[OPENMP] Replace calls of getAssociatedStmt().

getAssociatedStmt() returns the outermost captured statement for the
OpenMP directive. It may return incorrect region in case of combined
constructs. Reworked the code to reduce the number of calls of
getAssociatedStmt() and used getInnermostCapturedStmt() and
getCapturedStmt() functions instead.
In case of firstprivate variables it may lead to an extra allocas
generation for private copies even if the variable is passed by value
into outlined function and could be used directly as private copy.

llvm-svn: 322393

16 files changed, 151 insertions, 188 deletions

diff --git a/clang/include/clang/AST/StmtOpenMP.h b/clang/include/clang/AST/StmtOpenMP.h
index b6fd1a2c28c..88052097c2f 100644
--- a/clang/include/clang/AST/StmtOpenMP.h
+++ b/clang/include/clang/AST/StmtOpenMP.h
@@ -222,6 +222,25 @@ public:
     llvm_unreachable("Incorrect RegionKind specified for directive.");
   }
 
+  /// Get innermost captured statement for the construct.
+  CapturedStmt *getInnermostCapturedStmt() {
+    assert(hasAssociatedStmt() && getAssociatedStmt() &&
+           "Must have associated statement.");
+    SmallVector<OpenMPDirectiveKind, 4> CaptureRegions;
+    getOpenMPCaptureRegions(CaptureRegions, getDirectiveKind());
+    assert(!CaptureRegions.empty() &&
+           "At least one captured statement must be provided.");
+    auto *CS = cast<CapturedStmt>(getAssociatedStmt());
+    for (unsigned Level = CaptureRegions.size(); Level > 1; --Level)
+      CS = cast<CapturedStmt>(CS->getCapturedStmt());
+    return CS;
+  }
+
+  const CapturedStmt *getInnermostCapturedStmt() const {
+    return const_cast<OMPExecutableDirective *>(this)
+        ->getInnermostCapturedStmt();
+  }
+
   OpenMPDirectiveKind getDirectiveKind() const { return Kind; }
 
   static bool classof(const Stmt *S) {
@@ -903,9 +922,8 @@ public:
   }
   const Stmt *getBody() const {
     // This relies on the loop form is already checked by Sema.
-    const Stmt *Body = getAssociatedStmt()->IgnoreContainers(true);
-    while(const auto *CS = dyn_cast<CapturedStmt>(Body))
-      Body = CS->getCapturedStmt();
+    const Stmt *Body =
+        getInnermostCapturedStmt()->getCapturedStmt()->IgnoreContainers();
     Body = cast<ForStmt>(Body)->getBody();
     for (unsigned Cnt = 1; Cnt < CollapsedNum; ++Cnt) {
       Body = Body->IgnoreContainers();
diff --git a/clang/lib/AST/StmtPrinter.cpp b/clang/lib/AST/StmtPrinter.cpp
index d7e668a8328..de8f6a6481c 100644
--- a/clang/lib/AST/StmtPrinter.cpp
+++ b/clang/lib/AST/StmtPrinter.cpp
@@ -1034,12 +1034,8 @@ void StmtPrinter::PrintOMPExecutableDirective(OMPExecutableDirective *S,
       OS << ' ';
     }
   OS << "\n";
-  if (S->hasAssociatedStmt() && S->getAssociatedStmt() && !ForceNoStmt) {
-    assert(isa<CapturedStmt>(S->getAssociatedStmt()) &&
-           "Expected captured statement!");
-    Stmt *CS = cast<CapturedStmt>(S->getAssociatedStmt())->getCapturedStmt();
-    PrintStmt(CS);
-  }
+  if (!ForceNoStmt && S->hasAssociatedStmt())
+    PrintStmt(S->getInnermostCapturedStmt()->getCapturedStmt());
 }
 
 void StmtPrinter::VisitOMPParallelDirective(OMPParallelDirective *Node) {
@@ -1142,7 +1138,7 @@ void StmtPrinter::VisitOMPFlushDirective(OMPFlushDirective *Node) {
 
 void StmtPrinter::VisitOMPOrderedDirective(OMPOrderedDirective *Node) {
   Indent() << "#pragma omp ordered ";
-  PrintOMPExecutableDirective(Node);
+  PrintOMPExecutableDirective(Node, Node->hasClausesOfKind<OMPDependClause>());
 }
 
 void StmtPrinter::VisitOMPAtomicDirective(OMPAtomicDirective *Node) {
diff --git a/clang/lib/CodeGen/CGOpenMPRuntime.cpp b/clang/lib/CodeGen/CGOpenMPRuntime.cpp
index 9fa173b2e09..9cff2a35784 100644
--- a/clang/lib/CodeGen/CGOpenMPRuntime.cpp
+++ b/clang/lib/CodeGen/CGOpenMPRuntime.cpp
@@ -1371,7 +1371,10 @@ llvm::Value *CGOpenMPRuntime::emitTaskOutlinedFunction(
   CodeGen.setAction(Action);
   assert(!ThreadIDVar->getType()->isPointerType() &&
          "thread id variable must be of type kmp_int32 for tasks");
-  auto *CS = cast<CapturedStmt>(D.getAssociatedStmt());
+  const OpenMPDirectiveKind Region =
+      isOpenMPTaskLoopDirective(D.getDirectiveKind()) ? OMPD_taskloop
+                                                      : OMPD_task;
+  auto *CS = D.getCapturedStmt(Region);
   auto *TD = dyn_cast<OMPTaskDirective>(&D);
   CodeGenFunction CGF(CGM, true);
   CGOpenMPTaskOutlinedRegionInfo CGInfo(*CS, ThreadIDVar, CodeGen,
@@ -5885,7 +5888,7 @@ void CGOpenMPRuntime::emitTargetOutlinedFunctionHelper(
        << llvm::format("_%x_", FileID) << ParentName << "_l" << Line;
   }
 
-  const CapturedStmt &CS = *cast<CapturedStmt>(D.getAssociatedStmt());
+  const CapturedStmt &CS = *D.getCapturedStmt(OMPD_target);
 
   CodeGenFunction CGF(CGM, true);
   CGOpenMPTargetRegionInfo CGInfo(CS, CodeGen, EntryFnName);
@@ -5979,7 +5982,7 @@ emitNumTeamsForTargetDirective(CGOpenMPRuntime &OMPRuntime,
   // the expression is captured in the enclosing target environment when the
   // teams directive is not combined with target.
 
-  const CapturedStmt &CS = *cast<CapturedStmt>(D.getAssociatedStmt());
+  const CapturedStmt &CS = *D.getCapturedStmt(OMPD_target);
 
   if (auto *TeamsDir = dyn_cast_or_null<OMPExecutableDirective>(
           ignoreCompoundStmts(CS.getCapturedStmt()))) {
@@ -6082,7 +6085,7 @@ emitNumThreadsForTargetDirective(CGOpenMPRuntime &OMPRuntime,
   // the expression is captured in the enclosing target environment when the
   // teams directive is not combined with target.
 
