[llvm-branch-commits] [clang] [llvm] [openmp] [Clang][OpenMP] Add reverse directive (PR #92916)

Michael Kruse via llvm-branch-commits llvm-branch-commits at lists.llvm.org
Tue May 21 08:50:28 PDT 2024


https://github.com/Meinersbur updated https://github.com/llvm/llvm-project/pull/92916

>From b15caff37dfd8a30551d5a250969975e720ffa33 Mon Sep 17 00:00:00 2001
From: Michael Kruse <llvm-project at meinersbur.de>
Date: Tue, 21 May 2024 15:33:47 +0200
Subject: [PATCH 1/2] [Clang][OpenMP] Add reverse directive

---
 clang/include/clang-c/Index.h                 |    4 +
 clang/include/clang/AST/RecursiveASTVisitor.h |    3 +
 clang/include/clang/AST/StmtOpenMP.h          |   72 +-
 clang/include/clang/Basic/StmtNodes.td        |    1 +
 clang/include/clang/Sema/SemaOpenMP.h         |    5 +
 .../include/clang/Serialization/ASTBitCodes.h |    1 +
 clang/lib/AST/StmtOpenMP.cpp                  |   19 +
 clang/lib/AST/StmtPrinter.cpp                 |    5 +
 clang/lib/AST/StmtProfile.cpp                 |    4 +
 clang/lib/Basic/OpenMPKinds.cpp               |    3 +-
 clang/lib/CodeGen/CGStmt.cpp                  |    3 +
 clang/lib/CodeGen/CGStmtOpenMP.cpp            |    8 +
 clang/lib/CodeGen/CodeGenFunction.h           |    1 +
 clang/lib/Parse/ParseOpenMP.cpp               |    2 +
 clang/lib/Sema/SemaExceptionSpec.cpp          |    1 +
 clang/lib/Sema/SemaOpenMP.cpp                 |  190 ++
 clang/lib/Sema/TreeTransform.h                |   11 +
 clang/lib/Serialization/ASTReaderStmt.cpp     |   12 +
 clang/lib/Serialization/ASTWriterStmt.cpp     |    5 +
 clang/test/OpenMP/reverse_ast_print.cpp       |  159 ++
 clang/test/OpenMP/reverse_codegen.cpp         | 1554 +++++++++++++++++
 clang/test/OpenMP/reverse_messages.cpp        |   40 +
 clang/tools/libclang/CIndex.cpp               |    7 +
 clang/tools/libclang/CXCursor.cpp             |    3 +
 llvm/include/llvm/Frontend/OpenMP/OMP.td      |    3 +
 .../test/transform/reverse/foreach.cpp        |  162 ++
 .../runtime/test/transform/reverse/intfor.c   |   25 +
 .../test/transform/reverse/iterfor.cpp        |  164 ++
 .../parallel-wsloop-collapse-foreach.cpp      |  285 +++
 .../parallel-wsloop-collapse-intfor.cpp       |   51 +
 30 files changed, 2800 insertions(+), 3 deletions(-)
 create mode 100644 clang/test/OpenMP/reverse_ast_print.cpp
 create mode 100644 clang/test/OpenMP/reverse_codegen.cpp
 create mode 100644 clang/test/OpenMP/reverse_messages.cpp
 create mode 100644 openmp/runtime/test/transform/reverse/foreach.cpp
 create mode 100644 openmp/runtime/test/transform/reverse/intfor.c
 create mode 100644 openmp/runtime/test/transform/reverse/iterfor.cpp
 create mode 100644 openmp/runtime/test/transform/reverse/parallel-wsloop-collapse-foreach.cpp
 create mode 100644 openmp/runtime/test/transform/reverse/parallel-wsloop-collapse-intfor.cpp

diff --git a/clang/include/clang-c/Index.h b/clang/include/clang-c/Index.h
index 365b607c74117..c7d63818ece23 100644
--- a/clang/include/clang-c/Index.h
+++ b/clang/include/clang-c/Index.h
@@ -2146,6 +2146,10 @@ enum CXCursorKind {
    */
   CXCursor_OMPScopeDirective = 306,
 
+  /** OpenMP reverse directive.
+   */
+  CXCursor_OMPReverseDirective = 307,
+
   /** OpenACC Compute Construct.
    */
   CXCursor_OpenACCComputeConstruct = 320,
diff --git a/clang/include/clang/AST/RecursiveASTVisitor.h b/clang/include/clang/AST/RecursiveASTVisitor.h
index f5cefedb07e0e..06b29d59785f6 100644
--- a/clang/include/clang/AST/RecursiveASTVisitor.h
+++ b/clang/include/clang/AST/RecursiveASTVisitor.h
@@ -3021,6 +3021,9 @@ DEF_TRAVERSE_STMT(OMPTileDirective,
 DEF_TRAVERSE_STMT(OMPUnrollDirective,
                   { TRY_TO(TraverseOMPExecutableDirective(S)); })
 
+DEF_TRAVERSE_STMT(OMPReverseDirective,
+                  { TRY_TO(TraverseOMPExecutableDirective(S)); })
+
 DEF_TRAVERSE_STMT(OMPForDirective,
                   { TRY_TO(TraverseOMPExecutableDirective(S)); })
 
diff --git a/clang/include/clang/AST/StmtOpenMP.h b/clang/include/clang/AST/StmtOpenMP.h
index f735fa5643aec..4be2e2d3a4605 100644
--- a/clang/include/clang/AST/StmtOpenMP.h
+++ b/clang/include/clang/AST/StmtOpenMP.h
@@ -1007,8 +1007,9 @@ class OMPLoopTransformationDirective : public OMPLoopBasedDirective {
   Stmt *getPreInits() const;
 
   static bool classof(const Stmt *T) {
-    return T->getStmtClass() == OMPTileDirectiveClass ||
-           T->getStmtClass() == OMPUnrollDirectiveClass;
+    Stmt::StmtClass C = T->getStmtClass();
+    return C == OMPTileDirectiveClass || C == OMPUnrollDirectiveClass ||
+           C == OMPReverseDirectiveClass;
   }
 };
 
@@ -5711,6 +5712,73 @@ class OMPUnrollDirective final : public OMPLoopTransformationDirective {
   }
 };
 
+/// Represents the '#pragma omp reverse' loop transformation directive.
+///
+/// \code
+/// #pragma omp reverse
+/// for (int i = 0; i < n; ++i)
+///   ...
+/// \endcode
+class OMPReverseDirective final : public OMPLoopTransformationDirective {
+  friend class ASTStmtReader;
+  friend class OMPExecutableDirective;
+
+  /// Offsets of child members.
+  enum {
+    PreInitsOffset = 0,
+    TransformedStmtOffset,
+  };
+
+  explicit OMPReverseDirective(SourceLocation StartLoc, SourceLocation EndLoc)
+      : OMPLoopTransformationDirective(OMPReverseDirectiveClass,
+                                       llvm::omp::OMPD_reverse, StartLoc,
+                                       EndLoc, 1) {}
+
+  void setPreInits(Stmt *PreInits) {
+    Data->getChildren()[PreInitsOffset] = PreInits;
+  }
+
+  void setTransformedStmt(Stmt *S) {
+    Data->getChildren()[TransformedStmtOffset] = S;
+  }
+
+public:
+  /// Create a new AST node representation for '#pragma omp reverse'.
+  ///
+  /// \param C         Context of the AST.
+  /// \param StartLoc  Location of the introducer (e.g. the 'omp' token).
+  /// \param EndLoc    Location of the directive's end (e.g. the tok::eod).
+  /// \param Clauses   The directive's clauses.
+  /// \param AssociatedStmt  The outermost associated loop.
+  /// \param TransformedStmt The loop nest after tiling, or nullptr in
+  ///                        dependent contexts.
+  /// \param PreInits   Helper preinits statements for the loop nest.
+  static OMPReverseDirective *
+  Create(const ASTContext &C, SourceLocation StartLoc, SourceLocation EndLoc,
+         ArrayRef<OMPClause *> Clauses, Stmt *AssociatedStmt,
+         Stmt *TransformedStmt, Stmt *PreInits);
+
+  /// Build an empty '#pragma omp reverse' AST node for deserialization.
+  ///
+  /// \param C          Context of the AST.
+  /// \param NumClauses Number of clauses to allocate.
+  static OMPReverseDirective *CreateEmpty(const ASTContext &C,
+                                          unsigned NumClauses);
+
+  /// Gets/sets the associated loops after the transformation, i.e. after
+  /// de-sugaring.
+  Stmt *getTransformedStmt() const {
+    return Data->getChildren()[TransformedStmtOffset];
+  }
+
+  /// Return preinits statement.
+  Stmt *getPreInits() const { return Data->getChildren()[PreInitsOffset]; }
+
+  static bool classof(const Stmt *T) {
+    return T->getStmtClass() == OMPReverseDirectiveClass;
+  }
+};
+
 /// This represents '#pragma omp scan' directive.
 ///
 /// \code
diff --git a/clang/include/clang/Basic/StmtNodes.td b/clang/include/clang/Basic/StmtNodes.td
index 305f19daa4a92..b2e2be5c998bb 100644
--- a/clang/include/clang/Basic/StmtNodes.td
+++ b/clang/include/clang/Basic/StmtNodes.td
@@ -229,6 +229,7 @@ def OMPSimdDirective : StmtNode<OMPLoopDirective>;
 def OMPLoopTransformationDirective : StmtNode<OMPLoopBasedDirective, 1>;
 def OMPTileDirective : StmtNode<OMPLoopTransformationDirective>;
 def OMPUnrollDirective : StmtNode<OMPLoopTransformationDirective>;
+def OMPReverseDirective : StmtNode<OMPLoopTransformationDirective>;
 def OMPForDirective : StmtNode<OMPLoopDirective>;
 def OMPForSimdDirective : StmtNode<OMPLoopDirective>;
 def OMPSectionsDirective : StmtNode<OMPExecutableDirective>;
diff --git a/clang/include/clang/Sema/SemaOpenMP.h b/clang/include/clang/Sema/SemaOpenMP.h
index 51981e1c9a8b9..e36a90ba4e1b9 100644
--- a/clang/include/clang/Sema/SemaOpenMP.h
+++ b/clang/include/clang/Sema/SemaOpenMP.h
@@ -422,6 +422,11 @@ class SemaOpenMP : public SemaBase {
   StmtResult ActOnOpenMPUnrollDirective(ArrayRef<OMPClause *> Clauses,
                                         Stmt *AStmt, SourceLocation StartLoc,
                                         SourceLocation EndLoc);
+  /// Called on well-formed '#pragma omp reverse' after parsing of its clauses
+  /// and the associated statement.
+  StmtResult ActOnOpenMPReverseDirective(ArrayRef<OMPClause *> Clauses,
+                                         Stmt *AStmt, SourceLocation StartLoc,
+                                         SourceLocation EndLoc);
   /// Called on well-formed '\#pragma omp for' after parsing
   /// of the associated statement.
   StmtResult
diff --git a/clang/include/clang/Serialization/ASTBitCodes.h b/clang/include/clang/Serialization/ASTBitCodes.h
index fe1bd47348be1..dee0d073557cc 100644
--- a/clang/include/clang/Serialization/ASTBitCodes.h
+++ b/clang/include/clang/Serialization/ASTBitCodes.h
@@ -1856,6 +1856,7 @@ enum StmtCode {
   STMT_OMP_SIMD_DIRECTIVE,
   STMT_OMP_TILE_DIRECTIVE,
   STMT_OMP_UNROLL_DIRECTIVE,
+  STMT_OMP_REVERSE_DIRECTIVE,
   STMT_OMP_FOR_DIRECTIVE,
   STMT_OMP_FOR_SIMD_DIRECTIVE,
   STMT_OMP_SECTIONS_DIRECTIVE,
diff --git a/clang/lib/AST/StmtOpenMP.cpp b/clang/lib/AST/StmtOpenMP.cpp
index d8519b2071e6d..0be0d9d2cfa94 100644
--- a/clang/lib/AST/StmtOpenMP.cpp
+++ b/clang/lib/AST/StmtOpenMP.cpp
@@ -449,6 +449,25 @@ OMPUnrollDirective *OMPUnrollDirective::CreateEmpty(const ASTContext &C,
       SourceLocation(), SourceLocation());
 }
 
+OMPReverseDirective *
+OMPReverseDirective::Create(const ASTContext &C, SourceLocation StartLoc,
+                            SourceLocation EndLoc,
+                            ArrayRef<OMPClause *> Clauses, Stmt *AssociatedStmt,
+                            Stmt *TransformedStmt, Stmt *PreInits) {
+  OMPReverseDirective *Dir = createDirective<OMPReverseDirective>(
+      C, Clauses, AssociatedStmt, TransformedStmtOffset + 1, StartLoc, EndLoc);
+  Dir->setTransformedStmt(TransformedStmt);
+  Dir->setPreInits(PreInits);
+  return Dir;
+}
+
+OMPReverseDirective *OMPReverseDirective::CreateEmpty(const ASTContext &C,
+                                                      unsigned NumClauses) {
+  return createEmptyDirective<OMPReverseDirective>(
+      C, NumClauses, /*HasAssociatedStmt=*/true, TransformedStmtOffset + 1,
+      SourceLocation(), SourceLocation());
+}
+
 OMPForSimdDirective *
 OMPForSimdDirective::Create(const ASTContext &C, SourceLocation StartLoc,
                             SourceLocation EndLoc, unsigned CollapsedNum,
diff --git a/clang/lib/AST/StmtPrinter.cpp b/clang/lib/AST/StmtPrinter.cpp
index be2d5a2eb6b46..64b481f680311 100644
--- a/clang/lib/AST/StmtPrinter.cpp
+++ b/clang/lib/AST/StmtPrinter.cpp
@@ -763,6 +763,11 @@ void StmtPrinter::VisitOMPUnrollDirective(OMPUnrollDirective *Node) {
   PrintOMPExecutableDirective(Node);
 }
 
+void StmtPrinter::VisitOMPReverseDirective(OMPReverseDirective *Node) {
+  Indent() << "#pragma omp reverse";
+  PrintOMPExecutableDirective(Node);
+}
+
 void StmtPrinter::VisitOMPForDirective(OMPForDirective *Node) {
   Indent() << "#pragma omp for";
   PrintOMPExecutableDirective(Node);
diff --git a/clang/lib/AST/StmtProfile.cpp b/clang/lib/AST/StmtProfile.cpp
index caab4ab0ef160..f0e1c9548de72 100644
--- a/clang/lib/AST/StmtProfile.cpp
+++ b/clang/lib/AST/StmtProfile.cpp
@@ -985,6 +985,10 @@ void StmtProfiler::VisitOMPUnrollDirective(const OMPUnrollDirective *S) {
   VisitOMPLoopTransformationDirective(S);
 }
 
