the declaration is enclosed in a 'declare target' directive,
       // then it should not be captured.
       //
@@ -2204,7 +2229,8 @@
       return VD ? VD : Info.second;
     DSAStackTy::DSAVarData DVarTop =
         DSAStack->getTopDSA(D, DSAStack->isClauseParsingMode());
-    if (DVarTop.CKind != OMPC_unknown && isOpenMPPrivate(DVarTop.CKind))
+    if (DVarTop.CKind != OMPC_unknown && isOpenMPPrivate(DVarTop.CKind) &&
+        (!VD || VD->hasLocalStorage() || !DVarTop.AppliedToPointee))
       return VD ? VD : cast<VarDecl>(DVarTop.PrivateCopy->getDecl());
     // Threadprivate variables must not be captured.
     if (isOpenMPThreadPrivate(DVarTop.CKind))
@@ -2212,7 +2238,11 @@
     // The variable is not private or it is the variable in the directive with
     // default(none) clause and not used in any clause.
     DSAStackTy::DSAVarData DVarPrivate = DSAStack->hasDSA(
-        D, isOpenMPPrivate, [](OpenMPDirectiveKind) { return true; },
+        D,
+        [](OpenMPClauseKind C, bool AppliedToPointee) {
+          return isOpenMPPrivate(C) && !AppliedToPointee;
+        },
+        [](OpenMPDirectiveKind) { return true; },
         DSAStack->isClauseParsingMode());
     // Global shared must not be captured.
     if (VD && !VD->hasLocalStorage() && DVarPrivate.CKind == OMPC_unknown &&
@@ -2266,7 +2296,8 @@
         (IsTriviallyCopyable ||
          !isOpenMPTaskLoopDirective(CaptureRegions[CapLevel]))) {
       if (DSAStack->hasExplicitDSA(
-              D, [](OpenMPClauseKind K) { return K == OMPC_firstprivate; },
+              D,
+              [](OpenMPClauseKind K, bool) { return K == OMPC_firstprivate; },
               Level, /*NotLastprivate=*/true))
         return OMPC_firstprivate;
       DSAStackTy::DSAVarData DVar = DSAStack->getImplicitDSA(D, Level);
@@ -2287,7 +2318,8 @@
     if ((DSAStack->getPossiblyLoopCunter() == D->getCanonicalDecl() ||
          DSAStack->isLoopControlVariable(D).first) &&
         !DSAStack->hasExplicitDSA(
-            D, [](OpenMPClauseKind K) { return K != OMPC_private; }, Level) &&
+            D, [](OpenMPClauseKind K, bool) { return K != OMPC_private; },
+            Level) &&
         !isOpenMPSimdDirective(DSAStack->getCurrentDirective()))
       return OMPC_private;
   }
@@ -2295,7 +2327,8 @@
     if (DSAStack->isThreadPrivate(const_cast<VarDecl *>(VD)) &&
         DSAStack->isForceVarCapturing() &&
         !DSAStack->hasExplicitDSA(
-            D, [](OpenMPClauseKind K) { return K == OMPC_copyin; }, Level))
+            D, [](OpenMPClauseKind K, bool) { return K == OMPC_copyin; },
+            Level))
       return OMPC_private;
   }
   // User-defined allocators are private since they must be defined in the
@@ -2306,7 +2339,8 @@
           DSAStackTy::UsesAllocatorsDeclKind::UserDefinedAllocator)
     return OMPC_private;
   return (DSAStack->hasExplicitDSA(
-              D, [](OpenMPClauseKind K) { return K == OMPC_private; }, Level) ||
+              D, [](OpenMPClauseKind K, bool) { return K == OMPC_private; },
+              Level) ||
           (DSAStack->isClauseParsingMode() &&
            DSAStack->getClauseParsingMode() == OMPC_private) ||
           // Consider taskgroup reduction descriptor variable a private
@@ -2331,15 +2365,16 @@
   OpenMPClauseKind OMPC = OMPC_unknown;
   for (unsigned I = DSAStack->getNestingLevel() + 1; I > Level; --I) {
     const unsigned NewLevel = I - 1;
-    if (DSAStack->hasExplicitDSA(D,
-                                 [&OMPC](const OpenMPClauseKind K) {
-                                   if (isOpenMPPrivate(K)) {
-                                     OMPC = K;