-  const CapturedStmt &CS = *cast<CapturedStmt>(D.getAssociatedStmt());
+  const CapturedStmt &CS = *D.getCapturedStmt(OMPD_target);
 
   if (auto *TeamsDir = dyn_cast_or_null<OMPExecutableDirective>(
           ignoreCompoundStmts(CS.getCapturedStmt()))) {
@@ -7059,7 +7062,7 @@ void CGOpenMPRuntime::emitTargetCall(CodeGenFunction &CGF,
   // Get mappable expression information.
   MappableExprsHandler MEHandler(D, CGF);
 
-  const CapturedStmt &CS = *cast<CapturedStmt>(D.getAssociatedStmt());
+  const CapturedStmt &CS = *D.getCapturedStmt(OMPD_target);
   auto RI = CS.getCapturedRecordDecl()->field_begin();
   auto CV = CapturedVars.begin();
   for (CapturedStmt::const_capture_iterator CI = CS.capture_begin(),
@@ -7314,12 +7317,11 @@ void CGOpenMPRuntime::scanForTargetRegionsFunctions(const Stmt *S,
   }
 
   if (const OMPExecutableDirective *E = dyn_cast<OMPExecutableDirective>(S)) {
-    if (!E->hasAssociatedStmt())
+    if (!E->hasAssociatedStmt() || !E->getAssociatedStmt())
       return;
 
     scanForTargetRegionsFunctions(
-        cast<CapturedStmt>(E->getAssociatedStmt())->getCapturedStmt(),
-        ParentName);
+        E->getInnermostCapturedStmt()->getCapturedStmt(), ParentName);
     return;
   }
 
diff --git a/clang/lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp b/clang/lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp
index b90e87a16e5..6c0f00d10ca 100644
--- a/clang/lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp
+++ b/clang/lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp
@@ -2428,7 +2428,7 @@ void CGOpenMPRuntimeNVPTX::emitOutlinedFunctionCall(
 llvm::Function *CGOpenMPRuntimeNVPTX::createDataSharingWrapper(
     llvm::Function *OutlinedParallelFn, const OMPExecutableDirective &D) {
   ASTContext &Ctx = CGM.getContext();
-  const auto &CS = *cast<CapturedStmt>(D.getAssociatedStmt());
+  const CapturedStmt &CS = *D.getCapturedStmt(OMPD_parallel);
 
   // Create a function that takes as argument the source thread.
   FunctionArgList WrapperArgs;
diff --git a/clang/lib/CodeGen/CGStmtOpenMP.cpp b/clang/lib/CodeGen/CGStmtOpenMP.cpp
index 7221ad95b66..116647d7cfa 100644
--- a/clang/lib/CodeGen/CGStmtOpenMP.cpp
+++ b/clang/lib/CodeGen/CGStmtOpenMP.cpp
@@ -53,34 +53,35 @@ class OMPLexicalScope : public CodeGenFunction::LexicalScope {
   }
 
 public:
-  OMPLexicalScope(CodeGenFunction &CGF, const OMPExecutableDirective &S,
-                  bool AsInlined = false, bool EmitPreInitStmt = true)
+  OMPLexicalScope(
+      CodeGenFunction &CGF, const OMPExecutableDirective &S,
+      const llvm::Optional<OpenMPDirectiveKind> CapturedRegion = llvm::None,
+      const bool EmitPreInitStmt = true)
       : CodeGenFunction::LexicalScope(CGF, S.getSourceRange()),
         InlinedShareds(CGF) {
     if (EmitPreInitStmt)
       emitPreInitStmt(CGF, S);
-    if (AsInlined) {
-      if (S.hasAssociatedStmt()) {
-        auto *CS = cast<CapturedStmt>(S.getAssociatedStmt());
-        for (auto &C : CS->captures()) {
-          if (C.capturesVariable() || C.capturesVariableByCopy()) {
-            auto *VD = C.getCapturedVar();
-            assert(VD == VD->getCanonicalDecl() &&
-                        "Canonical decl must be captured.");
-            DeclRefExpr DRE(const_cast<VarDecl *>(VD),
-                            isCapturedVar(CGF, VD) ||
-                                (CGF.CapturedStmtInfo &&
-                                 InlinedShareds.isGlobalVarCaptured(VD)),
-                            VD->getType().getNonReferenceType(), VK_LValue,
-                            SourceLocation());
-            InlinedShareds.addPrivate(VD, [&CGF, &DRE]() -> Address {
-              return CGF.EmitLValue(&DRE).getAddress();
-            });
-          }
-        }
-        (void)InlinedShareds.Privatize();
+    if (!CapturedRegion.hasValue())
+      return;
+    assert(S.hasAssociatedStmt() &&
+           "Expected associated statement for inlined directive.");
+    const CapturedStmt *CS = S.getCapturedStmt(*CapturedRegion);
+    for (auto &C : CS->captures()) {
+      if (C.capturesVariable() || C.capturesVariableByCopy()) {
+        auto *VD = C.getCapturedVar();
+        assert(VD == VD->getCanonicalDecl() &&
+               "Canonical decl must be captured.");
+        DeclRefExpr DRE(
+            const_cast<VarDecl *>(VD),
+            isCapturedVar(CGF, VD) || (CGF.CapturedStmtInfo &&
+                                       InlinedShareds.isGlobalVarCaptured(VD)),
+            VD->getType().getNonReferenceType(), VK_LValue, SourceLocation());
+        InlinedShareds.addPrivate(VD, [&CGF, &DRE]() -> Address {
+          return CGF.EmitLValue(&DRE).getAddress();
+        });
       }
     }
+    (void)InlinedShareds.Privatize();
   }
 };
 
@@ -96,9 +97,8 @@ class OMPParallelScope final : public OMPLexicalScope {
 
 public:
   OMPParallelScope(CodeGenFunction &CGF, const OMPExecutableDirective &S)
-      : OMPLexicalScope(CGF, S,
-                        /*AsInlined=*/false,
-                        /*EmitPreInitStmt=*/EmitPreInitStmt(S)) {}
+      : OMPLexicalScope(CGF, S, /*CapturedRegion=*/llvm::None,
+                        EmitPreInitStmt(S)) {}
 };
 
 /// Lexical scope for OpenMP teams construct, that handles correct codegen
@@ -112,9 +112,8 @@ class OMPTeamsScope final : public OMPLexicalScope {
 
 public:
   OMPTeamsScope(CodeGenFunction &CGF, const OMPExecutableDirective &S)
-      : OMPLexicalScope(CGF, S,
-                        /*AsInlined=*/false,
-                        /*EmitPreInitStmt=*/EmitPreInitStmt(S)) {}
+      : OMPLexicalScope(CGF, S, /*CapturedRegion=*/llvm::None,
+                        EmitPreInitStmt(S)) {}
 };
 
 /// Private scope for OpenMP loop-based directives, that supports capturing
@@ -738,7 +737,12 @@ bool CodeGenFunction::EmitOMPFirstprivateClause(const OMPExecutableDirective &D,
           cast<VarDecl>(cast<DeclRefExpr>(D)->getDecl())->getCanonicalDecl());
   }
   llvm::DenseSet<const VarDecl *> EmittedAsFirstprivate;
-  CGCapturedStmtInfo CapturesInfo(cast<CapturedStmt>(*D.getAssociatedStmt()));
+  llvm::SmallVector<OpenMPDirectiveKind, 4> CaptureRegions;
+  getOpenMPCaptureRegions(CaptureRegions, D.getDirectiveKind());
+  // Force emission of the firstprivate copy if the directive does not emit
+  // outlined function, like omp for, omp simd, omp distribute etc.
+  bool MustEmitFirstprivateCopy =
+      CaptureRegions.size() == 1 && CaptureRegions.back() == OMPD_unknown;
   for (const auto *C : D.getClausesOfKind<OMPFirstprivateClause>()) {
     auto IRef = C->varlist_begin();
     auto InitsRef = C->inits().begin();
@@ -746,9 +750,8 @@ bool CodeGenFunction::EmitOMPFirstprivateClause(const OMPExecutableDirective &D,
       auto *OrigVD = cast<VarDecl>(cast<DeclRefExpr>(*IRef)->getDecl());
       bool ThisFirstprivateIsLastprivate =
           Lastprivates.count(OrigVD->getCanonicalDecl()) > 0;
-      auto *CapFD = CapturesInfo.lookup(OrigVD);
       auto *FD = CapturedStmtInfo->lookup(OrigVD);
-      if (!ThisFirstprivateIsLastprivate && FD && (FD == CapFD) &&
+      if (!MustEmitFirstprivateCopy && !ThisFirstprivateIsLastprivate && FD &&
           !FD->getType()->isReferenceType()) {
         EmittedAsFirstprivate.insert(OrigVD->getCanonicalDecl());
         ++IRef;
@@ -1272,7 +1275,7 @@ void CodeGenFunction::EmitOMPParallelDirective(const OMPParallelDirective &S) {
     CGF.EmitOMPPrivateClause(S, PrivateScope);
     CGF.EmitOMPReductionClauseInit(S, PrivateScope);
     (void)PrivateScope.Privatize();
-    CGF.EmitStmt(cast<CapturedStmt>(S.getAssociatedStmt())->getCapturedStmt());
+    CGF.EmitStmt(S.getCapturedStmt(OMPD_parallel)->getCapturedStmt());
     CGF.EmitOMPReductionClauseFinal(S, /*ReductionKind=*/OMPD_parallel);
   };
   emitCommonOMPParallelDirective(*this, S, OMPD_parallel, CodeGen,
@@ -1734,7 +1737,7 @@ void CodeGenFunction::EmitOMPSimdDirective(const OMPSimdDirective &S) {
   auto &&CodeGen = [&S](CodeGenFunction &CGF, PrePostActionTy &Action) {
     emitOMPSimdRegion(CGF, S, Action);
   };
-  OMPLexicalScope Scope(*this, S, /*AsInlined=*/true);
+  OMPLexicalScope Scope(*this, S, OMPD_unknown);
   CGM.getOpenMPRuntime().emitInlinedDirective(*this, OMPD_simd, CodeGen);
 }
 