+void StmtProfiler::VisitOMPReverseDirective(const OMPReverseDirective *S) {
+  VisitOMPLoopTransformationDirective(S);
+}
+
 void StmtProfiler::VisitOMPForDirective(const OMPForDirective *S) {
   VisitOMPLoopDirective(S);
 }
diff --git a/clang/lib/Basic/OpenMPKinds.cpp b/clang/lib/Basic/OpenMPKinds.cpp
index b3e9affbb3e58..803808c38e2fe 100644
--- a/clang/lib/Basic/OpenMPKinds.cpp
+++ b/clang/lib/Basic/OpenMPKinds.cpp
@@ -684,7 +684,7 @@ bool clang::isOpenMPLoopBoundSharingDirective(OpenMPDirectiveKind Kind) {
 }
 
 bool clang::isOpenMPLoopTransformationDirective(OpenMPDirectiveKind DKind) {
-  return DKind == OMPD_tile || DKind == OMPD_unroll;
+  return DKind == OMPD_tile || DKind == OMPD_unroll || DKind == OMPD_reverse;
 }
 
 bool clang::isOpenMPCombinedParallelADirective(OpenMPDirectiveKind DKind) {
@@ -808,6 +808,7 @@ void clang::getOpenMPCaptureRegions(
     break;
   case OMPD_tile:
   case OMPD_unroll:
+  case OMPD_reverse:
     // loop transformations do not introduce captures.
     break;
   case OMPD_threadprivate:
diff --git a/clang/lib/CodeGen/CGStmt.cpp b/clang/lib/CodeGen/CGStmt.cpp
index 99daaa14cf3fe..93c2f8900dd12 100644
--- a/clang/lib/CodeGen/CGStmt.cpp
+++ b/clang/lib/CodeGen/CGStmt.cpp
@@ -222,6 +222,9 @@ void CodeGenFunction::EmitStmt(const Stmt *S, ArrayRef<const Attr *> Attrs) {
   case Stmt::OMPUnrollDirectiveClass:
     EmitOMPUnrollDirective(cast<OMPUnrollDirective>(*S));
     break;
+  case Stmt::OMPReverseDirectiveClass:
+    EmitOMPReverseDirective(cast<OMPReverseDirective>(*S));
+    break;
   case Stmt::OMPForDirectiveClass:
     EmitOMPForDirective(cast<OMPForDirective>(*S));
     break;
diff --git a/clang/lib/CodeGen/CGStmtOpenMP.cpp b/clang/lib/CodeGen/CGStmtOpenMP.cpp
index 6410f9e102c90..ad6c044aa483b 100644
--- a/clang/lib/CodeGen/CGStmtOpenMP.cpp
+++ b/clang/lib/CodeGen/CGStmtOpenMP.cpp
@@ -187,6 +187,8 @@ class OMPLoopScope : public CodeGenFunction::RunCleanupsScope {
       PreInits = Tile->getPreInits();
     } else if (const auto *Unroll = dyn_cast<OMPUnrollDirective>(&S)) {
       PreInits = Unroll->getPreInits();
+    } else if (const auto *Reverse = dyn_cast<OMPReverseDirective>(&S)) {
+      PreInits = Reverse->getPreInits();
     } else {
       llvm_unreachable("Unknown loop-based directive kind.");
     }
@@ -2762,6 +2764,12 @@ void CodeGenFunction::EmitOMPTileDirective(const OMPTileDirective &S) {
   EmitStmt(S.getTransformedStmt());
 }
 
+void CodeGenFunction::EmitOMPReverseDirective(const OMPReverseDirective &S) {
+  // Emit the de-sugared statement.
+  OMPTransformDirectiveScopeRAII ReverseScope(*this, &S);
+  EmitStmt(S.getTransformedStmt());
+}
+
 void CodeGenFunction::EmitOMPUnrollDirective(const OMPUnrollDirective &S) {
   bool UseOMPIRBuilder = CGM.getLangOpts().OpenMPIRBuilder;
 
diff --git a/clang/lib/CodeGen/CodeGenFunction.h b/clang/lib/CodeGen/CodeGenFunction.h
index 5f3ee7eb943f9..ac738e1e82886 100644
--- a/clang/lib/CodeGen/CodeGenFunction.h
+++ b/clang/lib/CodeGen/CodeGenFunction.h
@@ -3807,6 +3807,7 @@ class CodeGenFunction : public CodeGenTypeCache {
   void EmitOMPSimdDirective(const OMPSimdDirective &S);
   void EmitOMPTileDirective(const OMPTileDirective &S);
   void EmitOMPUnrollDirective(const OMPUnrollDirective &S);
+  void EmitOMPReverseDirective(const OMPReverseDirective &S);
   void EmitOMPForDirective(const OMPForDirective &S);
   void EmitOMPForSimdDirective(const OMPForSimdDirective &S);
   void EmitOMPSectionsDirective(const OMPSectionsDirective &S);
diff --git a/clang/lib/Parse/ParseOpenMP.cpp b/clang/lib/Parse/ParseOpenMP.cpp
index e959dd6378f46..57fcf6ce520ac 100644
--- a/clang/lib/Parse/ParseOpenMP.cpp
+++ b/clang/lib/Parse/ParseOpenMP.cpp
@@ -2384,6 +2384,7 @@ Parser::DeclGroupPtrTy Parser::ParseOpenMPDeclarativeDirectiveWithExtDecl(
   case OMPD_simd:
   case OMPD_tile:
   case OMPD_unroll:
+  case OMPD_reverse:
   case OMPD_task:
   case OMPD_taskyield:
   case OMPD_barrier:
@@ -2802,6 +2803,7 @@ StmtResult Parser::ParseOpenMPDeclarativeOrExecutableDirective(
   case OMPD_simd:
   case OMPD_tile:
   case OMPD_unroll:
+  case OMPD_reverse:
   case OMPD_for:
   case OMPD_for_simd:
   case OMPD_sections:
diff --git a/clang/lib/Sema/SemaExceptionSpec.cpp b/clang/lib/Sema/SemaExceptionSpec.cpp
index 41bf273d12f2f..4de7183cde281 100644
--- a/clang/lib/Sema/SemaExceptionSpec.cpp
+++ b/clang/lib/Sema/SemaExceptionSpec.cpp
@@ -1486,6 +1486,7 @@ CanThrowResult Sema::canThrow(const Stmt *S) {
   case Stmt::OMPSimdDirectiveClass:
   case Stmt::OMPTileDirectiveClass:
   case Stmt::OMPUnrollDirectiveClass:
+  case Stmt::OMPReverseDirectiveClass:
   case Stmt::OMPSingleDirectiveClass:
   case Stmt::OMPTargetDataDirectiveClass:
   case Stmt::OMPTargetDirectiveClass:
diff --git a/clang/lib/Sema/SemaOpenMP.cpp b/clang/lib/Sema/SemaOpenMP.cpp
index 663dbb927250e..7b9898704eb1c 100644
--- a/clang/lib/Sema/SemaOpenMP.cpp
+++ b/clang/lib/Sema/SemaOpenMP.cpp
@@ -4334,6 +4334,7 @@ void SemaOpenMP::ActOnOpenMPRegionStart(OpenMPDirectiveKind DKind,
   case OMPD_masked:
   case OMPD_tile:
   case OMPD_unroll:
+  case OMPD_reverse:
     break;
   case OMPD_loop:
     // TODO: 'loop' may require additional parameters depending on the binding.
@@ -6546,6 +6547,10 @@ StmtResult SemaOpenMP::ActOnOpenMPExecutableDirective(
     Res = ActOnOpenMPUnrollDirective(ClausesWithImplicit, AStmt, StartLoc,
                                      EndLoc);
     break;
+  case OMPD_reverse:
+    Res = ActOnOpenMPReverseDirective(ClausesWithImplicit, AStmt, StartLoc,
+                                      EndLoc);
+    break;
   case OMPD_for:
     Res = ActOnOpenMPForDirective(ClausesWithImplicit, AStmt, StartLoc, EndLoc,
                                   VarsWithInheritedDSA);
@@ -15121,6 +15126,8 @@ bool SemaOpenMP::checkTransformableLoopNest(
           DependentPreInits = Dir->getPreInits();
         else if (auto *Dir = dyn_cast<OMPUnrollDirective>(Transform))
           DependentPreInits = Dir->getPreInits();
+        else if (auto *Dir = dyn_cast<OMPReverseDirective>(Transform))
+          DependentPreInits = Dir->getPreInits();
         else
           llvm_unreachable("Unhandled loop transformation");
         if (!DependentPreInits)
@@ -15746,6 +15753,189 @@ StmtResult SemaOpenMP::ActOnOpenMPUnrollDirective(ArrayRef<OMPClause *> Clauses,
                                     buildPreInits(Context, PreInits));
 }
 
+StmtResult
+SemaOpenMP::ActOnOpenMPReverseDirective(ArrayRef<OMPClause *> Clauses,
+                                        Stmt *AStmt, SourceLocation StartLoc,
+                                        SourceLocation EndLoc) {
+  ASTContext &Context = getASTContext();
+  Scope *CurScope = SemaRef.getCurScope();
+  assert(Clauses.empty() && "reverse directive does not accept any clauses; "
+                            "must have beed checked before");
+
+  // Empty statement should only be possible if there already was an error.
+  if (!AStmt)
+    return StmtError();
+
+  constexpr unsigned NumLoops = 1;
+  Stmt *Body = nullptr;
+  SmallVector<OMPLoopBasedDirective::HelperExprs, NumLoops> LoopHelpers(
+      NumLoops);
+  SmallVector<SmallVector<Stmt *, 0>, NumLoops + 1> OriginalInits;
+  if (!checkTransformableLoopNest(OMPD_reverse, AStmt, NumLoops, LoopHelpers,
+                                  Body, OriginalInits))
+    return StmtError();
+
+  // Delay applying the transformation to when template is completely
+  // instantiated.
+  if (SemaRef.CurContext->isDependentContext())
+    return OMPReverseDirective::Create(Context, StartLoc, EndLoc, Clauses,
+                                       AStmt, nullptr, nullptr);
+
+  assert(LoopHelpers.size() == NumLoops &&
+         "Expecting a single-dimensional loop iteration space");
+  assert(OriginalInits.size() == NumLoops &&
+         "Expecting a single-dimensional loop iteration space");
+  OMPLoopBasedDirective::HelperExprs &LoopHelper = LoopHelpers.front();
+
+  // Find the loop statement.
+  Stmt *LoopStmt = nullptr;
+  collectLoopStmts(AStmt, {LoopStmt});
+
+  // Determine the PreInit declarations.
+  SmallVector<Stmt *> PreInits;
+  addLoopPreInits(Context, LoopHelper, LoopStmt, OriginalInits[0], PreInits);
+
+  auto *IterationVarRef = cast<DeclRefExpr>(LoopHelper.IterationVarRef);
+  QualType IVTy = IterationVarRef->getType();
+  uint64_t IVWidth = Context.getTypeSize(IVTy);
+  auto *OrigVar = cast<DeclRefExpr>(LoopHelper.Counters.front());
+
+  // Iteration variable SourceLocations.
+  SourceLocation OrigVarLoc = OrigVar->getExprLoc();
+  SourceLocation OrigVarLocBegin = OrigVar->getBeginLoc();
+  SourceLocation OrigVarLocEnd = OrigVar->getEndLoc();
+
+  // Locations pointing to the transformation.
+  SourceLocation TransformLoc = StartLoc;
+  SourceLocation TransformLocBegin = StartLoc;
+  SourceLocation TransformLocEnd = EndLoc;
+
+  // Internal variable names.
+  std::string OrigVarName = OrigVar->getNameInfo().getAsString();
+  std::string TripCountName = (Twine(".tripcount.") + OrigVarName).str();
+  std::string ForwardIVName = (Twine(".forward.iv.") + OrigVarName).str();
+  std::string ReversedIVName = (Twine(".reversed.iv.") + OrigVarName).str();
+
+  // LoopHelper.Updates will read the logical iteration number from
+  // LoopHelper.IterationVarRef, compute the value of the user loop counter of
+  // that logical iteration from it, then assign it to the user loop counter
+  // variable. We cannot directly use LoopHelper.IterationVarRef as the
+  // induction variable of the generated loop because it may cause an underflow:
+  // \code
+  //   for (unsigned i = 0; i < n; ++i)
+  //     body(i);
+  // \endcode
+  //
+  // Naive reversal:
+  // \code
+  //   for (unsigned i = n-1; i >= 0; --i)
+  //     body(i);
+  // \endcode
+  //
+  // Instead, we introduce a new iteration variable representing the logical
+  // iteration counter of the original loop, convert it to the logical iteration
+  // number of the reversed loop, then let LoopHelper.Updates compute the user's
+  // loop iteration variable from it.
+  // \code
+  //   for (auto .forward.iv = 0; .forward.iv < n; ++.forward.iv) {
+  //     auto .reversed.iv = n - .forward.iv - 1;
+  //     i = (.reversed.iv + 0) * 1;                // LoopHelper.Updates
+  //     body(i);                                   // Body
+  //   }
+  // \endcode
+
+  // Subexpressions with more than one use. One of the constraints of an AST is
+  // that every node object must appear at most once, hence we define a lambda
+  // that creates a new AST node at every use.
+  CaptureVars CopyTransformer(SemaRef);
+  auto MakeNumIterations = [&CopyTransformer, &LoopHelper]() -> Expr * {
+    return AssertSuccess(
+        CopyTransformer.TransformExpr(LoopHelper.NumIterations));
+  };
+
+  // Create the iteration variable for the forward loop (from 0 to n-1).
+  VarDecl *ForwardIVDecl =
+      buildVarDecl(SemaRef, {}, IVTy, ForwardIVName, nullptr, OrigVar);
+  auto MakeForwardRef = [&SemaRef = this->SemaRef, ForwardIVDecl, IVTy,
+                         OrigVarLoc]() {
+    return buildDeclRefExpr(SemaRef, ForwardIVDecl, IVTy, OrigVarLoc);
+  };
+
+  // Iteration variable for the reversed induction variable (from n-1 downto 0):
+  // Reuse the iteration variable created by checkOpenMPLoop.
+  auto *ReversedIVDecl = cast<VarDecl>(IterationVarRef->getDecl());
+  ReversedIVDecl->setDeclName(
+      &SemaRef.PP.getIdentifierTable().get(ReversedIVName));
+
+  // For init-statement:
+  // \code
+  //   auto .forward.iv = 0
+  // \endcode
+  IntegerLiteral *Zero =
+      IntegerLiteral::Create(Context, llvm::APInt::getZero(IVWidth),
+                             ForwardIVDecl->getType(), OrigVarLoc);
+  SemaRef.AddInitializerToDecl(ForwardIVDecl, Zero, /*DirectInit=*/false);
+  StmtResult Init = new (Context)
+      DeclStmt(DeclGroupRef(ForwardIVDecl), OrigVarLocBegin, OrigVarLocEnd);
+  if (!Init.isUsable())
+    return StmtError();
+
+  // Forward iv cond-expression:
+  // \code
+  //   .forward.iv < NumIterations
+  // \endcode
+  ExprResult Cond =
+      SemaRef.BuildBinOp(CurScope, LoopHelper.Cond->getExprLoc(), BO_LT,
+                         MakeForwardRef(), MakeNumIterations());
+  if (!Cond.isUsable())
+    return StmtError();
+
+  // Forward incr-statement: ++.forward.iv
+  ExprResult Incr = SemaRef.BuildUnaryOp(CurScope, LoopHelper.Inc->getExprLoc(),
+                                         UO_PreInc, MakeForwardRef());
+  if (!Incr.isUsable())
+    return StmtError();
+
+  // Reverse the forward-iv: auto .reversed.iv = MakeNumIterations() - 1 -
+  // .forward.iv
+  IntegerLiteral *One = IntegerLiteral::Create(Context, llvm::APInt(IVWidth, 1),
+                                               IVTy, TransformLoc);
+  ExprResult Minus = SemaRef.BuildBinOp(CurScope, TransformLoc, BO_Sub,
+                                        MakeNumIterations(), One);
+  if (!Minus.isUsable())
+    return StmtError();
+  Minus = SemaRef.BuildBinOp(CurScope, TransformLoc, BO_Sub, Minus.get(),
+                             MakeForwardRef());
+  if (!Minus.isUsable())
+    return StmtError();
+  StmtResult InitReversed = new (Context) DeclStmt(
+      DeclGroupRef(ReversedIVDecl), TransformLocBegin, TransformLocEnd);
+  if (!InitReversed.isUsable())
+    return StmtError();
+  SemaRef.AddInitializerToDecl(ReversedIVDecl, Minus.get(),
+                               /*DirectInit=*/false);
+
+  // The new loop body.
+  SmallVector<Stmt *> BodyStmts;
+  BodyStmts.push_back(InitReversed.get());
+  llvm::append_range(BodyStmts, LoopHelper.Updates);
+  if (auto *CXXRangeFor = dyn_cast<CXXForRangeStmt>(LoopStmt))
+    BodyStmts.push_back(CXXRangeFor->getLoopVarStmt());
+  BodyStmts.push_back(Body);
+  auto *ReversedBody =
+      CompoundStmt::Create(Context, BodyStmts, FPOptionsOverride(),
+                           Body->getBeginLoc(), Body->getEndLoc());
+
+  // Finally create the reversed For-statement.
+  auto *ReversedFor = new (Context)
+      ForStmt(Context, Init.get(), Cond.get(), nullptr, Incr.get(),
+              ReversedBody, LoopHelper.Init->getBeginLoc(),
+              LoopHelper.Init->getBeginLoc(), LoopHelper.Inc->getEndLoc());
+  return OMPReverseDirective::Create(Context, StartLoc, EndLoc, Clauses, AStmt,
+                                     ReversedFor,
+                                     buildPreInits(Context, PreInits));
+}
+
 OMPClause *SemaOpenMP::ActOnOpenMPSingleExprClause(OpenMPClauseKind Kind,
                                                    Expr *Expr,
                                                    SourceLocation StartLoc,
diff --git a/clang/lib/Sema/TreeTransform.h b/clang/lib/Sema/TreeTransform.h
index 06ed0843ef504..f8713976fcc5c 100644
--- a/clang/lib/Sema/TreeTransform.h
+++ b/clang/lib/Sema/TreeTransform.h
@@ -9172,6 +9172,17 @@ TreeTransform<Derived>::TransformOMPUnrollDirective(OMPUnrollDirective *D) {
   return Res;
 }
 