-                                     return true;
-                                   }
-                                   return false;
-                                 },
-                                 NewLevel))
+    if (DSAStack->hasExplicitDSA(
+            D,
+            [&OMPC](const OpenMPClauseKind K, bool AppliedToPointee) {
+              if (isOpenMPPrivate(K) && !AppliedToPointee) {
+                OMPC = K;
+                return true;
+              }
+              return false;
+            },
+            NewLevel))
       break;
     if (DSAStack->checkMappableExprComponentListsForDeclAtLevel(
             D, NewLevel,
@@ -3474,7 +3509,10 @@
       //  enclosing worksharing or parallel construct may not be accessed in an
       //  explicit task.
       DVar = Stack->hasInnermostDSA(
-          VD, [](OpenMPClauseKind C) { return C == OMPC_reduction; },
+          VD,
+          [](OpenMPClauseKind C, bool AppliedToPointee) {
+            return C == OMPC_reduction && !AppliedToPointee;
+          },
           [](OpenMPDirectiveKind K) {
             return isOpenMPParallelDirective(K) ||
                    isOpenMPWorksharingDirective(K) || isOpenMPTeamsDirective(K);
@@ -3559,7 +3597,10 @@
       //  enclosing worksharing or parallel construct may not be accessed in
       //  an  explicit task.
       DVar = Stack->hasInnermostDSA(
-          FD, [](OpenMPClauseKind C) { return C == OMPC_reduction; },
+          FD,
+          [](OpenMPClauseKind C, bool AppliedToPointee) {
+            return C == OMPC_reduction && !AppliedToPointee;
+          },
           [](OpenMPDirectiveKind K) {
             return isOpenMPParallelDirective(K) ||
                    isOpenMPWorksharingDirective(K) || isOpenMPTeamsDirective(K);
@@ -14044,7 +14085,10 @@
       //  from the worksharing construct.
       if (isOpenMPTaskingDirective(CurrDir)) {
         DVar = DSAStack->hasInnermostDSA(
-            D, [](OpenMPClauseKind C) { return C == OMPC_reduction; },
+            D,
+            [](OpenMPClauseKind C, bool AppliedToPointee) {
+              return C == OMPC_reduction && !AppliedToPointee;
+            },
             [](OpenMPDirectiveKind K) {
               return isOpenMPParallelDirective(K) ||
                      isOpenMPWorksharingDirective(K) ||
@@ -14435,7 +14479,11 @@
       if (DVar.CKind != OMPC_unknown)
         return true;
       DSAStackTy::DSAVarData DVarPrivate = Stack->hasDSA(
-          VD, isOpenMPPrivate, [](OpenMPDirectiveKind) { return true; },
+          VD,
+          [](OpenMPClauseKind C, bool AppliedToPointee) {
+            return isOpenMPPrivate(C) && !AppliedToPointee;
+          },
+          [](OpenMPDirectiveKind) { return true; },
           /*FromParent=*/true);
       return DVarPrivate.CKind != OMPC_unknown;
     }
@@ -15513,7 +15561,8 @@
     // correct analysis of in_reduction clauses.
     if (CurrDir == OMPD_taskgroup && ClauseKind == OMPC_task_reduction)
       Modifier = OMPC_REDUCTION_task;
-    Stack->addDSA(D, RefExpr->IgnoreParens(), OMPC_reduction, Ref, Modifier);
+    Stack->addDSA(D, RefExpr->IgnoreParens(), OMPC_reduction, Ref, Modifier,
+                  ASE || OASE);
     if (Modifier == OMPC_REDUCTION_task &&
         (CurrDir == OMPD_taskgroup ||
          ((isOpenMPParallelDirective(CurrDir) ||
diff --git a/clang/test/OpenMP/distribute_parallel_for_reduction_task_codegen.cpp b/clang/test/OpenMP/distribute_parallel_for_reduction_task_codegen.cpp
--- a/clang/test/OpenMP/distribute_parallel_for_reduction_task_codegen.cpp
+++ b/clang/test/OpenMP/distribute_parallel_for_reduction_task_codegen.cpp
@@ -20,9 +20,9 @@
   }
 }
 