@@ -2122,7 +2125,7 @@ void CodeGenFunction::EmitOMPDistributeParallelForDirective(
     CGF.EmitOMPDistributeLoop(S, emitInnerParallelForWhenCombined,
                               S.getDistInc());
   };
-  OMPLexicalScope Scope(*this, S, /*AsInlined=*/true);
+  OMPLexicalScope Scope(*this, S, OMPD_parallel);
   CGM.getOpenMPRuntime().emitInlinedDirective(*this, OMPD_distribute, CodeGen);
 }
 
@@ -2132,7 +2135,7 @@ void CodeGenFunction::EmitOMPDistributeParallelForSimdDirective(
     CGF.EmitOMPDistributeLoop(S, emitInnerParallelForWhenCombined,
                               S.getDistInc());
   };
-  OMPLexicalScope Scope(*this, S, /*AsInlined=*/true);
+  OMPLexicalScope Scope(*this, S, OMPD_parallel);
   CGM.getOpenMPRuntime().emitInlinedDirective(*this, OMPD_distribute, CodeGen);
 }
 
@@ -2141,7 +2144,7 @@ void CodeGenFunction::EmitOMPDistributeSimdDirective(
   auto &&CodeGen = [&S](CodeGenFunction &CGF, PrePostActionTy &) {
     CGF.EmitOMPDistributeLoop(S, emitOMPLoopBodyWithStopPoint, S.getInc());
   };
-  OMPLexicalScope Scope(*this, S, /*AsInlined=*/true);
+  OMPLexicalScope Scope(*this, S, OMPD_unknown);
   CGM.getOpenMPRuntime().emitInlinedDirective(*this, OMPD_simd, CodeGen);
 }
 
@@ -2169,12 +2172,11 @@ void CodeGenFunction::EmitOMPTargetSimdDirective(
 
 void CodeGenFunction::EmitOMPTargetTeamsDistributeParallelForSimdDirective(
     const OMPTargetTeamsDistributeParallelForSimdDirective &S) {
-  OMPLexicalScope Scope(*this, S, /*AsInlined=*/true);
+  OMPLexicalScope Scope(*this, S, OMPD_unknown);
   CGM.getOpenMPRuntime().emitInlinedDirective(
       *this, OMPD_target_teams_distribute_parallel_for_simd,
       [&S](CodeGenFunction &CGF, PrePostActionTy &) {
-        CGF.EmitStmt(
-            cast<CapturedStmt>(S.getAssociatedStmt())->getCapturedStmt());
+        CGF.EmitStmt(S.getInnermostCapturedStmt()->getCapturedStmt());
       });
 }
 
@@ -2414,7 +2416,7 @@ void CodeGenFunction::EmitOMPForDirective(const OMPForDirective &S) {
                                                  emitDispatchForLoopBounds);
   };
   {
-    OMPLexicalScope Scope(*this, S, /*AsInlined=*/true);
+    OMPLexicalScope Scope(*this, S, OMPD_unknown);
     CGM.getOpenMPRuntime().emitInlinedDirective(*this, OMPD_for, CodeGen,
                                                 S.hasCancel());
   }
@@ -2434,7 +2436,7 @@ void CodeGenFunction::EmitOMPForSimdDirective(const OMPForSimdDirective &S) {
                                                  emitDispatchForLoopBounds);
   };
   {
-    OMPLexicalScope Scope(*this, S, /*AsInlined=*/true);
+    OMPLexicalScope Scope(*this, S, OMPD_unknown);
     CGM.getOpenMPRuntime().emitInlinedDirective(*this, OMPD_simd, CodeGen);
   }
 
@@ -2454,8 +2456,8 @@ static LValue createSectionLVal(CodeGenFunction &CGF, QualType Ty,
 }
 
 void CodeGenFunction::EmitSections(const OMPExecutableDirective &S) {
-  auto *Stmt = cast<CapturedStmt>(S.getAssociatedStmt())->getCapturedStmt();
-  auto *CS = dyn_cast<CompoundStmt>(Stmt);
+  const Stmt *Stmt = S.getInnermostCapturedStmt()->getCapturedStmt();
+  const auto *CS = dyn_cast<CompoundStmt>(Stmt);
   bool HasLastprivates = false;
   auto &&CodeGen = [&S, Stmt, CS, &HasLastprivates](CodeGenFunction &CGF,
                                                     PrePostActionTy &) {
@@ -2595,7 +2597,7 @@ void CodeGenFunction::EmitSections(const OMPExecutableDirective &S) {
 
 void CodeGenFunction::EmitOMPSectionsDirective(const OMPSectionsDirective &S) {
   {
-    OMPLexicalScope Scope(*this, S, /*AsInlined=*/true);
+    OMPLexicalScope Scope(*this, S, OMPD_unknown);
     EmitSections(S);
   }
   // Emit an implicit barrier at the end.
@@ -2607,9 +2609,9 @@ void CodeGenFunction::EmitOMPSectionsDirective(const OMPSectionsDirective &S) {
 
 void CodeGenFunction::EmitOMPSectionDirective(const OMPSectionDirective &S) {
   auto &&CodeGen = [&S](CodeGenFunction &CGF, PrePostActionTy &) {
-    CGF.EmitStmt(cast<CapturedStmt>(S.getAssociatedStmt())->getCapturedStmt());
+    CGF.EmitStmt(S.getInnermostCapturedStmt()->getCapturedStmt());
   };
-  OMPLexicalScope Scope(*this, S, /*AsInlined=*/true);
+  OMPLexicalScope Scope(*this, S, OMPD_unknown);
   CGM.getOpenMPRuntime().emitInlinedDirective(*this, OMPD_section, CodeGen,
                                               S.hasCancel());
 }
@@ -2638,10 +2640,10 @@ void CodeGenFunction::EmitOMPSingleDirective(const OMPSingleDirective &S) {
     (void)CGF.EmitOMPFirstprivateClause(S, SingleScope);
     CGF.EmitOMPPrivateClause(S, SingleScope);
     (void)SingleScope.Privatize();
-    CGF.EmitStmt(cast<CapturedStmt>(S.getAssociatedStmt())->getCapturedStmt());
+    CGF.EmitStmt(S.getInnermostCapturedStmt()->getCapturedStmt());
   };
   {
-    OMPLexicalScope Scope(*this, S, /*AsInlined=*/true);
+    OMPLexicalScope Scope(*this, S, OMPD_unknown);
     CGM.getOpenMPRuntime().emitSingleRegion(*this, CodeGen, S.getLocStart(),
                                             CopyprivateVars, DestExprs,
                                             SrcExprs, AssignmentOps);
@@ -2658,21 +2660,21 @@ void CodeGenFunction::EmitOMPSingleDirective(const OMPSingleDirective &S) {
 void CodeGenFunction::EmitOMPMasterDirective(const OMPMasterDirective &S) {
   auto &&CodeGen = [&S](CodeGenFunction &CGF, PrePostActionTy &Action) {
     Action.Enter(CGF);
-    CGF.EmitStmt(cast<CapturedStmt>(S.getAssociatedStmt())->getCapturedStmt());
+    CGF.EmitStmt(S.getInnermostCapturedStmt()->getCapturedStmt());
   };
-  OMPLexicalScope Scope(*this, S, /*AsInlined=*/true);
+  OMPLexicalScope Scope(*this, S, OMPD_unknown);
   CGM.getOpenMPRuntime().emitMasterRegion(*this, CodeGen, S.getLocStart());
 }
 