+template <typename Derived>
+StmtResult
+TreeTransform<Derived>::TransformOMPReverseDirective(OMPReverseDirective *D) {
+  DeclarationNameInfo DirName;
+  getDerived().getSema().OpenMP().StartOpenMPDSABlock(
+      D->getDirectiveKind(), DirName, nullptr, D->getBeginLoc());
+  StmtResult Res = getDerived().TransformOMPExecutableDirective(D);
+  getDerived().getSema().OpenMP().EndOpenMPDSABlock(Res.get());
+  return Res;
+}
+
 template <typename Derived>
 StmtResult
 TreeTransform<Derived>::TransformOMPForDirective(OMPForDirective *D) {
diff --git a/clang/lib/Serialization/ASTReaderStmt.cpp b/clang/lib/Serialization/ASTReaderStmt.cpp
index eac4faff28549..ff72679f993e2 100644
--- a/clang/lib/Serialization/ASTReaderStmt.cpp
+++ b/clang/lib/Serialization/ASTReaderStmt.cpp
@@ -2428,6 +2428,10 @@ void ASTStmtReader::VisitOMPUnrollDirective(OMPUnrollDirective *D) {
   VisitOMPLoopTransformationDirective(D);
 }
 
+void ASTStmtReader::VisitOMPReverseDirective(OMPReverseDirective *D) {
+  VisitOMPLoopTransformationDirective(D);
+}
+
 void ASTStmtReader::VisitOMPForDirective(OMPForDirective *D) {
   VisitOMPLoopDirective(D);
   D->setHasCancel(Record.readBool());
@@ -3436,6 +3440,14 @@ Stmt *ASTReader::ReadStmtFromStream(ModuleFile &F) {
       break;
     }
 
+    case STMT_OMP_REVERSE_DIRECTIVE: {
+      assert(Record[ASTStmtReader::NumStmtFields] == 1 &&
+             "Reverse directive accepts only a single loop");
+      unsigned NumClauses = Record[ASTStmtReader::NumStmtFields + 1];
+      S = OMPReverseDirective::CreateEmpty(Context, NumClauses);
+      break;
+    }
+
     case STMT_OMP_FOR_DIRECTIVE: {
       unsigned CollapsedNum = Record[ASTStmtReader::NumStmtFields];
       unsigned NumClauses = Record[ASTStmtReader::NumStmtFields + 1];
diff --git a/clang/lib/Serialization/ASTWriterStmt.cpp b/clang/lib/Serialization/ASTWriterStmt.cpp
index a44852af97bea..ffb09ece981e6 100644
--- a/clang/lib/Serialization/ASTWriterStmt.cpp
+++ b/clang/lib/Serialization/ASTWriterStmt.cpp
@@ -2412,6 +2412,11 @@ void ASTStmtWriter::VisitOMPUnrollDirective(OMPUnrollDirective *D) {
   Code = serialization::STMT_OMP_UNROLL_DIRECTIVE;
 }
 