-// CHECK: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @{{.+}}, i32 4, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i64, i64, i32*, i8***)* [[OUTLINED:@.+]] to void (i32*, i32*, ...)*), i64 %{{.+}}, i64 %{{.+}}, i32* %{{.+}}, i8*** %{{.+}})
+// CHECK: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @{{.+}}, i32 4, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i64, i64, i32*, i8**)* [[OUTLINED:@.+]] to void (i32*, i32*, ...)*), i64 %{{.+}}, i64 %{{.+}}, i32* %{{.+}}, i8** %{{.+}})
 
-// CHECK: define internal void [[OUTLINED]](i32* noalias %{{.+}}, i32* noalias %{{.+}}, i64 %{{.+}}, i64 %{{.+}}, i32* {{.+}}, i8*** {{.+}})
+// CHECK: define internal void [[OUTLINED]](i32* noalias %{{.+}}, i32* noalias %{{.+}}, i64 %{{.+}}, i64 %{{.+}}, i32* {{.+}}, i8** {{.+}})
 // CHECK: alloca i32,
 // CHECK: [[ARGC_FP_ADDR:%.+]] = alloca i32,
 // CHECK: [[TR:%.+]] = alloca [2 x %struct.kmp_taskred_input_t],
@@ -124,7 +124,6 @@
 // CHECK_DAG: [[TG]] = load i8*, i8** [[TG_ADDR]],
 // CHECK-DAG: [[ARGV_REF]] = load i8*, i8** [[ARGV_ADDR:%.+]],
 // CHECK-DAG: [[ARGV_ADDR]] = load i8**, i8*** [[ARGV_ADDR_REF:%.+]],
-// CHECK-DAG: [[ARGV_ADDR_REF:%.+]] = load i8***, i8**** [[ARGV:%.+]],
-// CHECK-DAG: [[ARGV]] = getelementptr inbounds [[CAPS_TY]], [[CAPS_TY]]* [[CAP]], i32 0, i32 2
+// CHECK-DAG: [[ARGV_ADDR_REF]] = getelementptr inbounds [[CAPS_TY]], [[CAPS_TY]]* [[CAP]], i32 0, i32 2
 
 #endif
diff --git a/clang/test/OpenMP/for_reduction_task_codegen.cpp b/clang/test/OpenMP/for_reduction_task_codegen.cpp
--- a/clang/test/OpenMP/for_reduction_task_codegen.cpp
+++ b/clang/test/OpenMP/for_reduction_task_codegen.cpp
@@ -124,7 +124,6 @@
 // CHECK_DAG: [[TG]] = load i8*, i8** [[TG_ADDR]],
 // CHECK-DAG: [[ARGV_REF]] = load i8*, i8** [[ARGV_ADDR:%.+]],
 // CHECK-DAG: [[ARGV_ADDR]] = load i8**, i8*** [[ARGV_ADDR_REF:%.+]],
-// CHECK-DAG: [[ARGV_ADDR_REF:%.+]] = load i8***, i8**** [[ARGV:%.+]],
-// CHECK-DAG: [[ARGV]] = getelementptr inbounds [[CAPS_TY]], [[CAPS_TY]]* [[CAP]], i32 0, i32 2
+// CHECK-DAG: [[ARGV_ADDR_REF]] = getelementptr inbounds [[CAPS_TY]], [[CAPS_TY]]* [[CAP]], i32 0, i32 2
 
 #endif
diff --git a/clang/test/OpenMP/parallel_for_reduction_task_codegen.cpp b/clang/test/OpenMP/parallel_for_reduction_task_codegen.cpp
--- a/clang/test/OpenMP/parallel_for_reduction_task_codegen.cpp
+++ b/clang/test/OpenMP/parallel_for_reduction_task_codegen.cpp
@@ -19,9 +19,9 @@
   }
 }
 