 void CodeGenFunction::EmitOMPCriticalDirective(const OMPCriticalDirective &S) {
   auto &&CodeGen = [&S](CodeGenFunction &CGF, PrePostActionTy &Action) {
     Action.Enter(CGF);
-    CGF.EmitStmt(cast<CapturedStmt>(S.getAssociatedStmt())->getCapturedStmt());
+    CGF.EmitStmt(S.getInnermostCapturedStmt()->getCapturedStmt());
   };
   Expr *Hint = nullptr;
   if (auto *HintClause = S.getSingleClause<OMPHintClause>())
     Hint = HintClause->getHint();
-  OMPLexicalScope Scope(*this, S, /*AsInlined=*/true);
+  OMPLexicalScope Scope(*this, S, OMPD_unknown);
   CGM.getOpenMPRuntime().emitCriticalRegion(*this,
                                             S.getDirectiveName().getAsString(),
                                             CodeGen, S.getLocStart(), Hint);
@@ -2714,12 +2716,12 @@ void CodeGenFunction::EmitOMPParallelSectionsDirective(
                                  emitEmptyBoundParameters);
 }
 
-void CodeGenFunction::EmitOMPTaskBasedDirective(const OMPExecutableDirective &S,
-                                                const RegionCodeGenTy &BodyGen,
-                                                const TaskGenTy &TaskGen,
-                                                OMPTaskDataTy &Data) {
+void CodeGenFunction::EmitOMPTaskBasedDirective(
+    const OMPExecutableDirective &S, const OpenMPDirectiveKind CapturedRegion,
+    const RegionCodeGenTy &BodyGen, const TaskGenTy &TaskGen,
+    OMPTaskDataTy &Data) {
   // Emit outlined function for task construct.
-  auto CS = cast<CapturedStmt>(S.getAssociatedStmt());
+  const CapturedStmt *CS = S.getCapturedStmt(CapturedRegion);
   auto *I = CS->getCapturedDecl()->param_begin();
   auto *PartId = std::next(I);
   auto *TaskT = std::next(I, 4);
@@ -2820,8 +2822,9 @@ void CodeGenFunction::EmitOMPTaskBasedDirective(const OMPExecutableDirective &S,
   for (const auto *C : S.getClausesOfKind<OMPDependClause>())
     for (auto *IRef : C->varlists())
       Data.Dependences.push_back(std::make_pair(C->getDependencyKind(), IRef));
-  auto &&CodeGen = [&Data, &S, CS, &BodyGen, &LastprivateDstsOrigs](
-      CodeGenFunction &CGF, PrePostActionTy &Action) {
+  auto &&CodeGen = [&Data, &S, CS, &BodyGen, &LastprivateDstsOrigs,
+                    CapturedRegion](CodeGenFunction &CGF,
+                                    PrePostActionTy &Action) {
     // Set proper addresses for generated private copies.
     OMPPrivateScope Scope(CGF);
     if (!Data.PrivateVars.empty() || !Data.FirstprivateVars.empty() ||
@@ -2878,7 +2881,7 @@ void CodeGenFunction::EmitOMPTaskBasedDirective(const OMPExecutableDirective &S,
       }
     }
     if (Data.Reductions) {
-      OMPLexicalScope LexScope(CGF, S, /*AsInlined=*/true);
+      OMPLexicalScope LexScope(CGF, S, CapturedRegion);
       ReductionCodeGen RedCG(Data.ReductionVars, Data.ReductionCopies,
                              Data.ReductionOps);
       llvm::Value *ReductionsPtr = CGF.Builder.CreateLoad(
@@ -3096,8 +3099,7 @@ void CodeGenFunction::EmitOMPTargetTaskBasedDirective(
         CGF.GetAddrOfLocalVar(SVD), /*Index=*/0, CGF.getSizeSize());
 
     Action.Enter(CGF);
-    OMPLexicalScope LexScope(CGF, S, /*AsInlined=*/true,
-                             /*EmitPreInitStmt=*/false);
+    OMPLexicalScope LexScope(CGF, S, OMPD_task, /*EmitPreInitStmt=*/false);
     BodyGen(CGF);
   };
   auto *OutlinedFn = CGM.getOpenMPRuntime().emitTaskOutlinedFunction(
@@ -3114,7 +3116,7 @@ void CodeGenFunction::EmitOMPTargetTaskBasedDirective(
 
 void CodeGenFunction::EmitOMPTaskDirective(const OMPTaskDirective &S) {
   // Emit outlined function for task construct.
-  auto CS = cast<CapturedStmt>(S.getAssociatedStmt());
+  const CapturedStmt *CS = S.getCapturedStmt(OMPD_task);
   auto CapturedStruct = GenerateCapturedStmtArgument(*CS);
   auto SharedsTy = getContext().getRecordType(CS->getCapturedRecordDecl());
   const Expr *IfCond = nullptr;
@@ -3139,7 +3141,7 @@ void CodeGenFunction::EmitOMPTaskDirective(const OMPTaskDirective &S) {
                                             SharedsTy, CapturedStruct, IfCond,
                                             Data);
   };
-  EmitOMPTaskBasedDirective(S, BodyGen, TaskGen, Data);
+  EmitOMPTaskBasedDirective(S, OMPD_task, BodyGen, TaskGen, Data);
 }
 
 void CodeGenFunction::EmitOMPTaskyieldDirective(
@@ -3188,9 +3190,9 @@ void CodeGenFunction::EmitOMPTaskgroupDirective(
       CGF.EmitStoreOfScalar(ReductionDesc, CGF.GetAddrOfLocalVar(VD),
                             /*Volatile=*/false, E->getType());
     }
-    CGF.EmitStmt(cast<CapturedStmt>(S.getAssociatedStmt())->getCapturedStmt());
+    CGF.EmitStmt(S.getInnermostCapturedStmt()->getCapturedStmt());
   };
-  OMPLexicalScope Scope(*this, S, /*AsInlined=*/true);
+  OMPLexicalScope Scope(*this, S, OMPD_unknown);
   CGM.getOpenMPRuntime().emitTaskgroupRegion(*this, CodeGen, S.getLocStart());
 }
 
@@ -3398,7 +3400,7 @@ void CodeGenFunction::EmitOMPDistributeDirective(
 
     CGF.EmitOMPDistributeLoop(S, emitOMPLoopBodyWithStopPoint, S.getInc());
   };
-  OMPLexicalScope Scope(*this, S, /*AsInlined=*/true);
+  OMPLexicalScope Scope(*this, S, OMPD_unknown);
   CGM.getOpenMPRuntime().emitInlinedDirective(*this, OMPD_distribute, CodeGen);
 }
 