+void ASTStmtWriter::VisitOMPReverseDirective(OMPReverseDirective *D) {
+  VisitOMPLoopTransformationDirective(D);
+  Code = serialization::STMT_OMP_REVERSE_DIRECTIVE;
+}
+
 void ASTStmtWriter::VisitOMPForDirective(OMPForDirective *D) {
   VisitOMPLoopDirective(D);
   Record.writeBool(D->hasCancel());
diff --git a/clang/test/OpenMP/reverse_ast_print.cpp b/clang/test/OpenMP/reverse_ast_print.cpp
new file mode 100644
index 0000000000000..3ff6d18cfdf8b
--- /dev/null
+++ b/clang/test/OpenMP/reverse_ast_print.cpp
@@ -0,0 +1,159 @@
+// Check no warnings/errors
+// RUN: %clang_cc1 -triple x86_64-pc-linux-gnu -fopenmp -fopenmp-version=60 -fsyntax-only -verify %s
+// expected-no-diagnostics
+
+// Check AST and unparsing
+// RUN: %clang_cc1 -triple x86_64-pc-linux-gnu -fopenmp -fopenmp-version=60 -ast-dump  %s | FileCheck %s --check-prefix=DUMP
+// RUN: %clang_cc1 -triple x86_64-pc-linux-gnu -fopenmp -fopenmp-version=60 -ast-print %s | FileCheck %s --check-prefix=PRINT
+
+// Check same results after serialization round-trip
+// RUN: %clang_cc1 -triple x86_64-pc-linux-gnu -fopenmp -fopenmp-version=60 -emit-pch -o %t %s
+// RUN: %clang_cc1 -triple x86_64-pc-linux-gnu -fopenmp -fopenmp-version=60 -include-pch %t -ast-dump-all %s | FileCheck %s --check-prefix=DUMP
+// RUN: %clang_cc1 -triple x86_64-pc-linux-gnu -fopenmp -fopenmp-version=60 -include-pch %t -ast-print    %s | FileCheck %s --check-prefix=PRINT
+
+#ifndef HEADER
+#define HEADER
+
+// placeholder for loop body code.
+extern "C" void body(...);
+
+// PRINT-LABEL: void foo1(
+// DUMP-LABEL:  FunctionDecl {{.*}} foo1
+void foo1() {
+  // PRINT:     #pragma omp reverse
+  // DUMP:      OMPReverseDirective
+  #pragma omp reverse
+  // PRINT: for (int i = 7; i < 17; i += 3)
+  // DUMP-NEXT: ForStmt
+  for (int i = 7; i < 17; i += 3)
+    // PRINT: body(i);
+    // DUMP:  CallExpr
+      body(i);
+}
+
+
+// PRINT-LABEL: void foo2(
+// DUMP-LABEL:  FunctionDecl {{.*}} foo2
+void foo2(int start, int end, int step) {
+  // PRINT:     #pragma omp reverse
+  // DUMP:      OMPReverseDirective
+  #pragma omp reverse
+  // PRINT: for (int i = start; i < end; i += step)
+  // DUMP-NEXT: ForStmt
+  for (int i = start; i < end; i += step)
+      // PRINT: body(i);
+      // DUMP:  CallExpr
+      body(i);
+}
+
+
+// PRINT-LABEL: void foo3(
+// DUMP-LABEL:  FunctionDecl {{.*}} foo3
+void foo3() {
+  // PRINT: #pragma omp for
+  // DUMP:  OMPForDirective
+  // DUMP-NEXT:    CapturedStmt
+  // DUMP-NEXT:      CapturedDecl
+  #pragma omp for
+  // PRINT:     #pragma omp reverse 
+  // DUMP-NEXT: OMPReverseDirective
+  #pragma omp reverse
+  for (int i = 7; i < 17; i += 3)
+    // PRINT: body(i);
+    // DUMP:  CallExpr
+    body(i);
+}
+
+
+// PRINT-LABEL: void foo4(
+// DUMP-LABEL:  FunctionDecl {{.*}} foo4
+void foo4() {
+  // PRINT: #pragma omp for collapse(2)
+  // DUMP: OMPForDirective
+  // DUMP-NEXT: OMPCollapseClause
+  // DUMP-NEXT:  ConstantExpr
+  // DUMP-NEXT:    value: Int 2
+  // DUMP-NEXT:  IntegerLiteral {{.*}} 2
+  // DUMP-NEXT:    CapturedStmt
+  // DUMP-NEXT:      CapturedDecl
+  #pragma omp for collapse(2)
+  // PRINT:     #pragma omp reverse
+  // DUMP:      OMPReverseDirective
+  #pragma omp reverse
+  // PRINT: for (int i = 7; i < 17; i += 1)
+  // DUMP-NEXT: ForStmt
+  for (int i = 7; i < 17; i += 1)
+    // PRINT: for (int j = 7; j < 17; j += 1)
+    // DUMP:  ForStmt
+    for (int j = 7; j < 17; j += 1)
+      // PRINT: body(i, j);
+      // DUMP:  CallExpr
+      body(i, j);
+}
+
+
+// PRINT-LABEL: void foo5(
+// DUMP-LABEL:  FunctionDecl {{.*}} foo5
+void foo5(int start, int end, int step) {
+  // PRINT:     #pragma omp for collapse(2)
+  // DUMP:      OMPForDirective
+  // DUMP-NEXT:   OMPCollapseClause
+  // DUMP-NEXT:    ConstantExpr
+  // DUMP-NEXT:      value: Int 2
+  // DUMP-NEXT:    IntegerLiteral {{.*}} 2
+  // DUMP-NEXT:  CapturedStmt
+  // DUMP-NEXT:    CapturedDecl
+  #pragma omp for collapse(2)
+  // PRINT:     for (int i = 7; i < 17; i += 1)
+  // DUMP-NEXT: ForStmt
+  for (int i = 7; i < 17; i += 1)
+    // PRINT: #pragma omp reverse
+    // DUMP:  OMPReverseDirective
+    #pragma omp reverse 
+    // PRINT:     for (int j = 7; j < 17; j += 1)
+    // DUMP-NEXT: ForStmt
+    for (int j = 7; j < 17; j += 1)
+      // PRINT: body(i, j);
+      // DUMP:  CallExpr
+      body(i, j);
+}
+
+
+// PRINT-LABEL: void foo6(
+// DUMP-LABEL:  FunctionTemplateDecl {{.*}} foo6
+template<typename T, T Step>
+void foo6(T start, T end) {
+  // PRINT: #pragma omp reverse
+  // DUMP:  OMPReverseDirective
+  #pragma omp reverse
+    // PRINT-NEXT: for (T i = start; i < end; i += Step)
+    // DUMP-NEXT:  ForStmt
+    for (T i = start; i < end; i += Step)
+      // PRINT-NEXT: body(i);
+      // DUMP:       CallExpr
+      body(i);
+}
+
+// Also test instantiating the template.
+void tfoo6() {
+  foo6<int,3>(0, 42);
+}
+
+
+// PRINT-LABEL: void foo7(
+// DUMP-LABEL:  FunctionDecl {{.*}} foo7
+void foo7() {
+  double arr[128];
+  // PRINT: #pragma omp reverse
+  // DUMP:  OMPReverseDirective
+  #pragma omp reverse
+  // PRINT-NEXT: for (auto &&v : arr)
+  // DUMP-NEXT:  CXXForRangeStmt
+  for (auto &&v : arr)
+    // PRINT-NEXT: body(v);
+    // DUMP:       CallExpr
+    body(v);
+}
+
+#endif
+
diff --git a/clang/test/OpenMP/reverse_codegen.cpp b/clang/test/OpenMP/reverse_codegen.cpp
new file mode 100644
index 0000000000000..9adaa6cc7d18d
--- /dev/null
+++ b/clang/test/OpenMP/reverse_codegen.cpp
@@ -0,0 +1,1554 @@
+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --function-signature --include-generated-funcs --replace-value-regex "__omp_offloading_[0-9a-z]+_[0-9a-z]+" "reduction_size[.].+[.]" "pl_cond[.].+[.|,]" --prefix-filecheck-ir-name _
+
+// expected-no-diagnostics
+
+// Check code generation
+// RUN: %clang_cc1 -verify -triple x86_64-pc-linux-gnu -std=c++20 -fclang-abi-compat=latest -fopenmp -fopenmp-version=60 -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK1
+
+// Check same results after serialization round-trip
+// RUN: %clang_cc1 -verify -triple x86_64-pc-linux-gnu -std=c++20 -fclang-abi-compat=latest -fopenmp -fopenmp-version=60 -emit-pch -o %t %s
+// RUN: %clang_cc1 -verify -triple x86_64-pc-linux-gnu -std=c++20 -fclang-abi-compat=latest -fopenmp -fopenmp-version=60 -include-pch %t -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK2
+
+#ifndef HEADER
+#define HEADER
+
+// placeholder for loop body code.
+extern "C" void body(...) {}
+
+
+struct S {
+  int i;
+  S() {
+#pragma omp reverse
+    for (i = 7; i < 17; i += 3)
+      body(i);
+  }
+} s;
+
+
+extern "C" void foo1(int start, int end, int step) {
+  int i;
+#pragma omp reverse
+  for (i = start; i < end; i += step)
+    body(i);
+}
+
+
+extern "C" void foo2() {
+#pragma omp for
+#pragma omp reverse
+    for (int i = 7; i < 17; i += 3)
+        body(i);
+}
+
+
+extern "C" void foo3() {
+#pragma omp for collapse(3)
+  for (int k = 7; k < 17; k += 3)
+#pragma omp reverse
+    for (int i = 7; i < 17; i += 3)
+      for (int j = 7; j < 17; j += 3)
+        body(k, i, j);
+}
+
+
+extern "C" void foo4() {
+#pragma omp parallel for
+#pragma omp reverse
+  for (int i = 7; i < 17; i += 3)
+    body(i);
+}
+
+
+template<typename T, T Step>
+void foo5(T start, T end) {
+#pragma omp reverse
+  for (T i = start; i < end; i += Step)
+    body(i);
+}
+
+extern "C" void tfoo5() {
+  foo5<int,3>(0, 42);
+}
+
+
+extern "C" void foo6() {
+  double arr[128];
+#pragma omp reverse
+  for (int c = 42; auto && v : arr)
+    body(v, c);
+}
+
+
+extern "C" void foo7() {
+  double A[128];
+
+#pragma omp for collapse(3)
+  for (int k = 7; k < 17; k += 3)
+#pragma omp reverse
+    for (int c = 42; auto && v : A)
+      for (int j = 7; j < 17; j += 3)
+        body(k, c, v, j);
+}
+
+#endif /* HEADER */
+
+// CHECK1-LABEL: define {{[^@]+}}@body
+// CHECK1-SAME: (...) #[[ATTR0:[0-9]+]] {
+// CHECK1-NEXT:  entry:
+// CHECK1-NEXT:    ret void
+//
+//
+// CHECK1-LABEL: define {{[^@]+}}@__cxx_global_var_init
+// CHECK1-SAME: () #[[ATTR1:[0-9]+]] section ".text.startup" {
+// CHECK1-NEXT:  entry:
+// CHECK1-NEXT:    call void @_ZN1SC1Ev(ptr noundef nonnull align 4 dereferenceable(4) @s)
+// CHECK1-NEXT:    ret void
+//
+//
+// CHECK1-LABEL: define {{[^@]+}}@_ZN1SC1Ev
+// CHECK1-SAME: (ptr noundef nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR0]] comdat align 2 {
+// CHECK1-NEXT:  entry:
+// CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
+// CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
+// CHECK1-NEXT:    call void @_ZN1SC2Ev(ptr noundef nonnull align 4 dereferenceable(4) [[THIS1]])
+// CHECK1-NEXT:    ret void
+//
+//
+// CHECK1-LABEL: define {{[^@]+}}@_ZN1SC2Ev
+// CHECK1-SAME: (ptr noundef nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR0]] comdat align 2 {
+// CHECK1-NEXT:  entry:
+// CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[I2:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[DOTFORWARD_IV_I:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[DOTREVERSED_IV_I:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
+// CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
+// CHECK1-NEXT:    [[I:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], ptr [[THIS1]], i32 0, i32 0
+// CHECK1-NEXT:    store i32 7, ptr [[I]], align 4
+// CHECK1-NEXT:    [[I3:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[THIS1]], i32 0, i32 0
+// CHECK1-NEXT:    store ptr [[I3]], ptr [[I2]], align 8
+// CHECK1-NEXT:    store i32 0, ptr [[DOTFORWARD_IV_I]], align 4
+// CHECK1-NEXT:    br label [[FOR_COND:%.*]]
+// CHECK1:       for.cond:
+// CHECK1-NEXT:    [[TMP0:%.*]] = load i32, ptr [[DOTFORWARD_IV_I]], align 4
+// CHECK1-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP0]], 4
+// CHECK1-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
+// CHECK1:       for.body:
+// CHECK1-NEXT:    [[TMP1:%.*]] = load i32, ptr [[DOTFORWARD_IV_I]], align 4
+// CHECK1-NEXT:    [[SUB:%.*]] = sub nsw i32 3, [[TMP1]]
+// CHECK1-NEXT:    store i32 [[SUB]], ptr [[DOTREVERSED_IV_I]], align 4
+// CHECK1-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTREVERSED_IV_I]], align 4
+// CHECK1-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP2]], 3
+// CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 7, [[MUL]]
+// CHECK1-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[I2]], align 8
+// CHECK1-NEXT:    store i32 [[ADD]], ptr [[TMP3]], align 4
+// CHECK1-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[I2]], align 8
+// CHECK1-NEXT:    [[TMP5:%.*]] = load i32, ptr [[TMP4]], align 4
+// CHECK1-NEXT:    call void (...) @body(i32 noundef [[TMP5]])
+// CHECK1-NEXT:    br label [[FOR_INC:%.*]]
+// CHECK1:       for.inc:
+// CHECK1-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTFORWARD_IV_I]], align 4
+// CHECK1-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP6]], 1
+// CHECK1-NEXT:    store i32 [[INC]], ptr [[DOTFORWARD_IV_I]], align 4
+// CHECK1-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP3:![0-9]+]]
+// CHECK1:       for.end:
+// CHECK1-NEXT:    ret void
+//
+//
+// CHECK1-LABEL: define {{[^@]+}}@foo1
+// CHECK1-SAME: (i32 noundef [[START:%.*]], i32 noundef [[END:%.*]], i32 noundef [[STEP:%.*]]) #[[ATTR0]] {
+// CHECK1-NEXT:  entry:
+// CHECK1-NEXT:    [[START_ADDR:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[END_ADDR:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[STEP_ADDR:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[I:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[DOTNEW_STEP:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[DOTFORWARD_IV_I:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[DOTREVERSED_IV_I:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    store i32 [[START]], ptr [[START_ADDR]], align 4
+// CHECK1-NEXT:    store i32 [[END]], ptr [[END_ADDR]], align 4
+// CHECK1-NEXT:    store i32 [[STEP]], ptr [[STEP_ADDR]], align 4
+// CHECK1-NEXT:    [[TMP0:%.*]] = load i32, ptr [[START_ADDR]], align 4
+// CHECK1-NEXT:    store i32 [[TMP0]], ptr [[I]], align 4
+// CHECK1-NEXT:    [[TMP1:%.*]] = load i32, ptr [[START_ADDR]], align 4
+// CHECK1-NEXT:    store i32 [[TMP1]], ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK1-NEXT:    [[TMP2:%.*]] = load i32, ptr [[END_ADDR]], align 4
+// CHECK1-NEXT:    store i32 [[TMP2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
+// CHECK1-NEXT:    [[TMP3:%.*]] = load i32, ptr [[STEP_ADDR]], align 4
+// CHECK1-NEXT:    store i32 [[TMP3]], ptr [[DOTNEW_STEP]], align 4
+// CHECK1-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
+// CHECK1-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK1-NEXT:    [[SUB:%.*]] = sub i32 [[TMP4]], [[TMP5]]
+// CHECK1-NEXT:    [[SUB3:%.*]] = sub i32 [[SUB]], 1
+// CHECK1-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTNEW_STEP]], align 4
+// CHECK1-NEXT:    [[ADD:%.*]] = add i32 [[SUB3]], [[TMP6]]
+// CHECK1-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTNEW_STEP]], align 4
+// CHECK1-NEXT:    [[DIV:%.*]] = udiv i32 [[ADD]], [[TMP7]]
+// CHECK1-NEXT:    [[SUB4:%.*]] = sub i32 [[DIV]], 1
+// CHECK1-NEXT:    store i32 [[SUB4]], ptr [[DOTCAPTURE_EXPR_2]], align 4
+// CHECK1-NEXT:    store i32 0, ptr [[DOTFORWARD_IV_I]], align 4
+// CHECK1-NEXT:    br label [[FOR_COND:%.*]]
+// CHECK1:       for.cond:
+// CHECK1-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTFORWARD_IV_I]], align 4
+// CHECK1-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