-// CHECK: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @{{.+}}, i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32*, i8***)* [[OUTLINED:@.+]] to void (i32*, i32*, ...)*), i32* %{{.+}}, i8*** %{{.+}})
+// CHECK: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @{{.+}}, i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32*, i8**)* [[OUTLINED:@.+]] to void (i32*, i32*, ...)*), i32* %{{.+}}, i8** %{{.+}})
 
-// CHECK: define internal void [[OUTLINED]](i32* noalias %{{.+}}, i32* noalias %{{.+}}, i32* {{.+}}, i8*** {{.+}})
+// CHECK: define internal void [[OUTLINED]](i32* noalias %{{.+}}, i32* noalias %{{.+}}, i32* {{.+}}, i8** {{.+}})
 // CHECK: alloca i32,
 // CHECK: [[ARGC_FP_ADDR:%.+]] = alloca i32,
 // CHECK: [[TR:%.+]] = alloca [2 x %struct.kmp_taskred_input_t],
@@ -123,7 +123,6 @@
 // CHECK_DAG: [[TG]] = load i8*, i8** [[TG_ADDR]],
 // CHECK-DAG: [[ARGV_REF]] = load i8*, i8** [[ARGV_ADDR:%.+]],
 // CHECK-DAG: [[ARGV_ADDR]] = load i8**, i8*** [[ARGV_ADDR_REF:%.+]],
-// CHECK-DAG: [[ARGV_ADDR_REF:%.+]] = load i8***, i8**** [[ARGV:%.+]],
-// CHECK-DAG: [[ARGV]] = getelementptr inbounds [[CAPS_TY]], [[CAPS_TY]]* [[CAP]], i32 0, i32 2
+// CHECK-DAG: [[ARGV_ADDR_REF]] = getelementptr inbounds [[CAPS_TY]], [[CAPS_TY]]* [[CAP]], i32 0, i32 2
 
 #endif
diff --git a/clang/test/OpenMP/parallel_master_reduction_task_codegen.cpp b/clang/test/OpenMP/parallel_master_reduction_task_codegen.cpp
--- a/clang/test/OpenMP/parallel_master_reduction_task_codegen.cpp
+++ b/clang/test/OpenMP/parallel_master_reduction_task_codegen.cpp
@@ -19,9 +19,9 @@
   }
 }
 
-// CHECK: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @{{.+}}, i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32*, i8***)* [[OUTLINED:@.+]] to void (i32*, i32*, ...)*), i32* %{{.+}}, i8*** %{{.+}})
+// CHECK: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @{{.+}}, i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32*, i8**)* [[OUTLINED:@.+]] to void (i32*, i32*, ...)*), i32* %{{.+}}, i8** %{{.+}})
 
-// CHECK: define internal void [[OUTLINED]](i32* noalias %{{.+}}, i32* noalias %{{.+}}, i32* {{.+}}, i8*** {{.+}})
+// CHECK: define internal void [[OUTLINED]](i32* noalias %{{.+}}, i32* noalias %{{.+}}, i32* {{.+}}, i8** {{.+}})
 // CHECK: [[ARGC_FP_ADDR:%.+]] = alloca i32,
 // CHECK: [[TR:%.+]] = alloca [2 x %struct.kmp_taskred_input_t],
 // CHECK: [[TG:%.+]] = alloca i8*,
@@ -122,7 +122,6 @@
 // CHECK_DAG: [[TG]] = load i8*, i8** [[TG_ADDR]],
 // CHECK-DAG: [[ARGV_REF]] = load i8*, i8** [[ARGV_ADDR:%.+]],
 // CHECK-DAG: [[ARGV_ADDR]] = load i8**, i8*** [[ARGV_ADDR_REF:%.+]],
-// CHECK-DAG: [[ARGV_ADDR_REF:%.+]] = load i8***, i8**** [[ARGV:%.+]],
-// CHECK-DAG: [[ARGV]] = getelementptr inbounds [[CAPS_TY]], [[CAPS_TY]]* [[CAP]], i32 0, i32 2
+// CHECK-DAG: [[ARGV_ADDR_REF]] = getelementptr inbounds [[CAPS_TY]], [[CAPS_TY]]* [[CAP]], i32 0, i32 2
 