@@ -3413,7 +3415,9 @@ static llvm::Function *emitOutlinedOrderedFunction(CodeGenModule &CGM,
 }
 
 void CodeGenFunction::EmitOMPOrderedDirective(const OMPOrderedDirective &S) {
-  if (!S.getAssociatedStmt()) {
+  if (S.hasClausesOfKind<OMPDependClause>()) {
+    assert(!S.getAssociatedStmt() &&
+           "No associated statement must be in ordered depend construct.");
     for (const auto *DC : S.getClausesOfKind<OMPDependClause>())
       CGM.getOpenMPRuntime().emitDoacrossOrdered(*this, DC);
     return;
@@ -3421,8 +3425,8 @@ void CodeGenFunction::EmitOMPOrderedDirective(const OMPOrderedDirective &S) {
   auto *C = S.getSingleClause<OMPSIMDClause>();
   auto &&CodeGen = [&S, C, this](CodeGenFunction &CGF,
                                  PrePostActionTy &Action) {
+    const CapturedStmt *CS = S.getInnermostCapturedStmt();
     if (C) {
-      auto CS = cast<CapturedStmt>(S.getAssociatedStmt());
       llvm::SmallVector<llvm::Value *, 16> CapturedVars;
       CGF.GenerateOpenMPCapturedVars(*CS, CapturedVars);
       auto *OutlinedFn = emitOutlinedOrderedFunction(CGM, CS);
@@ -3430,11 +3434,10 @@ void CodeGenFunction::EmitOMPOrderedDirective(const OMPOrderedDirective &S) {
                                                       OutlinedFn, CapturedVars);
     } else {
       Action.Enter(CGF);
-      CGF.EmitStmt(
-          cast<CapturedStmt>(S.getAssociatedStmt())->getCapturedStmt());
+      CGF.EmitStmt(CS->getCapturedStmt());
     }
   };
-  OMPLexicalScope Scope(*this, S, /*AsInlined=*/true);
+  OMPLexicalScope Scope(*this, S, OMPD_unknown);
   CGM.getOpenMPRuntime().emitOrderedRegion(*this, CodeGen, S.getLocStart(), !C);
 }
 
@@ -3878,8 +3881,7 @@ void CodeGenFunction::EmitOMPAtomicDirective(const OMPAtomicDirective &S) {
     }
   }
 
-  const auto *CS =
-      S.getAssociatedStmt()->IgnoreContainers(/*IgnoreCaptured=*/true);
+  const auto *CS = S.getInnermostCapturedStmt()->IgnoreContainers();
   if (const auto *EWC = dyn_cast<ExprWithCleanups>(CS)) {
     enterFullExpression(EWC);
   }
@@ -3899,7 +3901,7 @@ void CodeGenFunction::EmitOMPAtomicDirective(const OMPAtomicDirective &S) {
                       S.getV(), S.getExpr(), S.getUpdateExpr(),
                       S.isXLHSInRHSPart(), S.getLocStart());
   };
-  OMPLexicalScope Scope(*this, S, /*AsInlined=*/true);
+  OMPLexicalScope Scope(*this, S, OMPD_unknown);
   CGM.getOpenMPRuntime().emitInlinedDirective(*this, OMPD_atomic, CodeGen);
 }
 
@@ -3971,7 +3973,7 @@ static void emitTargetRegion(CodeGenFunction &CGF, const OMPTargetDirective &S,
   (void)PrivateScope.Privatize();
 
   Action.Enter(CGF);
-  CGF.EmitStmt(cast<CapturedStmt>(S.getAssociatedStmt())->getCapturedStmt());
+  CGF.EmitStmt(S.getCapturedStmt(OMPD_target)->getCapturedStmt());
 }
 
 void CodeGenFunction::EmitOMPTargetDeviceFunction(CodeGenModule &CGM,
@@ -4028,7 +4030,7 @@ void CodeGenFunction::EmitOMPTeamsDirective(const OMPTeamsDirective &S) {
     CGF.EmitOMPPrivateClause(S, PrivateScope);
     CGF.EmitOMPReductionClauseInit(S, PrivateScope);
     (void)PrivateScope.Privatize();
-    CGF.EmitStmt(cast<CapturedStmt>(S.getAssociatedStmt())->getCapturedStmt());
+    CGF.EmitStmt(S.getCapturedStmt(OMPD_teams)->getCapturedStmt());
     CGF.EmitOMPReductionClauseFinal(S, /*ReductionKind=*/OMPD_teams);
   };
   emitCommonOMPTeamsDirective(*this, S, OMPD_distribute, CodeGen);
@@ -4421,10 +4423,9 @@ void CodeGenFunction::EmitOMPTargetDataDirective(
   DevicePointerPrivActionTy PrivAction(PrivatizeDevicePointers);
 
   auto &&CodeGen = [&S, &Info, &PrivatizeDevicePointers](
-      CodeGenFunction &CGF, PrePostActionTy &Action) {
+                       CodeGenFunction &CGF, PrePostActionTy &Action) {
     auto &&InnermostCodeGen = [&S](CodeGenFunction &CGF, PrePostActionTy &) {
-      CGF.EmitStmt(
-          cast<CapturedStmt>(S.getAssociatedStmt())->getCapturedStmt());
+      CGF.EmitStmt(S.getInnermostCapturedStmt()->getCapturedStmt());
     };
 
     // Codegen that selects wheather to generate the privatization code or not.
@@ -4506,7 +4507,7 @@ void CodeGenFunction::EmitOMPTargetEnterDataDirective(
   if (auto *C = S.getSingleClause<OMPDeviceClause>())
     Device = C->getDevice();
 
-  OMPLexicalScope Scope(*this, S, /*AsInlined=*/true);
+  OMPLexicalScope Scope(*this, S, OMPD_task);
   CGM.getOpenMPRuntime().emitTargetDataStandAloneCall(*this, S, IfCond, Device);
 }
 
@@ -4527,7 +4528,7 @@ void CodeGenFunction::EmitOMPTargetExitDataDirective(
   if (auto *C = S.getSingleClause<OMPDeviceClause>())
     Device = C->getDevice();
 
-  OMPLexicalScope Scope(*this, S, /*AsInlined=*/true);
+  OMPLexicalScope Scope(*this, S, OMPD_task);
   CGM.getOpenMPRuntime().emitTargetDataStandAloneCall(*this, S, IfCond, Device);
 }
 
@@ -4664,7 +4665,7 @@ static void mapParam(CodeGenFunction &CGF, const DeclRefExpr *Helper,
 void CodeGenFunction::EmitOMPTaskLoopBasedDirective(const OMPLoopDirective &S) {
   assert(isOpenMPTaskLoopDirective(S.getDirectiveKind()));
   // Emit outlined function for task construct.
-  auto CS = cast<CapturedStmt>(S.getAssociatedStmt());
+  const CapturedStmt *CS = S.getCapturedStmt(OMPD_taskloop);
   auto CapturedStruct = GenerateCapturedStmtArgument(*CS);
   auto SharedsTy = getContext().getRecordType(CS->getCapturedRecordDecl());
   const Expr *IfCond = nullptr;
@@ -4786,15 +4787,16 @@ void CodeGenFunction::EmitOMPTaskLoopBasedDirective(const OMPLoopDirective &S) {
     CGF.CGM.getOpenMPRuntime().emitInlinedDirective(CGF, OMPD_taskloop,
                                                     CodeGen);
   };
-  if (Data.Nogroup)
-    EmitOMPTaskBasedDirective(S, BodyGen, TaskGen, Data);
-  else {
+  if (Data.Nogroup) {
+    EmitOMPTaskBasedDirective(S, OMPD_taskloop, BodyGen, TaskGen, Data);
+  } else {
     CGM.getOpenMPRuntime().emitTaskgroupRegion(
         *this,
         [&S, &BodyGen, &TaskGen, &Data](CodeGenFunction &CGF,
                                         PrePostActionTy &Action) {
           Action.Enter(CGF);
-          CGF.EmitOMPTaskBasedDirective(S, BodyGen, TaskGen, Data);
+          CGF.EmitOMPTaskBasedDirective(S, OMPD_taskloop, BodyGen, TaskGen,
+                                        Data);
         },
         S.getLocStart());
   }
@@ -4827,7 +4829,7 @@ void CodeGenFunction::EmitOMPTargetUpdateDirective(
   if (auto *C = S.getSingleClause<OMPDeviceClause>())
     Device = C->getDevice();
 