+// CHECK1-NEXT:    [[ADD5:%.*]] = add i32 [[TMP9]], 1
+// CHECK1-NEXT:    [[CMP:%.*]] = icmp ult i32 [[TMP8]], [[ADD5]]
+// CHECK1-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
+// CHECK1:       for.body:
+// CHECK1-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
+// CHECK1-NEXT:    [[ADD6:%.*]] = add i32 [[TMP10]], 1
+// CHECK1-NEXT:    [[SUB7:%.*]] = sub i32 [[ADD6]], 1
+// CHECK1-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTFORWARD_IV_I]], align 4
+// CHECK1-NEXT:    [[SUB8:%.*]] = sub i32 [[SUB7]], [[TMP11]]
+// CHECK1-NEXT:    store i32 [[SUB8]], ptr [[DOTREVERSED_IV_I]], align 4
+// CHECK1-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK1-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTREVERSED_IV_I]], align 4
+// CHECK1-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTNEW_STEP]], align 4
+// CHECK1-NEXT:    [[MUL:%.*]] = mul i32 [[TMP13]], [[TMP14]]
+// CHECK1-NEXT:    [[ADD9:%.*]] = add i32 [[TMP12]], [[MUL]]
+// CHECK1-NEXT:    store i32 [[ADD9]], ptr [[I]], align 4
+// CHECK1-NEXT:    [[TMP15:%.*]] = load i32, ptr [[I]], align 4
+// CHECK1-NEXT:    call void (...) @body(i32 noundef [[TMP15]])
+// CHECK1-NEXT:    br label [[FOR_INC:%.*]]
+// CHECK1:       for.inc:
+// CHECK1-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTFORWARD_IV_I]], align 4
+// CHECK1-NEXT:    [[INC:%.*]] = add i32 [[TMP16]], 1
+// CHECK1-NEXT:    store i32 [[INC]], ptr [[DOTFORWARD_IV_I]], align 4
+// CHECK1-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP5:![0-9]+]]
+// CHECK1:       for.end:
+// CHECK1-NEXT:    ret void
+//
+//
+// CHECK1-LABEL: define {{[^@]+}}@foo2
+// CHECK1-SAME: () #[[ATTR0]] {
+// CHECK1-NEXT:  entry:
+// CHECK1-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[TMP:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[I:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[DOTFORWARD_IV_I:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[DOTREVERSED_IV_I:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB2:[0-9]+]])
+// CHECK1-NEXT:    store i32 7, ptr [[I]], align 4
+// CHECK1-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
+// CHECK1-NEXT:    store i32 3, ptr [[DOTOMP_UB]], align 4
+// CHECK1-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
+// CHECK1-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
+// CHECK1-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1:[0-9]+]], i32 [[TMP0]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
+// CHECK1-NEXT:    [[TMP1:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
+// CHECK1-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP1]], 3
+// CHECK1-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
+// CHECK1:       cond.true:
+// CHECK1-NEXT:    br label [[COND_END:%.*]]
+// CHECK1:       cond.false:
+// CHECK1-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
+// CHECK1-NEXT:    br label [[COND_END]]
+// CHECK1:       cond.end:
+// CHECK1-NEXT:    [[COND:%.*]] = phi i32 [ 3, [[COND_TRUE]] ], [ [[TMP2]], [[COND_FALSE]] ]
+// CHECK1-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
+// CHECK1-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
+// CHECK1-NEXT:    store i32 [[TMP3]], ptr [[DOTOMP_IV]], align 4
+// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
+// CHECK1:       omp.inner.for.cond:
+// CHECK1-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
+// CHECK1-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
+// CHECK1-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP4]], [[TMP5]]
+// CHECK1-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
+// CHECK1:       omp.inner.for.body:
+// CHECK1-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
+// CHECK1-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP6]], 1
+// CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
+// CHECK1-NEXT:    store i32 [[ADD]], ptr [[DOTFORWARD_IV_I]], align 4
+// CHECK1-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTFORWARD_IV_I]], align 4
+// CHECK1-NEXT:    [[SUB:%.*]] = sub nsw i32 3, [[TMP7]]
+// CHECK1-NEXT:    store i32 [[SUB]], ptr [[DOTREVERSED_IV_I]], align 4
+// CHECK1-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTREVERSED_IV_I]], align 4
+// CHECK1-NEXT:    [[MUL2:%.*]] = mul nsw i32 [[TMP8]], 3
+// CHECK1-NEXT:    [[ADD3:%.*]] = add nsw i32 7, [[MUL2]]
+// CHECK1-NEXT:    store i32 [[ADD3]], ptr [[I]], align 4
+// CHECK1-NEXT:    [[TMP9:%.*]] = load i32, ptr [[I]], align 4
+// CHECK1-NEXT:    call void (...) @body(i32 noundef [[TMP9]])
+// CHECK1-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
+// CHECK1:       omp.body.continue:
+// CHECK1-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
+// CHECK1:       omp.inner.for.inc:
+// CHECK1-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
+// CHECK1-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP10]], 1
+// CHECK1-NEXT:    store i32 [[ADD4]], ptr [[DOTOMP_IV]], align 4
+// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]]
+// CHECK1:       omp.inner.for.end:
+// CHECK1-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
+// CHECK1:       omp.loop.exit:
+// CHECK1-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP0]])
+// CHECK1-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB3:[0-9]+]], i32 [[TMP0]])
+// CHECK1-NEXT:    ret void
+//
+//
+// CHECK1-LABEL: define {{[^@]+}}@foo3
+// CHECK1-SAME: () #[[ATTR0]] {
+// CHECK1-NEXT:  entry:
+// CHECK1-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[TMP:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[_TMP2:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[I:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[K:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[DOTFORWARD_IV_I:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[J:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[DOTREVERSED_IV_I:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB2]])
+// CHECK1-NEXT:    store i32 7, ptr [[I]], align 4
+// CHECK1-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
+// CHECK1-NEXT:    store i32 63, ptr [[DOTOMP_UB]], align 4
+// CHECK1-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
+// CHECK1-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
+// CHECK1-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP0]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
+// CHECK1-NEXT:    [[TMP1:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
+// CHECK1-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP1]], 63
+// CHECK1-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
+// CHECK1:       cond.true:
+// CHECK1-NEXT:    br label [[COND_END:%.*]]
+// CHECK1:       cond.false:
+// CHECK1-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
+// CHECK1-NEXT:    br label [[COND_END]]
+// CHECK1:       cond.end:
+// CHECK1-NEXT:    [[COND:%.*]] = phi i32 [ 63, [[COND_TRUE]] ], [ [[TMP2]], [[COND_FALSE]] ]
+// CHECK1-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
+// CHECK1-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
+// CHECK1-NEXT:    store i32 [[TMP3]], ptr [[DOTOMP_IV]], align 4
+// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
+// CHECK1:       omp.inner.for.cond:
+// CHECK1-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
+// CHECK1-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
+// CHECK1-NEXT:    [[CMP3:%.*]] = icmp sle i32 [[TMP4]], [[TMP5]]
+// CHECK1-NEXT:    br i1 [[CMP3]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
+// CHECK1:       omp.inner.for.body:
+// CHECK1-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
+// CHECK1-NEXT:    [[DIV:%.*]] = sdiv i32 [[TMP6]], 16
+// CHECK1-NEXT:    [[MUL:%.*]] = mul nsw i32 [[DIV]], 3
+// CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 7, [[MUL]]
+// CHECK1-NEXT:    store i32 [[ADD]], ptr [[K]], align 4
+// CHECK1-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
+// CHECK1-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
+// CHECK1-NEXT:    [[DIV4:%.*]] = sdiv i32 [[TMP8]], 16
+// CHECK1-NEXT:    [[MUL5:%.*]] = mul nsw i32 [[DIV4]], 16
+// CHECK1-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP7]], [[MUL5]]
+// CHECK1-NEXT:    [[DIV6:%.*]] = sdiv i32 [[SUB]], 4
+// CHECK1-NEXT:    [[MUL7:%.*]] = mul nsw i32 [[DIV6]], 1
+// CHECK1-NEXT:    [[ADD8:%.*]] = add nsw i32 0, [[MUL7]]
+// CHECK1-NEXT:    store i32 [[ADD8]], ptr [[DOTFORWARD_IV_I]], align 4
+// CHECK1-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
+// CHECK1-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
+// CHECK1-NEXT:    [[DIV9:%.*]] = sdiv i32 [[TMP10]], 16
+// CHECK1-NEXT:    [[MUL10:%.*]] = mul nsw i32 [[DIV9]], 16
+// CHECK1-NEXT:    [[SUB11:%.*]] = sub nsw i32 [[TMP9]], [[MUL10]]
+// CHECK1-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
+// CHECK1-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
+// CHECK1-NEXT:    [[DIV12:%.*]] = sdiv i32 [[TMP12]], 16
+// CHECK1-NEXT:    [[MUL13:%.*]] = mul nsw i32 [[DIV12]], 16
+// CHECK1-NEXT:    [[SUB14:%.*]] = sub nsw i32 [[TMP11]], [[MUL13]]
+// CHECK1-NEXT:    [[DIV15:%.*]] = sdiv i32 [[SUB14]], 4
+// CHECK1-NEXT:    [[MUL16:%.*]] = mul nsw i32 [[DIV15]], 4
+// CHECK1-NEXT:    [[SUB17:%.*]] = sub nsw i32 [[SUB11]], [[MUL16]]
+// CHECK1-NEXT:    [[MUL18:%.*]] = mul nsw i32 [[SUB17]], 3
+// CHECK1-NEXT:    [[ADD19:%.*]] = add nsw i32 7, [[MUL18]]
+// CHECK1-NEXT:    store i32 [[ADD19]], ptr [[J]], align 4
+// CHECK1-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTFORWARD_IV_I]], align 4
+// CHECK1-NEXT:    [[SUB20:%.*]] = sub nsw i32 3, [[TMP13]]
+// CHECK1-NEXT:    store i32 [[SUB20]], ptr [[DOTREVERSED_IV_I]], align 4
+// CHECK1-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTREVERSED_IV_I]], align 4
+// CHECK1-NEXT:    [[MUL21:%.*]] = mul nsw i32 [[TMP14]], 3
+// CHECK1-NEXT:    [[ADD22:%.*]] = add nsw i32 7, [[MUL21]]
+// CHECK1-NEXT:    store i32 [[ADD22]], ptr [[I]], align 4
+// CHECK1-NEXT:    [[TMP15:%.*]] = load i32, ptr [[K]], align 4
+// CHECK1-NEXT:    [[TMP16:%.*]] = load i32, ptr [[I]], align 4
+// CHECK1-NEXT:    [[TMP17:%.*]] = load i32, ptr [[J]], align 4
+// CHECK1-NEXT:    call void (...) @body(i32 noundef [[TMP15]], i32 noundef [[TMP16]], i32 noundef [[TMP17]])
+// CHECK1-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
+// CHECK1:       omp.body.continue:
+// CHECK1-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
+// CHECK1:       omp.inner.for.inc:
+// CHECK1-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
+// CHECK1-NEXT:    [[ADD23:%.*]] = add nsw i32 [[TMP18]], 1
+// CHECK1-NEXT:    store i32 [[ADD23]], ptr [[DOTOMP_IV]], align 4
+// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]]
+// CHECK1:       omp.inner.for.end:
+// CHECK1-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
+// CHECK1:       omp.loop.exit:
+// CHECK1-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP0]])
+// CHECK1-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB3]], i32 [[TMP0]])
+// CHECK1-NEXT:    ret void
+//
+//
+// CHECK1-LABEL: define {{[^@]+}}@foo4
+// CHECK1-SAME: () #[[ATTR0]] {
+// CHECK1-NEXT:  entry:
+// CHECK1-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB2]], i32 0, ptr @foo4.omp_outlined)
+// CHECK1-NEXT:    ret void
+//
+//
+// CHECK1-LABEL: define {{[^@]+}}@foo4.omp_outlined
+// CHECK1-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]]) #[[ATTR4:[0-9]+]] {
+// CHECK1-NEXT:  entry:
+// CHECK1-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[TMP:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[I:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[DOTFORWARD_IV_I:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[DOTREVERSED_IV_I:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
+// CHECK1-NEXT:    store i32 7, ptr [[I]], align 4
+// CHECK1-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
+// CHECK1-NEXT:    store i32 3, ptr [[DOTOMP_UB]], align 4
+// CHECK1-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
+// CHECK1-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
+// CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
+// CHECK1-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
+// CHECK1-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP1]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
+// CHECK1-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
+// CHECK1-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP2]], 3
+// CHECK1-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
+// CHECK1:       cond.true:
+// CHECK1-NEXT:    br label [[COND_END:%.*]]
+// CHECK1:       cond.false:
+// CHECK1-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
+// CHECK1-NEXT:    br label [[COND_END]]
+// CHECK1:       cond.end:
+// CHECK1-NEXT:    [[COND:%.*]] = phi i32 [ 3, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
+// CHECK1-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
+// CHECK1-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
+// CHECK1-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_IV]], align 4
+// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
+// CHECK1:       omp.inner.for.cond:
+// CHECK1-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
+// CHECK1-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
+// CHECK1-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
+// CHECK1-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
+// CHECK1:       omp.inner.for.body:
+// CHECK1-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
+// CHECK1-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP7]], 1
+// CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
+// CHECK1-NEXT:    store i32 [[ADD]], ptr [[DOTFORWARD_IV_I]], align 4
+// CHECK1-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTFORWARD_IV_I]], align 4
+// CHECK1-NEXT:    [[SUB:%.*]] = sub nsw i32 3, [[TMP8]]
+// CHECK1-NEXT:    store i32 [[SUB]], ptr [[DOTREVERSED_IV_I]], align 4
+// CHECK1-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTREVERSED_IV_I]], align 4
+// CHECK1-NEXT:    [[MUL2:%.*]] = mul nsw i32 [[TMP9]], 3
+// CHECK1-NEXT:    [[ADD3:%.*]] = add nsw i32 7, [[MUL2]]
+// CHECK1-NEXT:    store i32 [[ADD3]], ptr [[I]], align 4
+// CHECK1-NEXT:    [[TMP10:%.*]] = load i32, ptr [[I]], align 4
+// CHECK1-NEXT:    call void (...) @body(i32 noundef [[TMP10]])
+// CHECK1-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
+// CHECK1:       omp.body.continue:
+// CHECK1-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
+// CHECK1:       omp.inner.for.inc:
+// CHECK1-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
+// CHECK1-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP11]], 1
+// CHECK1-NEXT:    store i32 [[ADD4]], ptr [[DOTOMP_IV]], align 4
+// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]]
+// CHECK1:       omp.inner.for.end:
+// CHECK1-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
+// CHECK1:       omp.loop.exit:
+// CHECK1-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP1]])
+// CHECK1-NEXT:    ret void
+//
+//
+// CHECK1-LABEL: define {{[^@]+}}@tfoo5
+// CHECK1-SAME: () #[[ATTR0]] {
+// CHECK1-NEXT:  entry:
+// CHECK1-NEXT:    call void @_Z4foo5IiTnT_Li3EEvS0_S0_(i32 noundef 0, i32 noundef 42)
+// CHECK1-NEXT:    ret void
+//
+//
+// CHECK1-LABEL: define {{[^@]+}}@_Z4foo5IiTnT_Li3EEvS0_S0_
+// CHECK1-SAME: (i32 noundef [[START:%.*]], i32 noundef [[END:%.*]]) #[[ATTR0]] comdat {
+// CHECK1-NEXT:  entry:
+// CHECK1-NEXT:    [[START_ADDR:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[END_ADDR:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[I:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[DOTFORWARD_IV_I:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[DOTREVERSED_IV_I:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    store i32 [[START]], ptr [[START_ADDR]], align 4
+// CHECK1-NEXT:    store i32 [[END]], ptr [[END_ADDR]], align 4
+// CHECK1-NEXT:    [[TMP0:%.*]] = load i32, ptr [[START_ADDR]], align 4
+// CHECK1-NEXT:    store i32 [[TMP0]], ptr [[I]], align 4
+// CHECK1-NEXT:    [[TMP1:%.*]] = load i32, ptr [[START_ADDR]], align 4
+// CHECK1-NEXT:    store i32 [[TMP1]], ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK1-NEXT:    [[TMP2:%.*]] = load i32, ptr [[END_ADDR]], align 4
+// CHECK1-NEXT:    store i32 [[TMP2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
+// CHECK1-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
+// CHECK1-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK1-NEXT:    [[SUB:%.*]] = sub i32 [[TMP3]], [[TMP4]]
+// CHECK1-NEXT:    [[SUB3:%.*]] = sub i32 [[SUB]], 1
+// CHECK1-NEXT:    [[ADD:%.*]] = add i32 [[SUB3]], 3
+// CHECK1-NEXT:    [[DIV:%.*]] = udiv i32 [[ADD]], 3
+// CHECK1-NEXT:    [[SUB4:%.*]] = sub i32 [[DIV]], 1
+// CHECK1-NEXT:    store i32 [[SUB4]], ptr [[DOTCAPTURE_EXPR_2]], align 4
+// CHECK1-NEXT:    store i32 0, ptr [[DOTFORWARD_IV_I]], align 4
+// CHECK1-NEXT:    br label [[FOR_COND:%.*]]
+// CHECK1:       for.cond:
+// CHECK1-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTFORWARD_IV_I]], align 4
+// CHECK1-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
+// CHECK1-NEXT:    [[ADD5:%.*]] = add i32 [[TMP6]], 1
+// CHECK1-NEXT:    [[CMP:%.*]] = icmp ult i32 [[TMP5]], [[ADD5]]
+// CHECK1-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
+// CHECK1:       for.body:
+// CHECK1-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
+// CHECK1-NEXT:    [[ADD6:%.*]] = add i32 [[TMP7]], 1
+// CHECK1-NEXT:    [[SUB7:%.*]] = sub i32 [[ADD6]], 1
+// CHECK1-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTFORWARD_IV_I]], align 4
+// CHECK1-NEXT:    [[SUB8:%.*]] = sub i32 [[SUB7]], [[TMP8]]
+// CHECK1-NEXT:    store i32 [[SUB8]], ptr [[DOTREVERSED_IV_I]], align 4
+// CHECK1-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK1-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTREVERSED_IV_I]], align 4
+// CHECK1-NEXT:    [[MUL:%.*]] = mul i32 [[TMP10]], 3
+// CHECK1-NEXT:    [[ADD9:%.*]] = add i32 [[TMP9]], [[MUL]]
+// CHECK1-NEXT:    store i32 [[ADD9]], ptr [[I]], align 4
+// CHECK1-NEXT:    [[TMP11:%.*]] = load i32, ptr [[I]], align 4
+// CHECK1-NEXT:    call void (...) @body(i32 noundef [[TMP11]])
+// CHECK1-NEXT:    br label [[FOR_INC:%.*]]
+// CHECK1:       for.inc:
+// CHECK1-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTFORWARD_IV_I]], align 4
+// CHECK1-NEXT:    [[INC:%.*]] = add i32 [[TMP12]], 1
+// CHECK1-NEXT:    store i32 [[INC]], ptr [[DOTFORWARD_IV_I]], align 4
+// CHECK1-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP8:![0-9]+]]
+// CHECK1:       for.end:
+// CHECK1-NEXT:    ret void
+//
+//
+// CHECK1-LABEL: define {{[^@]+}}@foo6
+// CHECK1-SAME: () #[[ATTR0]] {
+// CHECK1-NEXT:  entry:
+// CHECK1-NEXT:    [[ARR:%.*]] = alloca [128 x double], align 16
+// CHECK1-NEXT:    [[C:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[__RANGE2:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[__END2:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[__BEGIN2:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[DOTCAPTURE_EXPR_3:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[DOTCAPTURE_EXPR_4:%.*]] = alloca i64, align 8
+// CHECK1-NEXT:    [[DOTFORWARD_IV___BEGIN2:%.*]] = alloca i64, align 8
+// CHECK1-NEXT:    [[DOTREVERSED_IV___BEGIN2:%.*]] = alloca i64, align 8
+// CHECK1-NEXT:    [[V:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    store i32 42, ptr [[C]], align 4
+// CHECK1-NEXT:    store ptr [[ARR]], ptr [[__RANGE2]], align 8
+// CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[__RANGE2]], align 8
+// CHECK1-NEXT:    [[ARRAYDECAY:%.*]] = getelementptr inbounds [128 x double], ptr [[TMP0]], i64 0, i64 0
+// CHECK1-NEXT:    [[ADD_PTR:%.*]] = getelementptr inbounds double, ptr [[ARRAYDECAY]], i64 128
+// CHECK1-NEXT:    store ptr [[ADD_PTR]], ptr [[__END2]], align 8
+// CHECK1-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[__RANGE2]], align 8
+// CHECK1-NEXT:    [[ARRAYDECAY1:%.*]] = getelementptr inbounds [128 x double], ptr [[TMP1]], i64 0, i64 0
+// CHECK1-NEXT:    store ptr [[ARRAYDECAY1]], ptr [[__BEGIN2]], align 8
+// CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[__RANGE2]], align 8
+// CHECK1-NEXT:    [[ARRAYDECAY2:%.*]] = getelementptr inbounds [128 x double], ptr [[TMP2]], i64 0, i64 0
+// CHECK1-NEXT:    store ptr [[ARRAYDECAY2]], ptr [[DOTCAPTURE_EXPR_]], align 8
+// CHECK1-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[__END2]], align 8
+// CHECK1-NEXT:    store ptr [[TMP3]], ptr [[DOTCAPTURE_EXPR_3]], align 8
+// CHECK1-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[DOTCAPTURE_EXPR_3]], align 8
+// CHECK1-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[DOTCAPTURE_EXPR_]], align 8
+// CHECK1-NEXT:    [[SUB_PTR_LHS_CAST:%.*]] = ptrtoint ptr [[TMP4]] to i64
+// CHECK1-NEXT:    [[SUB_PTR_RHS_CAST:%.*]] = ptrtoint ptr [[TMP5]] to i64
+// CHECK1-NEXT:    [[SUB_PTR_SUB:%.*]] = sub i64 [[SUB_PTR_LHS_CAST]], [[SUB_PTR_RHS_CAST]]
+// CHECK1-NEXT:    [[SUB_PTR_DIV:%.*]] = sdiv exact i64 [[SUB_PTR_SUB]], 8
+// CHECK1-NEXT:    [[SUB:%.*]] = sub nsw i64 [[SUB_PTR_DIV]], 1
+// CHECK1-NEXT:    [[ADD:%.*]] = add nsw i64 [[SUB]], 1
+// CHECK1-NEXT:    [[DIV:%.*]] = sdiv i64 [[ADD]], 1
+// CHECK1-NEXT:    [[SUB5:%.*]] = sub nsw i64 [[DIV]], 1
+// CHECK1-NEXT:    store i64 [[SUB5]], ptr [[DOTCAPTURE_EXPR_4]], align 8
+// CHECK1-NEXT:    store i64 0, ptr [[DOTFORWARD_IV___BEGIN2]], align 8
+// CHECK1-NEXT:    br label [[FOR_COND:%.*]]
+// CHECK1:       for.cond:
+// CHECK1-NEXT:    [[TMP6:%.*]] = load i64, ptr [[DOTFORWARD_IV___BEGIN2]], align 8
+// CHECK1-NEXT:    [[TMP7:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR_4]], align 8
+// CHECK1-NEXT:    [[ADD6:%.*]] = add nsw i64 [[TMP7]], 1
+// CHECK1-NEXT:    [[CMP:%.*]] = icmp slt i64 [[TMP6]], [[ADD6]]
+// CHECK1-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
+// CHECK1:       for.body:
+// CHECK1-NEXT:    [[TMP8:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR_4]], align 8
+// CHECK1-NEXT:    [[ADD7:%.*]] = add nsw i64 [[TMP8]], 1
+// CHECK1-NEXT:    [[SUB8:%.*]] = sub nsw i64 [[ADD7]], 1
+// CHECK1-NEXT:    [[TMP9:%.*]] = load i64, ptr [[DOTFORWARD_IV___BEGIN2]], align 8
+// CHECK1-NEXT:    [[SUB9:%.*]] = sub nsw i64 [[SUB8]], [[TMP9]]
+// CHECK1-NEXT:    store i64 [[SUB9]], ptr [[DOTREVERSED_IV___BEGIN2]], align 8
+// CHECK1-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DOTCAPTURE_EXPR_]], align 8
+// CHECK1-NEXT:    [[TMP11:%.*]] = load i64, ptr [[DOTREVERSED_IV___BEGIN2]], align 8
+// CHECK1-NEXT:    [[MUL:%.*]] = mul nsw i64 [[TMP11]], 1
+// CHECK1-NEXT:    [[ADD_PTR10:%.*]] = getelementptr inbounds double, ptr [[TMP10]], i64 [[MUL]]
+// CHECK1-NEXT:    store ptr [[ADD_PTR10]], ptr [[__BEGIN2]], align 8
+// CHECK1-NEXT:    [[TMP12:%.*]] = load ptr, ptr [[__BEGIN2]], align 8
+// CHECK1-NEXT:    store ptr [[TMP12]], ptr [[V]], align 8
+// CHECK1-NEXT:    [[TMP13:%.*]] = load ptr, ptr [[V]], align 8
+// CHECK1-NEXT:    [[TMP14:%.*]] = load double, ptr [[TMP13]], align 8
+// CHECK1-NEXT:    [[TMP15:%.*]] = load i32, ptr [[C]], align 4
+// CHECK1-NEXT:    call void (...) @body(double noundef [[TMP14]], i32 noundef [[TMP15]])
+// CHECK1-NEXT:    br label [[FOR_INC:%.*]]
+// CHECK1:       for.inc:
+// CHECK1-NEXT:    [[TMP16:%.*]] = load i64, ptr [[DOTFORWARD_IV___BEGIN2]], align 8
+// CHECK1-NEXT:    [[INC:%.*]] = add nsw i64 [[TMP16]], 1
+// CHECK1-NEXT:    store i64 [[INC]], ptr [[DOTFORWARD_IV___BEGIN2]], align 8
+// CHECK1-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP9:![0-9]+]]
+// CHECK1:       for.end:
+// CHECK1-NEXT:    ret void
+//
+//
+// CHECK1-LABEL: define {{[^@]+}}@foo7
+// CHECK1-SAME: () #[[ATTR0]] {
+// CHECK1-NEXT:  entry:
+// CHECK1-NEXT:    [[A:%.*]] = alloca [128 x double], align 16
+// CHECK1-NEXT:    [[DOTOMP_IV:%.*]] = alloca i64, align 8
+// CHECK1-NEXT:    [[TMP:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[_TMP1:%.*]] = alloca i64, align 8
+// CHECK1-NEXT:    [[_TMP2:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[C:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[__RANGE3:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[__END3:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[__BEGIN3:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[DOTCAPTURE_EXPR_5:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[DOTCAPTURE_EXPR_6:%.*]] = alloca i64, align 8
+// CHECK1-NEXT:    [[DOTCAPTURE_EXPR_8:%.*]] = alloca i64, align 8
+// CHECK1-NEXT:    [[DOTCAPTURE_EXPR_10:%.*]] = alloca i64, align 8
+// CHECK1-NEXT:    [[K:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[DOTFORWARD_IV___BEGIN3:%.*]] = alloca i64, align 8
+// CHECK1-NEXT:    [[J:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[DOTOMP_LB:%.*]] = alloca i64, align 8
+// CHECK1-NEXT:    [[DOTOMP_UB:%.*]] = alloca i64, align 8
+// CHECK1-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i64, align 8
+// CHECK1-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[K15:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[DOTFORWARD_IV___BEGIN316:%.*]] = alloca i64, align 8
+// CHECK1-NEXT:    [[J17:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[DOTREVERSED_IV___BEGIN3:%.*]] = alloca i64, align 8
+// CHECK1-NEXT:    [[V:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB2]])
+// CHECK1-NEXT:    store i32 42, ptr [[C]], align 4
+// CHECK1-NEXT:    store ptr [[A]], ptr [[__RANGE3]], align 8
+// CHECK1-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[__RANGE3]], align 8
+// CHECK1-NEXT:    [[ARRAYDECAY:%.*]] = getelementptr inbounds [128 x double], ptr [[TMP1]], i64 0, i64 0
+// CHECK1-NEXT:    [[ADD_PTR:%.*]] = getelementptr inbounds double, ptr [[ARRAYDECAY]], i64 128
+// CHECK1-NEXT:    store ptr [[ADD_PTR]], ptr [[__END3]], align 8
+// CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[__RANGE3]], align 8
+// CHECK1-NEXT:    [[ARRAYDECAY3:%.*]] = getelementptr inbounds [128 x double], ptr [[TMP2]], i64 0, i64 0
+// CHECK1-NEXT:    store ptr [[ARRAYDECAY3]], ptr [[__BEGIN3]], align 8
+// CHECK1-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[__RANGE3]], align 8
+// CHECK1-NEXT:    [[ARRAYDECAY4:%.*]] = getelementptr inbounds [128 x double], ptr [[TMP3]], i64 0, i64 0
+// CHECK1-NEXT:    store ptr [[ARRAYDECAY4]], ptr [[DOTCAPTURE_EXPR_]], align 8
+// CHECK1-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[__END3]], align 8
+// CHECK1-NEXT:    store ptr [[TMP4]], ptr [[DOTCAPTURE_EXPR_5]], align 8
+// CHECK1-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[DOTCAPTURE_EXPR_5]], align 8
+// CHECK1-NEXT:    [[TMP6:%.*]] = load ptr, ptr [[DOTCAPTURE_EXPR_]], align 8
+// CHECK1-NEXT:    [[SUB_PTR_LHS_CAST:%.*]] = ptrtoint ptr [[TMP5]] to i64
+// CHECK1-NEXT:    [[SUB_PTR_RHS_CAST:%.*]] = ptrtoint ptr [[TMP6]] to i64
+// CHECK1-NEXT:    [[SUB_PTR_SUB:%.*]] = sub i64 [[SUB_PTR_LHS_CAST]], [[SUB_PTR_RHS_CAST]]