 #endif
diff --git a/clang/test/OpenMP/parallel_reduction_task_codegen.cpp b/clang/test/OpenMP/parallel_reduction_task_codegen.cpp
--- a/clang/test/OpenMP/parallel_reduction_task_codegen.cpp
+++ b/clang/test/OpenMP/parallel_reduction_task_codegen.cpp
@@ -19,9 +19,9 @@
   }
 }
 
-// CHECK: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @{{.+}}, i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32*, i8***)* [[OUTLINED:@.+]] to void (i32*, i32*, ...)*), i32* %{{.+}}, i8*** %{{.+}})
+// CHECK: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @{{.+}}, i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32*, i8**)* [[OUTLINED:@.+]] to void (i32*, i32*, ...)*), i32* %{{.+}}, i8** %{{.+}})
 
-// CHECK: define internal void [[OUTLINED]](i32* noalias %{{.+}}, i32* noalias %{{.+}}, i32* {{.+}}, i8*** {{.+}})
+// CHECK: define internal void [[OUTLINED]](i32* noalias %{{.+}}, i32* noalias %{{.+}}, i32* {{.+}}, i8** {{.+}})
 // CHECK: [[ARGC_FP_ADDR:%.+]] = alloca i32,
 // CHECK: [[TR:%.+]] = alloca [2 x %struct.kmp_taskred_input_t],
 // CHECK: [[TG:%.+]] = alloca i8*,
@@ -122,7 +122,6 @@
 // CHECK_DAG: [[TG]] = load i8*, i8** [[TG_ADDR]],
 // CHECK-DAG: [[ARGV_REF]] = load i8*, i8** [[ARGV_ADDR:%.+]],
 // CHECK-DAG: [[ARGV_ADDR]] = load i8**, i8*** [[ARGV_ADDR_REF:%.+]],
-// CHECK-DAG: [[ARGV_ADDR_REF:%.+]] = load i8***, i8**** [[ARGV:%.+]],
-// CHECK-DAG: [[ARGV]] = getelementptr inbounds [[CAPS_TY]], [[CAPS_TY]]* [[CAP]], i32 0, i32 2
+// CHECK-DAG: [[ARGV_ADDR_REF]] = getelementptr inbounds [[CAPS_TY]], [[CAPS_TY]]* [[CAP]], i32 0, i32 2
 
 #endif
diff --git a/clang/test/OpenMP/parallel_sections_reduction_task_codegen.cpp b/clang/test/OpenMP/parallel_sections_reduction_task_codegen.cpp
--- a/clang/test/OpenMP/parallel_sections_reduction_task_codegen.cpp
+++ b/clang/test/OpenMP/parallel_sections_reduction_task_codegen.cpp
@@ -19,9 +19,9 @@
   }
 }
 
-// CHECK: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @{{.+}}, i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32*, i8***)* [[OUTLINED:@.+]] to void (i32*, i32*, ...)*), i32* %{{.+}}, i8*** %{{.+}})
+// CHECK: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @{{.+}}, i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32*, i8**)* [[OUTLINED:@.+]] to void (i32*, i32*, ...)*), i32* %{{.+}}, i8** %{{.+}})
 