-  OMPLexicalScope Scope(*this, S, /*AsInlined=*/true);
+  OMPLexicalScope Scope(*this, S, OMPD_task);
   CGM.getOpenMPRuntime().emitTargetDataStandAloneCall(*this, S, IfCond, Device);
 }
 
@@ -4849,10 +4851,7 @@ void CodeGenFunction::EmitSimpleOMPExecutableDirective(
           }
         }
       }
-      const auto *CS = cast<CapturedStmt>(D.getAssociatedStmt());
-      while (const auto *CCS = dyn_cast<CapturedStmt>(CS->getCapturedStmt()))
-        CS = CCS;
-      CGF.EmitStmt(CS->getCapturedStmt());
+      CGF.EmitStmt(D.getInnermostCapturedStmt()->getCapturedStmt());
     }
   };
   OMPSimdLexicalScope Scope(*this, D);
diff --git a/clang/lib/CodeGen/CodeGenFunction.h b/clang/lib/CodeGen/CodeGenFunction.h
index cedf327a480..228cbe69acb 100644
--- a/clang/lib/CodeGen/CodeGenFunction.h
+++ b/clang/lib/CodeGen/CodeGenFunction.h
@@ -2832,6 +2832,7 @@ public:
                                         const OMPTaskDataTy & /*Data*/)>
       TaskGenTy;
   void EmitOMPTaskBasedDirective(const OMPExecutableDirective &S,
+                                 const OpenMPDirectiveKind CapturedRegion,
                                  const RegionCodeGenTy &BodyGen,
                                  const TaskGenTy &TaskGen, OMPTaskDataTy &Data);
   struct OMPTargetDataInfo {
diff --git a/clang/lib/Sema/TreeTransform.h b/clang/lib/Sema/TreeTransform.h
index 96969ea87a1..ebfdbe83ef8 100644
--- a/clang/lib/Sema/TreeTransform.h
+++ b/clang/lib/Sema/TreeTransform.h
@@ -7642,11 +7642,7 @@ StmtResult TreeTransform<Derived>::TransformOMPExecutableDirective(
     StmtResult Body;
     {
       Sema::CompoundScopeRAII CompoundScope(getSema());
-      int ThisCaptureLevel =
-          Sema::getOpenMPCaptureLevels(D->getDirectiveKind());
-      Stmt *CS = D->getAssociatedStmt();
-      while (--ThisCaptureLevel >= 0)
-        CS = cast<CapturedStmt>(CS)->getCapturedStmt();
+      Stmt *CS = D->getInnermostCapturedStmt()->getCapturedStmt();
       Body = getDerived().TransformStmt(CS);
     }
     AssociatedStmt =
diff --git a/clang/test/OpenMP/nvptx_target_teams_codegen.cpp b/clang/test/OpenMP/nvptx_target_teams_codegen.cpp
index 8b016cc307d..ff5967b5612 100644
--- a/clang/test/OpenMP/nvptx_target_teams_codegen.cpp
+++ b/clang/test/OpenMP/nvptx_target_teams_codegen.cpp
@@ -121,9 +121,7 @@ int bar(int n){
   // CHECK: [[ACV:%.+]] = load i[[SZ]], i[[SZ]]* [[AC]], align
   // CHECK: store i[[SZ]] [[ACV]], i[[SZ]]* [[A_ADDR_T:%.+]], align
   // CHECK: [[CONV2:%.+]] = bitcast i[[SZ]]* [[A_ADDR_T]] to i8*
-  // CHECK: [[LD_CONV2:%.+]] = load i8, i8* [[CONV2]],
-  // CHECK: store i8 [[LD_CONV2]], i8* [[A_PRIV:%[^,]+]],
-  // CHECK: store i8 49, i8* [[A_PRIV]], align
+  // CHECK: store i8 49, i8* [[CONV2]], align
   // CHECK: br label {{%?}}[[TERMINATE:.+]]
   //
   // CHECK: [[TERMINATE]]
@@ -209,9 +207,7 @@ int bar(int n){
   // CHECK: [[ACV:%.+]] = load i[[SZ]], i[[SZ]]* [[AC]], align
   // CHECK: store i[[SZ]] [[ACV]], i[[SZ]]* [[AA_ADDR_T:%.+]], align
   // CHECK: [[CONV2:%.+]] = bitcast i[[SZ]]* [[AA_ADDR_T]] to i16*
-  // CHECK: [[LD_CONV2:%.+]] = load i16, i16* [[CONV2]],
-  // CHECK: store i16 [[LD_CONV2]], i16* [[A_PRIV:%[^,]+]],
-  // CHECK: store i16 1, i16* [[A_PRIV]], align
+  // CHECK: store i16 1, i16* [[CONV2]], align
   // CHECK: br label {{%?}}[[TERMINATE:.+]]
   //
   // CHECK: [[TERMINATE]]
diff --git a/clang/test/OpenMP/target_teams_codegen.cpp b/clang/test/OpenMP/target_teams_codegen.cpp
index f3cc9352078..4acc149053f 100644
--- a/clang/test/OpenMP/target_teams_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_codegen.cpp
@@ -345,7 +345,8 @@ int foo(int n) {
 //
 //
 // CHECK:       define internal {{.*}}void [[OMP_OUTLINED]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] {{[^)]+}})
-// CHECK:       alloca i16,
+// CHECK:       alloca i[[SZ]],
+// CHECK:       bitcast i[[SZ]]* {{.+}} to i16*
 // CHECK:       ret void
 // CHECK-NEXT:  }
 
diff --git a/clang/test/OpenMP/target_teams_distribute_codegen.cpp b/clang/test/OpenMP/target_teams_distribute_codegen.cpp
index 33e4288905f..aff7326cbe4 100644
--- a/clang/test/OpenMP/target_teams_distribute_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_distribute_codegen.cpp
@@ -355,7 +355,8 @@ int foo(int n) {
 //
 //
 // CHECK:       define internal {{.*}}void [[OMP_OUTLINED]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] {{[^)]+}})
-// CHECK:       alloca i16,
+// CHECK:       alloca i[[SZ]],
+// CHECK:       bitcast i[[SZ]]* {{.+}} to i16*
 // CHECK:       ret void
 // CHECK-NEXT:  }
 