+// CHECK1-NEXT:    [[SUB_PTR_DIV:%.*]] = sdiv exact i64 [[SUB_PTR_SUB]], 8
+// CHECK1-NEXT:    [[SUB:%.*]] = sub nsw i64 [[SUB_PTR_DIV]], 1
+// CHECK1-NEXT:    [[ADD:%.*]] = add nsw i64 [[SUB]], 1
+// CHECK1-NEXT:    [[DIV:%.*]] = sdiv i64 [[ADD]], 1
+// CHECK1-NEXT:    [[SUB7:%.*]] = sub nsw i64 [[DIV]], 1
+// CHECK1-NEXT:    store i64 [[SUB7]], ptr [[DOTCAPTURE_EXPR_6]], align 8
+// CHECK1-NEXT:    [[TMP7:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR_6]], align 8
+// CHECK1-NEXT:    [[ADD9:%.*]] = add nsw i64 [[TMP7]], 1
+// CHECK1-NEXT:    store i64 [[ADD9]], ptr [[DOTCAPTURE_EXPR_8]], align 8
+// CHECK1-NEXT:    [[TMP8:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR_8]], align 8
+// CHECK1-NEXT:    [[SUB11:%.*]] = sub nsw i64 [[TMP8]], 0
+// CHECK1-NEXT:    [[DIV12:%.*]] = sdiv i64 [[SUB11]], 1
+// CHECK1-NEXT:    [[MUL:%.*]] = mul nsw i64 4, [[DIV12]]
+// CHECK1-NEXT:    [[MUL13:%.*]] = mul nsw i64 [[MUL]], 4
+// CHECK1-NEXT:    [[SUB14:%.*]] = sub nsw i64 [[MUL13]], 1
+// CHECK1-NEXT:    store i64 [[SUB14]], ptr [[DOTCAPTURE_EXPR_10]], align 8
+// CHECK1-NEXT:    store i32 7, ptr [[K]], align 4
+// CHECK1-NEXT:    store i64 0, ptr [[DOTFORWARD_IV___BEGIN3]], align 8
+// CHECK1-NEXT:    store i32 7, ptr [[J]], align 4
+// CHECK1-NEXT:    [[TMP9:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR_8]], align 8
+// CHECK1-NEXT:    [[CMP:%.*]] = icmp slt i64 0, [[TMP9]]
+// CHECK1-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
+// CHECK1:       omp.precond.then:
+// CHECK1-NEXT:    store i64 0, ptr [[DOTOMP_LB]], align 8
+// CHECK1-NEXT:    [[TMP10:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR_10]], align 8
+// CHECK1-NEXT:    store i64 [[TMP10]], ptr [[DOTOMP_UB]], align 8
+// CHECK1-NEXT:    store i64 1, ptr [[DOTOMP_STRIDE]], align 8
+// CHECK1-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
+// CHECK1-NEXT:    call void @__kmpc_for_static_init_8(ptr @[[GLOB1]], i32 [[TMP0]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i64 1, i64 1)
+// CHECK1-NEXT:    [[TMP11:%.*]] = load i64, ptr [[DOTOMP_UB]], align 8
+// CHECK1-NEXT:    [[TMP12:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR_10]], align 8
+// CHECK1-NEXT:    [[CMP18:%.*]] = icmp sgt i64 [[TMP11]], [[TMP12]]
+// CHECK1-NEXT:    br i1 [[CMP18]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
+// CHECK1:       cond.true:
+// CHECK1-NEXT:    [[TMP13:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR_10]], align 8
+// CHECK1-NEXT:    br label [[COND_END:%.*]]
+// CHECK1:       cond.false:
+// CHECK1-NEXT:    [[TMP14:%.*]] = load i64, ptr [[DOTOMP_UB]], align 8
+// CHECK1-NEXT:    br label [[COND_END]]
+// CHECK1:       cond.end:
+// CHECK1-NEXT:    [[COND:%.*]] = phi i64 [ [[TMP13]], [[COND_TRUE]] ], [ [[TMP14]], [[COND_FALSE]] ]
+// CHECK1-NEXT:    store i64 [[COND]], ptr [[DOTOMP_UB]], align 8
+// CHECK1-NEXT:    [[TMP15:%.*]] = load i64, ptr [[DOTOMP_LB]], align 8
+// CHECK1-NEXT:    store i64 [[TMP15]], ptr [[DOTOMP_IV]], align 8
+// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
+// CHECK1:       omp.inner.for.cond:
+// CHECK1-NEXT:    [[TMP16:%.*]] = load i64, ptr [[DOTOMP_IV]], align 8
+// CHECK1-NEXT:    [[TMP17:%.*]] = load i64, ptr [[DOTOMP_UB]], align 8
+// CHECK1-NEXT:    [[CMP19:%.*]] = icmp sle i64 [[TMP16]], [[TMP17]]
+// CHECK1-NEXT:    br i1 [[CMP19]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
+// CHECK1:       omp.inner.for.body:
+// CHECK1-NEXT:    [[TMP18:%.*]] = load i64, ptr [[DOTOMP_IV]], align 8
+// CHECK1-NEXT:    [[TMP19:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR_8]], align 8
+// CHECK1-NEXT:    [[SUB20:%.*]] = sub nsw i64 [[TMP19]], 0
+// CHECK1-NEXT:    [[DIV21:%.*]] = sdiv i64 [[SUB20]], 1
+// CHECK1-NEXT:    [[MUL22:%.*]] = mul nsw i64 1, [[DIV21]]
+// CHECK1-NEXT:    [[MUL23:%.*]] = mul nsw i64 [[MUL22]], 4
+// CHECK1-NEXT:    [[DIV24:%.*]] = sdiv i64 [[TMP18]], [[MUL23]]
+// CHECK1-NEXT:    [[MUL25:%.*]] = mul nsw i64 [[DIV24]], 3
+// CHECK1-NEXT:    [[ADD26:%.*]] = add nsw i64 7, [[MUL25]]
+// CHECK1-NEXT:    [[CONV:%.*]] = trunc i64 [[ADD26]] to i32
+// CHECK1-NEXT:    store i32 [[CONV]], ptr [[K15]], align 4
+// CHECK1-NEXT:    [[TMP20:%.*]] = load i64, ptr [[DOTOMP_IV]], align 8
+// CHECK1-NEXT:    [[TMP21:%.*]] = load i64, ptr [[DOTOMP_IV]], align 8
+// CHECK1-NEXT:    [[TMP22:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR_8]], align 8
+// CHECK1-NEXT:    [[SUB27:%.*]] = sub nsw i64 [[TMP22]], 0
+// CHECK1-NEXT:    [[DIV28:%.*]] = sdiv i64 [[SUB27]], 1
+// CHECK1-NEXT:    [[MUL29:%.*]] = mul nsw i64 1, [[DIV28]]
+// CHECK1-NEXT:    [[MUL30:%.*]] = mul nsw i64 [[MUL29]], 4
+// CHECK1-NEXT:    [[DIV31:%.*]] = sdiv i64 [[TMP21]], [[MUL30]]
+// CHECK1-NEXT:    [[TMP23:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR_8]], align 8
+// CHECK1-NEXT:    [[SUB32:%.*]] = sub nsw i64 [[TMP23]], 0
+// CHECK1-NEXT:    [[DIV33:%.*]] = sdiv i64 [[SUB32]], 1
+// CHECK1-NEXT:    [[MUL34:%.*]] = mul nsw i64 1, [[DIV33]]
+// CHECK1-NEXT:    [[MUL35:%.*]] = mul nsw i64 [[MUL34]], 4
+// CHECK1-NEXT:    [[MUL36:%.*]] = mul nsw i64 [[DIV31]], [[MUL35]]
+// CHECK1-NEXT:    [[SUB37:%.*]] = sub nsw i64 [[TMP20]], [[MUL36]]
+// CHECK1-NEXT:    [[DIV38:%.*]] = sdiv i64 [[SUB37]], 4
+// CHECK1-NEXT:    [[MUL39:%.*]] = mul nsw i64 [[DIV38]], 1
+// CHECK1-NEXT:    [[ADD40:%.*]] = add nsw i64 0, [[MUL39]]
+// CHECK1-NEXT:    store i64 [[ADD40]], ptr [[DOTFORWARD_IV___BEGIN316]], align 8
+// CHECK1-NEXT:    [[TMP24:%.*]] = load i64, ptr [[DOTOMP_IV]], align 8
+// CHECK1-NEXT:    [[TMP25:%.*]] = load i64, ptr [[DOTOMP_IV]], align 8
+// CHECK1-NEXT:    [[TMP26:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR_8]], align 8
+// CHECK1-NEXT:    [[SUB41:%.*]] = sub nsw i64 [[TMP26]], 0
+// CHECK1-NEXT:    [[DIV42:%.*]] = sdiv i64 [[SUB41]], 1
+// CHECK1-NEXT:    [[MUL43:%.*]] = mul nsw i64 1, [[DIV42]]
+// CHECK1-NEXT:    [[MUL44:%.*]] = mul nsw i64 [[MUL43]], 4
+// CHECK1-NEXT:    [[DIV45:%.*]] = sdiv i64 [[TMP25]], [[MUL44]]
+// CHECK1-NEXT:    [[TMP27:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR_8]], align 8
+// CHECK1-NEXT:    [[SUB46:%.*]] = sub nsw i64 [[TMP27]], 0
+// CHECK1-NEXT:    [[DIV47:%.*]] = sdiv i64 [[SUB46]], 1
+// CHECK1-NEXT:    [[MUL48:%.*]] = mul nsw i64 1, [[DIV47]]
+// CHECK1-NEXT:    [[MUL49:%.*]] = mul nsw i64 [[MUL48]], 4
+// CHECK1-NEXT:    [[MUL50:%.*]] = mul nsw i64 [[DIV45]], [[MUL49]]
+// CHECK1-NEXT:    [[SUB51:%.*]] = sub nsw i64 [[TMP24]], [[MUL50]]
+// CHECK1-NEXT:    [[TMP28:%.*]] = load i64, ptr [[DOTOMP_IV]], align 8
+// CHECK1-NEXT:    [[TMP29:%.*]] = load i64, ptr [[DOTOMP_IV]], align 8
+// CHECK1-NEXT:    [[TMP30:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR_8]], align 8
+// CHECK1-NEXT:    [[SUB52:%.*]] = sub nsw i64 [[TMP30]], 0
+// CHECK1-NEXT:    [[DIV53:%.*]] = sdiv i64 [[SUB52]], 1
+// CHECK1-NEXT:    [[MUL54:%.*]] = mul nsw i64 1, [[DIV53]]
+// CHECK1-NEXT:    [[MUL55:%.*]] = mul nsw i64 [[MUL54]], 4
+// CHECK1-NEXT:    [[DIV56:%.*]] = sdiv i64 [[TMP29]], [[MUL55]]
+// CHECK1-NEXT:    [[TMP31:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR_8]], align 8
+// CHECK1-NEXT:    [[SUB57:%.*]] = sub nsw i64 [[TMP31]], 0
+// CHECK1-NEXT:    [[DIV58:%.*]] = sdiv i64 [[SUB57]], 1
+// CHECK1-NEXT:    [[MUL59:%.*]] = mul nsw i64 1, [[DIV58]]
+// CHECK1-NEXT:    [[MUL60:%.*]] = mul nsw i64 [[MUL59]], 4
+// CHECK1-NEXT:    [[MUL61:%.*]] = mul nsw i64 [[DIV56]], [[MUL60]]
+// CHECK1-NEXT:    [[SUB62:%.*]] = sub nsw i64 [[TMP28]], [[MUL61]]
+// CHECK1-NEXT:    [[DIV63:%.*]] = sdiv i64 [[SUB62]], 4
+// CHECK1-NEXT:    [[MUL64:%.*]] = mul nsw i64 [[DIV63]], 4
+// CHECK1-NEXT:    [[SUB65:%.*]] = sub nsw i64 [[SUB51]], [[MUL64]]
+// CHECK1-NEXT:    [[MUL66:%.*]] = mul nsw i64 [[SUB65]], 3
+// CHECK1-NEXT:    [[ADD67:%.*]] = add nsw i64 7, [[MUL66]]
+// CHECK1-NEXT:    [[CONV68:%.*]] = trunc i64 [[ADD67]] to i32
+// CHECK1-NEXT:    store i32 [[CONV68]], ptr [[J17]], align 4
+// CHECK1-NEXT:    [[TMP32:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR_6]], align 8
+// CHECK1-NEXT:    [[ADD69:%.*]] = add nsw i64 [[TMP32]], 1
+// CHECK1-NEXT:    [[SUB70:%.*]] = sub nsw i64 [[ADD69]], 1
+// CHECK1-NEXT:    [[TMP33:%.*]] = load i64, ptr [[DOTFORWARD_IV___BEGIN316]], align 8
+// CHECK1-NEXT:    [[SUB71:%.*]] = sub nsw i64 [[SUB70]], [[TMP33]]
+// CHECK1-NEXT:    store i64 [[SUB71]], ptr [[DOTREVERSED_IV___BEGIN3]], align 8
+// CHECK1-NEXT:    [[TMP34:%.*]] = load ptr, ptr [[DOTCAPTURE_EXPR_]], align 8
+// CHECK1-NEXT:    [[TMP35:%.*]] = load i64, ptr [[DOTREVERSED_IV___BEGIN3]], align 8
+// CHECK1-NEXT:    [[MUL72:%.*]] = mul nsw i64 [[TMP35]], 1
+// CHECK1-NEXT:    [[ADD_PTR73:%.*]] = getelementptr inbounds double, ptr [[TMP34]], i64 [[MUL72]]
+// CHECK1-NEXT:    store ptr [[ADD_PTR73]], ptr [[__BEGIN3]], align 8
+// CHECK1-NEXT:    [[TMP36:%.*]] = load ptr, ptr [[__BEGIN3]], align 8
+// CHECK1-NEXT:    store ptr [[TMP36]], ptr [[V]], align 8
+// CHECK1-NEXT:    [[TMP37:%.*]] = load i32, ptr [[K15]], align 4
+// CHECK1-NEXT:    [[TMP38:%.*]] = load i32, ptr [[C]], align 4
+// CHECK1-NEXT:    [[TMP39:%.*]] = load ptr, ptr [[V]], align 8
+// CHECK1-NEXT:    [[TMP40:%.*]] = load double, ptr [[TMP39]], align 8
+// CHECK1-NEXT:    [[TMP41:%.*]] = load i32, ptr [[J17]], align 4
+// CHECK1-NEXT:    call void (...) @body(i32 noundef [[TMP37]], i32 noundef [[TMP38]], double noundef [[TMP40]], i32 noundef [[TMP41]])
+// CHECK1-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
+// CHECK1:       omp.body.continue:
+// CHECK1-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
+// CHECK1:       omp.inner.for.inc:
+// CHECK1-NEXT:    [[TMP42:%.*]] = load i64, ptr [[DOTOMP_IV]], align 8
+// CHECK1-NEXT:    [[ADD74:%.*]] = add nsw i64 [[TMP42]], 1
+// CHECK1-NEXT:    store i64 [[ADD74]], ptr [[DOTOMP_IV]], align 8
+// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]]
+// CHECK1:       omp.inner.for.end:
+// CHECK1-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
+// CHECK1:       omp.loop.exit:
+// CHECK1-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP0]])
+// CHECK1-NEXT:    br label [[OMP_PRECOND_END]]
+// CHECK1:       omp.precond.end:
+// CHECK1-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB3]], i32 [[TMP0]])
+// CHECK1-NEXT:    ret void
+//
+//
+// CHECK1-LABEL: define {{[^@]+}}@_GLOBAL__sub_I_reverse_codegen.cpp
+// CHECK1-SAME: () #[[ATTR1]] section ".text.startup" {
+// CHECK1-NEXT:  entry:
+// CHECK1-NEXT:    call void @__cxx_global_var_init()
+// CHECK1-NEXT:    ret void
+//
+//
+// CHECK2-LABEL: define {{[^@]+}}@__cxx_global_var_init
+// CHECK2-SAME: () #[[ATTR0:[0-9]+]] section ".text.startup" {
+// CHECK2-NEXT:  entry:
+// CHECK2-NEXT:    call void @_ZN1SC1Ev(ptr noundef nonnull align 4 dereferenceable(4) @s)
+// CHECK2-NEXT:    ret void
+//
+//
+// CHECK2-LABEL: define {{[^@]+}}@_ZN1SC1Ev
+// CHECK2-SAME: (ptr noundef nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR1:[0-9]+]] comdat align 2 {
+// CHECK2-NEXT:  entry:
+// CHECK2-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
+// CHECK2-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
+// CHECK2-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
+// CHECK2-NEXT:    call void @_ZN1SC2Ev(ptr noundef nonnull align 4 dereferenceable(4) [[THIS1]])
+// CHECK2-NEXT:    ret void
+//
+//
+// CHECK2-LABEL: define {{[^@]+}}@_ZN1SC2Ev
+// CHECK2-SAME: (ptr noundef nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat align 2 {
+// CHECK2-NEXT:  entry:
+// CHECK2-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
+// CHECK2-NEXT:    [[I2:%.*]] = alloca ptr, align 8
+// CHECK2-NEXT:    [[DOTFORWARD_IV_I:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[DOTREVERSED_IV_I:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
+// CHECK2-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
+// CHECK2-NEXT:    [[I:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], ptr [[THIS1]], i32 0, i32 0
+// CHECK2-NEXT:    store i32 7, ptr [[I]], align 4
+// CHECK2-NEXT:    [[I3:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[THIS1]], i32 0, i32 0
+// CHECK2-NEXT:    store ptr [[I3]], ptr [[I2]], align 8
+// CHECK2-NEXT:    store i32 0, ptr [[DOTFORWARD_IV_I]], align 4
+// CHECK2-NEXT:    br label [[FOR_COND:%.*]]
+// CHECK2:       for.cond:
+// CHECK2-NEXT:    [[TMP0:%.*]] = load i32, ptr [[DOTFORWARD_IV_I]], align 4
+// CHECK2-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP0]], 4
+// CHECK2-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
+// CHECK2:       for.body:
+// CHECK2-NEXT:    [[TMP1:%.*]] = load i32, ptr [[DOTFORWARD_IV_I]], align 4
+// CHECK2-NEXT:    [[SUB:%.*]] = sub nsw i32 3, [[TMP1]]
+// CHECK2-NEXT:    store i32 [[SUB]], ptr [[DOTREVERSED_IV_I]], align 4
+// CHECK2-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTREVERSED_IV_I]], align 4
+// CHECK2-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP2]], 3
+// CHECK2-NEXT:    [[ADD:%.*]] = add nsw i32 7, [[MUL]]
+// CHECK2-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[I2]], align 8
+// CHECK2-NEXT:    store i32 [[ADD]], ptr [[TMP3]], align 4
+// CHECK2-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[I2]], align 8
+// CHECK2-NEXT:    [[TMP5:%.*]] = load i32, ptr [[TMP4]], align 4
+// CHECK2-NEXT:    call void (...) @body(i32 noundef [[TMP5]])
+// CHECK2-NEXT:    br label [[FOR_INC:%.*]]
+// CHECK2:       for.inc:
+// CHECK2-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTFORWARD_IV_I]], align 4