-// CHECK: define internal void [[OUTLINED]](i32* noalias %{{.+}}, i32* noalias %{{.+}}, i32* {{.+}}, i8*** {{.+}})
+// CHECK: define internal void [[OUTLINED]](i32* noalias %{{.+}}, i32* noalias %{{.+}}, i32* {{.+}}, i8** {{.+}})
 // CHECK: alloca i32,
 // CHECK: alloca i32,
 // CHECK: alloca i32,
@@ -127,7 +127,6 @@
 // CHECK_DAG: [[TG]] = load i8*, i8** [[TG_ADDR]],
 // CHECK-DAG: [[ARGV_REF]] = load i8*, i8** [[ARGV_ADDR:%.+]],
 // CHECK-DAG: [[ARGV_ADDR]] = load i8**, i8*** [[ARGV_ADDR_REF:%.+]],
-// CHECK-DAG: [[ARGV_ADDR_REF:%.+]] = load i8***, i8**** [[ARGV:%.+]],
-// CHECK-DAG: [[ARGV]] = getelementptr inbounds [[CAPS_TY]], [[CAPS_TY]]* [[CAP]], i32 0, i32 2
+// CHECK-DAG: [[ARGV_ADDR_REF]] = getelementptr inbounds [[CAPS_TY]], [[CAPS_TY]]* [[CAP]], i32 0, i32 2
 
 #endif
diff --git a/clang/test/OpenMP/sections_reduction_task_codegen.cpp b/clang/test/OpenMP/sections_reduction_task_codegen.cpp
--- a/clang/test/OpenMP/sections_reduction_task_codegen.cpp
+++ b/clang/test/OpenMP/sections_reduction_task_codegen.cpp
@@ -128,7 +128,6 @@
 // CHECK_DAG: [[TG]] = load i8*, i8** [[TG_ADDR]],
 // CHECK-DAG: [[ARGV_REF]] = load i8*, i8** [[ARGV_ADDR:%.+]],
 // CHECK-DAG: [[ARGV_ADDR]] = load i8**, i8*** [[ARGV_ADDR_REF:%.+]],
-// CHECK-DAG: [[ARGV_ADDR_REF:%.+]] = load i8***, i8**** [[ARGV:%.+]],
-// CHECK-DAG: [[ARGV]] = getelementptr inbounds [[CAPS_TY]], [[CAPS_TY]]* [[CAP]], i32 0, i32 2
+// CHECK-DAG: [[ARGV_ADDR_REF]] = getelementptr inbounds [[CAPS_TY]], [[CAPS_TY]]* [[CAP]], i32 0, i32 2
 
 #endif
diff --git a/clang/test/OpenMP/target_parallel_for_reduction_task_codegen.cpp b/clang/test/OpenMP/target_parallel_for_reduction_task_codegen.cpp
--- a/clang/test/OpenMP/target_parallel_for_reduction_task_codegen.cpp
+++ b/clang/test/OpenMP/target_parallel_for_reduction_task_codegen.cpp
@@ -19,9 +19,9 @@
   }
 }
 
-// CHECK: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @{{.+}}, i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32*, i8***)* [[OUTLINED:@.+]] to void (i32*, i32*, ...)*), i32* %{{.+}}, i8*** %{{.+}})
+// CHECK: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @{{.+}}, i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32*, i8**)* [[OUTLINED:@.+]] to void (i32*, i32*, ...)*), i32* %{{.+}}, i8** %{{.+}})
 
-// CHECK: define internal void [[OUTLINED]](i32* noalias %{{.+}}, i32* noalias %{{.+}}, i32* {{.+}}, i8*** {{.+}})
+// CHECK: define internal void [[OUTLINED]](i32* noalias %{{.+}}, i32* noalias %{{.+}}, i32* {{.+}}, i8** {{.+}})
 // CHECK: alloca i32,
 // CHECK: [[ARGC_FP_ADDR:%.+]] = alloca i32,
 // CHECK: [[TR:%.+]] = alloca [2 x %struct.kmp_taskred_input_t],
@@ -123,7 +123,6 @@
 // CHECK_DAG: [[TG]] = load i8*, i8** [[TG_ADDR]],
 // CHECK-DAG: [[ARGV_REF]] = load i8*, i8** [[ARGV_ADDR:%.+]],
 // CHECK-DAG: [[ARGV_ADDR]] = load i8**, i8*** [[ARGV_ADDR_REF:%.+]],
-// CHECK-DAG: [[ARGV_ADDR_REF:%.+]] = load i8***, i8**** [[ARGV:%.+]],
-// CHECK-DAG: [[ARGV]] = getelementptr inbounds [[CAPS_TY]], [[CAPS_TY]]* [[CAP]], i32 0, i32 2
+// CHECK-DAG: [[ARGV_ADDR_REF]] = getelementptr inbounds [[CAPS_TY]], [[CAPS_TY]]* [[CAP]], i32 0, i32 2
 