diff --git a/clang/test/OpenMP/target_teams_distribute_firstprivate_codegen.cpp b/clang/test/OpenMP/target_teams_distribute_firstprivate_codegen.cpp
index a9f64f384fa..d5b6b336a08 100644
--- a/clang/test/OpenMP/target_teams_distribute_firstprivate_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_distribute_firstprivate_codegen.cpp
@@ -110,16 +110,7 @@ int main() {
     // LAMBDA: [[G_ADDR:%.+]] = alloca i{{[0-9]+}},
     // LAMBDA: [[G1_ADDR:%.+]] = alloca i{{[0-9]+}},
     // LAMBDA: [[SIVAR_ADDR:%.+]] = alloca i{{[0-9]+}},
-    // LAMBDA: {{%.+}} = alloca i{{[0-9]+}},
-    // LAMBDA: {{%.+}} = alloca i{{[0-9]+}},
-    // LAMBDA: {{%.+}} = alloca i{{[0-9]+}},
-    // LAMBDA: {{%.+}} = alloca i{{[0-9]+}},
-    // LAMBDA: {{%.+}} = alloca i{{[0-9]+}},
-    // LAMBDA: {{%.+}} = alloca i{{[0-9]+}},
-    // LAMBDA: [[G_PRIV_ADDR:%.+]] = alloca i{{[0-9]+}},
-    // LAMBDA: [[G1_PRIV_ADDR:%.+]] = alloca i{{[0-9]+}},
     // LAMBDA: [[G1_TMP:%.+]] = alloca i32*,
-    // LAMBDA: [[SIVAR_PRIV_ADDR:%.+]] = alloca i{{[0-9]+}},
     // skip loop vars
     // LAMBDA-DAG: store {{.+}}, {{.+}} [[G_ADDR]],
     // LAMBDA-DAG: store {{.+}}, {{.+}} [[G1_ADDR]],
@@ -127,15 +118,15 @@ int main() {
     // LAMBDA-DAG: [[G_CONV:%.+]] = bitcast {{.+}} [[G_ADDR]] to
     // LAMBDA-DAG: [[G1_CONV:%.+]] = bitcast {{.+}} [[G1_ADDR]] to
     // LAMBDA-DAG: [[SIVAR_CONV:%.+]] = bitcast {{.+}} [[SIVAR_ADDR]] to
-    // LAMBDA-DAG: store{{.+}} [[G1_PRIV_ADDR]], {{.+}} [[G1_TMP]],
+    // LAMBDA-DAG: store{{.+}} [[G1_CONV]], {{.+}} [[G1_TMP]],
     g = 1;
     g1 = 1;
     sivar = 2;
     // LAMBDA: call void @__kmpc_for_static_init_4(
-    // LAMBDA-DAG: store{{.+}} 1, {{.+}} [[G_PRIV_ADDR]],
+    // LAMBDA-DAG: store{{.+}} 1, {{.+}} [[G_CONV]],
     // LAMBDA-DAG: [[G1:%.+]] = load{{.+}}, {{.+}}* [[G1_TMP]]
     // LAMBDA-DAG: store{{.+}} 1, {{.+}} [[G1]],
-    // LAMBDA-DAG: store{{.+}} 2, {{.+}} [[SIVAR_PRIV_ADDR]],
+    // LAMBDA-DAG: store{{.+}} 2, {{.+}} [[SIVAR_CONV]],
     // LAMBDA-DAG: [[G1_REF:%.+]] = load{{.+}}, {{.+}} [[G1_TMP]],
     // LAMBDA-DAG: store{{.+}} 1, {{.+}} [[G1_REF]],
     // LAMBDA: call void [[INNER_LAMBDA:@.+]](
diff --git a/clang/test/OpenMP/target_teams_distribute_parallel_for_firstprivate_codegen.cpp b/clang/test/OpenMP/target_teams_distribute_parallel_for_firstprivate_codegen.cpp
index df887662199..c17de1a8954 100644
--- a/clang/test/OpenMP/target_teams_distribute_parallel_for_firstprivate_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_distribute_parallel_for_firstprivate_codegen.cpp
@@ -163,16 +163,6 @@ int main() {
     // LAMBDA: [[SIVAR_ADDR:%.+]] = alloca i{{[0-9]+}},
     // LAMBDA: [[G1_TMP:%.+]] = alloca i32*,
     // skip loop vars
-    // LAMBDA: alloca i32,
-    // LAMBDA: alloca i32,
-    // LAMBDA: alloca i32,
-    // LAMBDA: alloca i32,
-    // LAMBDA: alloca i32,
-    // LAMBDA: alloca i32,
-    // LAMBDA: [[G_PRIV:%.+]] = alloca i{{[0-9]+}},
-    // LAMBDA: [[G1_PRIV:%.+]] = alloca i{{[0-9]+}},
-    // LAMBDA: [[G1_TMP_PRIV:%.+]] = alloca i{{[0-9]+}}*,
-    // LAMBDA: [[SIVAR_PRIV:%.+]] = alloca i{{[0-9]+}},
     // LAMBDA-DAG: store {{.+}}, {{.+}} [[G_ADDR]],
     // LAMBDA-DAG: store {{.+}}, {{.+}} [[G1_ADDR]],
     // LAMBDA-DAG: store {{.+}}, {{.+}} [[SIVAR_ADDR]],
@@ -182,10 +172,10 @@ int main() {
     // LAMBDA-DAG: store{{.+}} [[G1_CONV]], {{.+}} [[G1_TMP]],
 
     // use of private vars
-    // LAMBDA-DAG: store{{.+}} 1, {{.+}} [[G_PRIV]],
-    // LAMBDA-DAG: [[G1:%.+]] = load{{.+}}, {{.+}}* [[G1_TMP_PRIV]]
+    // LAMBDA-DAG: store{{.+}} 1, {{.+}} [[G_CONV]],
+    // LAMBDA-DAG: [[G1:%.+]] = load{{.+}}, {{.+}}* [[G1_TMP]]
     // LAMBDA-DAG: store{{.+}} 1, {{.+}} [[G1]],
-    // LAMBDA-DAG: store{{.+}} 2, {{.+}} [[SIVAR_PRIV]],
+    // LAMBDA-DAG: store{{.+}} 2, {{.+}} [[SIVAR_CONV]],
     // LAMBDA-DAG: [[G1_REF:%.+]] = load{{.+}}, {{.+}} [[G1_TMP]],
     // LAMBDA: call void [[INNER_LAMBDA:@.+]](
     // LAMBDA: call void @__kmpc_for_static_fini(
diff --git a/clang/test/OpenMP/target_teams_distribute_simd_codegen.cpp b/clang/test/OpenMP/target_teams_distribute_simd_codegen.cpp
index 3ddc6044051..37a35ffdb87 100644
--- a/clang/test/OpenMP/target_teams_distribute_simd_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_distribute_simd_codegen.cpp
@@ -345,7 +345,8 @@ int foo(int n) {
 //
 //
 // CHECK:       define internal {{.*}}void [[OMP_OUTLINED]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] {{[^)]+}})
-// CHECK:       alloca i16,
+// CHECK:       alloca i[[SZ]],
+// CHECK:       bitcast i[[SZ]]* {{.+}} to i16*
 // CHECK:       ret void
 // CHECK-NEXT:  }
 
diff --git a/clang/test/OpenMP/target_teams_distribute_simd_firstprivate_codegen.cpp b/clang/test/OpenMP/target_teams_distribute_simd_firstprivate_codegen.cpp
index 8074fba7360..23579580ee6 100644
--- a/clang/test/OpenMP/target_teams_distribute_simd_firstprivate_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_distribute_simd_firstprivate_codegen.cpp
@@ -110,16 +110,7 @@ int main() {
     // LAMBDA: [[G_ADDR:%.+]] = alloca i{{[0-9]+}},
     // LAMBDA: [[G1_ADDR:%.+]] = alloca i{{[0-9]+}},
     // LAMBDA: [[SIVAR_ADDR:%.+]] = alloca i{{[0-9]+}},
-    // LAMBDA: {{%.+}} = alloca i{{[0-9]+}},
-    // LAMBDA: {{%.+}} = alloca i{{[0-9]+}},
-    // LAMBDA: {{%.+}} = alloca i{{[0-9]+}},
-    // LAMBDA: {{%.+}} = alloca i{{[0-9]+}},
-    // LAMBDA: {{%.+}} = alloca i{{[0-9]+}},
-    // LAMBDA: {{%.+}} = alloca i{{[0-9]+}},
-    // LAMBDA: [[G_PRIV_ADDR:%.+]] = alloca i{{[0-9]+}},
-    // LAMBDA: [[G1_PRIV_ADDR:%.+]] = alloca i{{[0-9]+}},
     // LAMBDA: [[G1_TMP:%.+]] = alloca i32*,
-    // LAMBDA: [[SIVAR_PRIV_ADDR:%.+]] = alloca i{{[0-9]+}},
     // skip loop vars
     // LAMBDA-DAG: store {{.+}}, {{.+}} [[G_ADDR]],
     // LAMBDA-DAG: store {{.+}}, {{.+}} [[G1_ADDR]],
@@ -127,15 +118,15 @@ int main() {
     // LAMBDA-DAG: [[G_CONV:%.+]] = bitcast {{.+}} [[G_ADDR]] to
     // LAMBDA-DAG: [[G1_CONV:%.+]] = bitcast {{.+}} [[G1_ADDR]] to
     // LAMBDA-DAG: [[SIVAR_CONV:%.+]] = bitcast {{.+}} [[SIVAR_ADDR]] to
-    // LAMBDA-DAG: store{{.+}} [[G1_PRIV_ADDR]], {{.+}} [[G1_TMP]],
+    // LAMBDA-DAG: store{{.+}} [[G1_CONV]], {{.+}} [[G1_TMP]],
     g = 1;
     g1 = 1;
     sivar = 2;
     // LAMBDA: call void @__kmpc_for_static_init_4(
-    // LAMBDA-DAG: store{{.+}} 1, {{.+}} [[G_PRIV_ADDR]],
+    // LAMBDA-DAG: store{{.+}} 1, {{.+}} [[G_CONV]],
     // LAMBDA-DAG: [[G1:%.+]] = load{{.+}}, {{.+}}* [[G1_TMP]]
     // LAMBDA-DAG: store{{.+}} 1, {{.+}} [[G1]],
-    // LAMBDA-DAG: store{{.+}} 2, {{.+}} [[SIVAR_PRIV_ADDR]],
+    // LAMBDA-DAG: store{{.+}} 2, {{.+}} [[SIVAR_CONV]],
     // LAMBDA-DAG: [[G1_REF:%.+]] = load{{.+}}, {{.+}} [[G1_TMP]],
     // LAMBDA-DAG: store{{.+}} 1, {{.+}} [[G1_REF]],
     // LAMBDA: call void [[INNER_LAMBDA:@.+]](
diff --git a/clang/test/OpenMP/teams_distribute_parallel_for_firstprivate_codegen.cpp b/clang/test/OpenMP/teams_distribute_parallel_for_firstprivate_codegen.cpp
index 796642707df..578c3cc65b1 100644
--- a/clang/test/OpenMP/teams_distribute_parallel_for_firstprivate_codegen.cpp
+++ b/clang/test/OpenMP/teams_distribute_parallel_for_firstprivate_codegen.cpp
@@ -140,16 +140,6 @@ int main() {
     // LAMBDA: [[SIVAR_ADDR:%.+]] = alloca i{{[0-9]+}},
     // LAMBDA: [[G1_TMP:%.+]] = alloca i32*,
     // skip loop vars
-    // LAMBDA: alloca i32,
-    // LAMBDA: alloca i32,
-    // LAMBDA: alloca i32,
-    // LAMBDA: alloca i32,
-    // LAMBDA: alloca i32,
-    // LAMBDA: alloca i32,
-    // LAMBDA: [[G_PRIV:%.+]] = alloca i{{[0-9]+}},
-    // LAMBDA: [[G1_PRIV:%.+]] = alloca i{{[0-9]+}},
-    // LAMBDA: [[G1_TMP_PRIV:%.+]] = alloca i{{[0-9]+}}*,
-    // LAMBDA: [[SIVAR_PRIV:%.+]] = alloca i{{[0-9]+}},
     // LAMBDA-DAG: store {{.+}}, {{.+}} [[G_ADDR]],
     // LAMBDA-DAG: store {{.+}}, {{.+}} [[G1_ADDR]],
     // LAMBDA-DAG: store {{.+}}, {{.+}} [[SIVAR_ADDR]],
@@ -159,10 +149,10 @@ int main() {
     // LAMBDA-DAG: store{{.+}} [[G1_CONV]], {{.+}} [[G1_TMP]],
 
     // use of private vars
-    // LAMBDA-DAG: store{{.+}} 1, {{.+}} [[G_PRIV]],
-    // LAMBDA-DAG: [[G1:%.+]] = load{{.+}}, {{.+}}* [[G1_TMP_PRIV]]
+    // LAMBDA-DAG: store{{.+}} 1, {{.+}} [[G_CONV]],
+    // LAMBDA-DAG: [[G1:%.+]] = load{{.+}}, {{.+}}* [[G1_TMP]]
     // LAMBDA-DAG: store{{.+}} 1, {{.+}} [[G1]],
-    // LAMBDA-DAG: store{{.+}} 2, {{.+}} [[SIVAR_PRIV]],
+    // LAMBDA-DAG: store{{.+}} 2, {{.+}} [[SIVAR_CONV]],
     // LAMBDA-DAG: [[G1_REF:%.+]] = load{{.+}}, {{.+}} [[G1_TMP]],
     // LAMBDA: call void [[INNER_LAMBDA:@.+]](
     // LAMBDA: call void @__kmpc_for_static_fini(
diff --git a/clang/test/OpenMP/teams_distribute_parallel_for_simd_firstprivate_codegen.cpp b/clang/test/OpenMP/teams_distribute_parallel_for_simd_firstprivate_codegen.cpp
index 1b996b38ffe..711fd2eaf21 100644
--- a/clang/test/OpenMP/teams_distribute_parallel_for_simd_firstprivate_codegen.cpp
+++ b/clang/test/OpenMP/teams_distribute_parallel_for_simd_firstprivate_codegen.cpp
@@ -140,16 +140,6 @@ int main() {
     // LAMBDA: [[SIVAR_ADDR:%.+]] = alloca i{{[0-9]+}},
     // LAMBDA: [[G1_TMP:%.+]] = alloca i32*,
     // skip loop vars
-    // LAMBDA: alloca i32,
-    // LAMBDA: alloca i32,
-    // LAMBDA: alloca i32,
-    // LAMBDA: alloca i32,
-    // LAMBDA: alloca i32,
-    // LAMBDA: alloca i32,
-    // LAMBDA: [[G_PRIV:%.+]] = alloca i{{[0-9]+}},
-    // LAMBDA: [[G1_PRIV:%.+]] = alloca i{{[0-9]+}},
-    // LAMBDA: [[G1_TMP_PRIV:%.+]] = alloca i{{[0-9]+}}*,
-    // LAMBDA: [[SIVAR_PRIV:%.+]] = alloca i{{[0-9]+}},
     // LAMBDA-DAG: store {{.+}}, {{.+}} [[G_ADDR]],
     // LAMBDA-DAG: store {{.+}}, {{.+}} [[G1_ADDR]],
     // LAMBDA-DAG: store {{.+}}, {{.+}} [[SIVAR_ADDR]],
@@ -159,10 +149,10 @@ int main() {
     // LAMBDA-DAG: store{{.+}} [[G1_CONV]], {{.+}} [[G1_TMP]],
 
     // use of private vars
-    // LAMBDA-DAG: store{{.+}} 1, {{.+}} [[G_PRIV]],
-    // LAMBDA-DAG: [[G1:%.+]] = load{{.+}}, {{.+}}* [[G1_TMP_PRIV]]
+    // LAMBDA-DAG: store{{.+}} 1, {{.+}} [[G_CONV]],
+    // LAMBDA-DAG: [[G1:%.+]] = load{{.+}}, {{.+}}* [[G1_TMP]]
     // LAMBDA-DAG: store{{.+}} 1, {{.+}} [[G1]],
-    // LAMBDA-DAG: store{{.+}} 2, {{.+}} [[SIVAR_PRIV]],
+    // LAMBDA-DAG: store{{.+}} 2, {{.+}} [[SIVAR_CONV]],
     // LAMBDA-DAG: [[G1_REF:%.+]] = load{{.+}}, {{.+}} [[G1_TMP]],
     // LAMBDA: call void [[INNER_LAMBDA:@.+]](
     // LAMBDA: call void @__kmpc_for_static_fini(