 #endif
diff --git a/clang/test/OpenMP/target_parallel_reduction_task_codegen.cpp b/clang/test/OpenMP/target_parallel_reduction_task_codegen.cpp
--- a/clang/test/OpenMP/target_parallel_reduction_task_codegen.cpp
+++ b/clang/test/OpenMP/target_parallel_reduction_task_codegen.cpp
@@ -19,9 +19,9 @@
   }
 }
 
-// CHECK: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @{{.+}}, i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32*, i8***)* [[OUTLINED:@.+]] to void (i32*, i32*, ...)*), i32* %{{.+}}, i8*** %{{.+}})
+// CHECK: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @{{.+}}, i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32*, i8**)* [[OUTLINED:@.+]] to void (i32*, i32*, ...)*), i32* %{{.+}}, i8** %{{.+}})
 
-// CHECK: define internal void [[OUTLINED]](i32* noalias %{{.+}}, i32* noalias %{{.+}}, i32* {{.+}}, i8*** {{.+}})
+// CHECK: define internal void [[OUTLINED]](i32* noalias %{{.+}}, i32* noalias %{{.+}}, i32* {{.+}}, i8** {{.+}})
 // CHECK: [[ARGC_FP_ADDR:%.+]] = alloca i32,
 // CHECK: [[TR:%.+]] = alloca [2 x %struct.kmp_taskred_input_t],
 // CHECK: [[TG:%.+]] = alloca i8*,
@@ -122,7 +122,6 @@
 // CHECK_DAG: [[TG]] = load i8*, i8** [[TG_ADDR]],
 // CHECK-DAG: [[ARGV_REF]] = load i8*, i8** [[ARGV_ADDR:%.+]],
 // CHECK-DAG: [[ARGV_ADDR]] = load i8**, i8*** [[ARGV_ADDR_REF:%.+]],
-// CHECK-DAG: [[ARGV_ADDR_REF:%.+]] = load i8***, i8**** [[ARGV:%.+]],
-// CHECK-DAG: [[ARGV]] = getelementptr inbounds [[CAPS_TY]], [[CAPS_TY]]* [[CAP]], i32 0, i32 2
+// CHECK-DAG: [[ARGV_ADDR_REF]] = getelementptr inbounds [[CAPS_TY]], [[CAPS_TY]]* [[CAP]], i32 0, i32 2
 
 #endif
diff --git a/clang/test/OpenMP/target_teams_distribute_parallel_for_reduction_task_codegen.cpp b/clang/test/OpenMP/target_teams_distribute_parallel_for_reduction_task_codegen.cpp
--- a/clang/test/OpenMP/target_teams_distribute_parallel_for_reduction_task_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_distribute_parallel_for_reduction_task_codegen.cpp
@@ -19,9 +19,9 @@
   }
 }
 
-// CHECK: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @{{.+}}, i32 4, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i64, i64, i32*, i8***)* [[OUTLINED:@.+]] to void (i32*, i32*, ...)*), i64 %{{.+}}, i64 %{{.+}}, i32* %{{.+}}, i8*** %{{.+}})
+// CHECK: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @{{.+}}, i32 4, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i64, i64, i32*, i8**)* [[OUTLINED:@.+]] to void (i32*, i32*, ...)*), i64 %{{.+}}, i64 %{{.+}}, i32* %{{.+}}, i8** %{{.+}})
 
-// CHECK: define internal void [[OUTLINED]](i32* noalias %{{.+}}, i32* noalias %{{.+}}, i64 %{{.+}}, i64 %{{.+}}, i32* {{.+}}, i8*** {{.+}})
+// CHECK: define internal void [[OUTLINED]](i32* noalias %{{.+}}, i32* noalias %{{.+}}, i64 %{{.+}}, i64 %{{.+}}, i32* {{.+}}, i8** {{.+}})
 // CHECK: alloca i32,
 // CHECK: [[ARGC_FP_ADDR:%.+]] = alloca i32,
 // CHECK: [[TR:%.+]] = alloca [2 x [[TASKRED_TY:%struct.kmp_taskred_input_t.*]]],
@@ -123,7 +123,6 @@
 // CHECK_DAG: [[TG]] = load i8*, i8** [[TG_ADDR]],
 // CHECK-DAG: [[ARGV_REF]] = load i8*, i8** [[ARGV_ADDR:%.+]],
 // CHECK-DAG: [[ARGV_ADDR]] = load i8**, i8*** [[ARGV_ADDR_REF:%.+]],
-// CHECK-DAG: [[ARGV_ADDR_REF:%.+]] = load i8***, i8**** [[ARGV:%.+]],
-// CHECK-DAG: [[ARGV]] = getelementptr inbounds [[CAPS_TY]], [[CAPS_TY]]* [[CAP]], i32 0, i32 2
+// CHECK-DAG: [[ARGV_ADDR_REF]] = getelementptr inbounds [[CAPS_TY]], [[CAPS_TY]]* [[CAP]], i32 0, i32 2
 
 #endif
diff --git a/clang/test/OpenMP/teams_distribute_parallel_for_reduction_task_codegen.cpp b/clang/test/OpenMP/teams_distribute_parallel_for_reduction_task_codegen.cpp
--- a/clang/test/OpenMP/teams_distribute_parallel_for_reduction_task_codegen.cpp
+++ b/clang/test/OpenMP/teams_distribute_parallel_for_reduction_task_codegen.cpp
@@ -20,9 +20,9 @@
   }
 }
 
-// CHECK: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @{{.+}}, i32 4, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i64, i64, i32*, i8***)* [[OUTLINED:@.+]] to void (i32*, i32*, ...)*), i64 %{{.+}}, i64 %{{.+}}, i32* %{{.+}}, i8*** %{{.+}})
+// CHECK: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @{{.+}}, i32 4, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i64, i64, i32*, i8**)* [[OUTLINED:@.+]] to void (i32*, i32*, ...)*), i64 %{{.+}}, i64 %{{.+}}, i32* %{{.+}}, i8** %{{.+}})
 
-// CHECK: define internal void [[OUTLINED]](i32* noalias %{{.+}}, i32* noalias %{{.+}}, i64 %{{.+}}, i64 %{{.+}}, i32* {{.+}}, i8*** {{.+}})
+// CHECK: define internal void [[OUTLINED]](i32* noalias %{{.+}}, i32* noalias %{{.+}}, i64 %{{.+}}, i64 %{{.+}}, i32* {{.+}}, i8** {{.+}})
 // CHECK: alloca i32,
 // CHECK: [[ARGC_FP_ADDR:%.+]] = alloca i32,
 // CHECK: [[TR:%.+]] = alloca [2 x [[TASKRED_TY:%struct.kmp_taskred_input_t.*]]],
@@ -124,7 +124,6 @@
 // CHECK_DAG: [[TG]] = load i8*, i8** [[TG_ADDR]],
 // CHECK-DAG: [[ARGV_REF]] = load i8*, i8** [[ARGV_ADDR:%.+]],
 // CHECK-DAG: [[ARGV_ADDR]] = load i8**, i8*** [[ARGV_ADDR_REF:%.+]],
-// CHECK-DAG: [[ARGV_ADDR_REF:%.+]] = load i8***, i8**** [[ARGV:%.+]],
-// CHECK-DAG: [[ARGV]] = getelementptr inbounds [[CAPS_TY]], [[CAPS_TY]]* [[CAP]], i32 0, i32 2
+// CHECK-DAG: [[ARGV_ADDR_REF]] = getelementptr inbounds [[CAPS_TY]], [[CAPS_TY]]* [[CAP]], i32 0, i32 2
 
 #endif