+// CHECK2-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP6]], 1
+// CHECK2-NEXT:    store i32 [[INC]], ptr [[DOTFORWARD_IV_I]], align 4
+// CHECK2-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP3:![0-9]+]]
+// CHECK2:       for.end:
+// CHECK2-NEXT:    ret void
+//
+//
+// CHECK2-LABEL: define {{[^@]+}}@body
+// CHECK2-SAME: (...) #[[ATTR1]] {
+// CHECK2-NEXT:  entry:
+// CHECK2-NEXT:    ret void
+//
+//
+// CHECK2-LABEL: define {{[^@]+}}@foo1
+// CHECK2-SAME: (i32 noundef [[START:%.*]], i32 noundef [[END:%.*]], i32 noundef [[STEP:%.*]]) #[[ATTR1]] {
+// CHECK2-NEXT:  entry:
+// CHECK2-NEXT:    [[START_ADDR:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[END_ADDR:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[STEP_ADDR:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[I:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[DOTNEW_STEP:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[DOTFORWARD_IV_I:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[DOTREVERSED_IV_I:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    store i32 [[START]], ptr [[START_ADDR]], align 4
+// CHECK2-NEXT:    store i32 [[END]], ptr [[END_ADDR]], align 4
+// CHECK2-NEXT:    store i32 [[STEP]], ptr [[STEP_ADDR]], align 4
+// CHECK2-NEXT:    [[TMP0:%.*]] = load i32, ptr [[START_ADDR]], align 4
+// CHECK2-NEXT:    store i32 [[TMP0]], ptr [[I]], align 4
+// CHECK2-NEXT:    [[TMP1:%.*]] = load i32, ptr [[START_ADDR]], align 4
+// CHECK2-NEXT:    store i32 [[TMP1]], ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK2-NEXT:    [[TMP2:%.*]] = load i32, ptr [[END_ADDR]], align 4
+// CHECK2-NEXT:    store i32 [[TMP2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
+// CHECK2-NEXT:    [[TMP3:%.*]] = load i32, ptr [[STEP_ADDR]], align 4
+// CHECK2-NEXT:    store i32 [[TMP3]], ptr [[DOTNEW_STEP]], align 4
+// CHECK2-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
+// CHECK2-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK2-NEXT:    [[SUB:%.*]] = sub i32 [[TMP4]], [[TMP5]]
+// CHECK2-NEXT:    [[SUB3:%.*]] = sub i32 [[SUB]], 1
+// CHECK2-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTNEW_STEP]], align 4
+// CHECK2-NEXT:    [[ADD:%.*]] = add i32 [[SUB3]], [[TMP6]]
+// CHECK2-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTNEW_STEP]], align 4
+// CHECK2-NEXT:    [[DIV:%.*]] = udiv i32 [[ADD]], [[TMP7]]
+// CHECK2-NEXT:    [[SUB4:%.*]] = sub i32 [[DIV]], 1
+// CHECK2-NEXT:    store i32 [[SUB4]], ptr [[DOTCAPTURE_EXPR_2]], align 4
+// CHECK2-NEXT:    store i32 0, ptr [[DOTFORWARD_IV_I]], align 4
+// CHECK2-NEXT:    br label [[FOR_COND:%.*]]
+// CHECK2:       for.cond:
+// CHECK2-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTFORWARD_IV_I]], align 4
+// CHECK2-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
+// CHECK2-NEXT:    [[ADD5:%.*]] = add i32 [[TMP9]], 1
+// CHECK2-NEXT:    [[CMP:%.*]] = icmp ult i32 [[TMP8]], [[ADD5]]
+// CHECK2-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
+// CHECK2:       for.body:
+// CHECK2-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
+// CHECK2-NEXT:    [[ADD6:%.*]] = add i32 [[TMP10]], 1
+// CHECK2-NEXT:    [[SUB7:%.*]] = sub i32 [[ADD6]], 1
+// CHECK2-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTFORWARD_IV_I]], align 4
+// CHECK2-NEXT:    [[SUB8:%.*]] = sub i32 [[SUB7]], [[TMP11]]
+// CHECK2-NEXT:    store i32 [[SUB8]], ptr [[DOTREVERSED_IV_I]], align 4
+// CHECK2-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK2-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTREVERSED_IV_I]], align 4
+// CHECK2-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTNEW_STEP]], align 4
+// CHECK2-NEXT:    [[MUL:%.*]] = mul i32 [[TMP13]], [[TMP14]]
+// CHECK2-NEXT:    [[ADD9:%.*]] = add i32 [[TMP12]], [[MUL]]
+// CHECK2-NEXT:    store i32 [[ADD9]], ptr [[I]], align 4
+// CHECK2-NEXT:    [[TMP15:%.*]] = load i32, ptr [[I]], align 4
+// CHECK2-NEXT:    call void (...) @body(i32 noundef [[TMP15]])
+// CHECK2-NEXT:    br label [[FOR_INC:%.*]]
+// CHECK2:       for.inc:
+// CHECK2-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTFORWARD_IV_I]], align 4
+// CHECK2-NEXT:    [[INC:%.*]] = add i32 [[TMP16]], 1
+// CHECK2-NEXT:    store i32 [[INC]], ptr [[DOTFORWARD_IV_I]], align 4
+// CHECK2-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP5:![0-9]+]]
+// CHECK2:       for.end:
+// CHECK2-NEXT:    ret void
+//
+//
+// CHECK2-LABEL: define {{[^@]+}}@foo2
+// CHECK2-SAME: () #[[ATTR1]] {
+// CHECK2-NEXT:  entry:
+// CHECK2-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[TMP:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[I:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[DOTFORWARD_IV_I:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[DOTREVERSED_IV_I:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB2:[0-9]+]])
+// CHECK2-NEXT:    store i32 7, ptr [[I]], align 4
+// CHECK2-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
+// CHECK2-NEXT:    store i32 3, ptr [[DOTOMP_UB]], align 4
+// CHECK2-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
+// CHECK2-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
+// CHECK2-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1:[0-9]+]], i32 [[TMP0]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
+// CHECK2-NEXT:    [[TMP1:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
+// CHECK2-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP1]], 3
+// CHECK2-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
+// CHECK2:       cond.true:
+// CHECK2-NEXT:    br label [[COND_END:%.*]]
+// CHECK2:       cond.false:
+// CHECK2-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
+// CHECK2-NEXT:    br label [[COND_END]]
+// CHECK2:       cond.end:
+// CHECK2-NEXT:    [[COND:%.*]] = phi i32 [ 3, [[COND_TRUE]] ], [ [[TMP2]], [[COND_FALSE]] ]
+// CHECK2-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
+// CHECK2-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
+// CHECK2-NEXT:    store i32 [[TMP3]], ptr [[DOTOMP_IV]], align 4
+// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
+// CHECK2:       omp.inner.for.cond:
+// CHECK2-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
+// CHECK2-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
+// CHECK2-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP4]], [[TMP5]]
+// CHECK2-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
+// CHECK2:       omp.inner.for.body:
+// CHECK2-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
+// CHECK2-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP6]], 1
+// CHECK2-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
+// CHECK2-NEXT:    store i32 [[ADD]], ptr [[DOTFORWARD_IV_I]], align 4
+// CHECK2-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTFORWARD_IV_I]], align 4
+// CHECK2-NEXT:    [[SUB:%.*]] = sub nsw i32 3, [[TMP7]]
+// CHECK2-NEXT:    store i32 [[SUB]], ptr [[DOTREVERSED_IV_I]], align 4
+// CHECK2-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTREVERSED_IV_I]], align 4
+// CHECK2-NEXT:    [[MUL2:%.*]] = mul nsw i32 [[TMP8]], 3
+// CHECK2-NEXT:    [[ADD3:%.*]] = add nsw i32 7, [[MUL2]]
+// CHECK2-NEXT:    store i32 [[ADD3]], ptr [[I]], align 4
+// CHECK2-NEXT:    [[TMP9:%.*]] = load i32, ptr [[I]], align 4
+// CHECK2-NEXT:    call void (...) @body(i32 noundef [[TMP9]])
+// CHECK2-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
+// CHECK2:       omp.body.continue:
+// CHECK2-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
+// CHECK2:       omp.inner.for.inc:
+// CHECK2-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
+// CHECK2-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP10]], 1
+// CHECK2-NEXT:    store i32 [[ADD4]], ptr [[DOTOMP_IV]], align 4
+// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND]]
+// CHECK2:       omp.inner.for.end:
+// CHECK2-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
+// CHECK2:       omp.loop.exit:
+// CHECK2-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP0]])
+// CHECK2-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB3:[0-9]+]], i32 [[TMP0]])
+// CHECK2-NEXT:    ret void
+//
+//
+// CHECK2-LABEL: define {{[^@]+}}@foo3
+// CHECK2-SAME: () #[[ATTR1]] {
+// CHECK2-NEXT:  entry:
+// CHECK2-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[TMP:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[_TMP2:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[I:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[K:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[DOTFORWARD_IV_I:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[J:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[DOTREVERSED_IV_I:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB2]])
+// CHECK2-NEXT:    store i32 7, ptr [[I]], align 4
+// CHECK2-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
+// CHECK2-NEXT:    store i32 63, ptr [[DOTOMP_UB]], align 4
+// CHECK2-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
+// CHECK2-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
+// CHECK2-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP0]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
+// CHECK2-NEXT:    [[TMP1:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
+// CHECK2-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP1]], 63
+// CHECK2-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
+// CHECK2:       cond.true:
+// CHECK2-NEXT:    br label [[COND_END:%.*]]
+// CHECK2:       cond.false:
+// CHECK2-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
+// CHECK2-NEXT:    br label [[COND_END]]
+// CHECK2:       cond.end:
+// CHECK2-NEXT:    [[COND:%.*]] = phi i32 [ 63, [[COND_TRUE]] ], [ [[TMP2]], [[COND_FALSE]] ]
+// CHECK2-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
+// CHECK2-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
+// CHECK2-NEXT:    store i32 [[TMP3]], ptr [[DOTOMP_IV]], align 4
+// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
+// CHECK2:       omp.inner.for.cond:
+// CHECK2-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
+// CHECK2-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
+// CHECK2-NEXT:    [[CMP3:%.*]] = icmp sle i32 [[TMP4]], [[TMP5]]
+// CHECK2-NEXT:    br i1 [[CMP3]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
+// CHECK2:       omp.inner.for.body:
+// CHECK2-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
+// CHECK2-NEXT:    [[DIV:%.*]] = sdiv i32 [[TMP6]], 16
+// CHECK2-NEXT:    [[MUL:%.*]] = mul nsw i32 [[DIV]], 3
+// CHECK2-NEXT:    [[ADD:%.*]] = add nsw i32 7, [[MUL]]
+// CHECK2-NEXT:    store i32 [[ADD]], ptr [[K]], align 4
+// CHECK2-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
+// CHECK2-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
+// CHECK2-NEXT:    [[DIV4:%.*]] = sdiv i32 [[TMP8]], 16
+// CHECK2-NEXT:    [[MUL5:%.*]] = mul nsw i32 [[DIV4]], 16
+// CHECK2-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP7]], [[MUL5]]
+// CHECK2-NEXT:    [[DIV6:%.*]] = sdiv i32 [[SUB]], 4
+// CHECK2-NEXT:    [[MUL7:%.*]] = mul nsw i32 [[DIV6]], 1
+// CHECK2-NEXT:    [[ADD8:%.*]] = add nsw i32 0, [[MUL7]]
+// CHECK2-NEXT:    store i32 [[ADD8]], ptr [[DOTFORWARD_IV_I]], align 4
+// CHECK2-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
+// CHECK2-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
+// CHECK2-NEXT:    [[DIV9:%.*]] = sdiv i32 [[TMP10]], 16
+// CHECK2-NEXT:    [[MUL10:%.*]] = mul nsw i32 [[DIV9]], 16
+// CHECK2-NEXT:    [[SUB11:%.*]] = sub nsw i32 [[TMP9]], [[MUL10]]
+// CHECK2-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
+// CHECK2-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
+// CHECK2-NEXT:    [[DIV12:%.*]] = sdiv i32 [[TMP12]], 16
+// CHECK2-NEXT:    [[MUL13:%.*]] = mul nsw i32 [[DIV12]], 16
+// CHECK2-NEXT:    [[SUB14:%.*]] = sub nsw i32 [[TMP11]], [[MUL13]]
+// CHECK2-NEXT:    [[DIV15:%.*]] = sdiv i32 [[SUB14]], 4
+// CHECK2-NEXT:    [[MUL16:%.*]] = mul nsw i32 [[DIV15]], 4
+// CHECK2-NEXT:    [[SUB17:%.*]] = sub nsw i32 [[SUB11]], [[MUL16]]
+// CHECK2-NEXT:    [[MUL18:%.*]] = mul nsw i32 [[SUB17]], 3
+// CHECK2-NEXT:    [[ADD19:%.*]] = add nsw i32 7, [[MUL18]]
+// CHECK2-NEXT:    store i32 [[ADD19]], ptr [[J]], align 4
+// CHECK2-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTFORWARD_IV_I]], align 4
+// CHECK2-NEXT:    [[SUB20:%.*]] = sub nsw i32 3, [[TMP13]]
+// CHECK2-NEXT:    store i32 [[SUB20]], ptr [[DOTREVERSED_IV_I]], align 4
+// CHECK2-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTREVERSED_IV_I]], align 4
+// CHECK2-NEXT:    [[MUL21:%.*]] = mul nsw i32 [[TMP14]], 3
+// CHECK2-NEXT:    [[ADD22:%.*]] = add nsw i32 7, [[MUL21]]
+// CHECK2-NEXT:    store i32 [[ADD22]], ptr [[I]], align 4
+// CHECK2-NEXT:    [[TMP15:%.*]] = load i32, ptr [[K]], align 4
+// CHECK2-NEXT:    [[TMP16:%.*]] = load i32, ptr [[I]], align 4
+// CHECK2-NEXT:    [[TMP17:%.*]] = load i32, ptr [[J]], align 4
+// CHECK2-NEXT:    call void (...) @body(i32 noundef [[TMP15]], i32 noundef [[TMP16]], i32 noundef [[TMP17]])
+// CHECK2-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
+// CHECK2:       omp.body.continue:
+// CHECK2-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
+// CHECK2:       omp.inner.for.inc:
+// CHECK2-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
+// CHECK2-NEXT:    [[ADD23:%.*]] = add nsw i32 [[TMP18]], 1
+// CHECK2-NEXT:    store i32 [[ADD23]], ptr [[DOTOMP_IV]], align 4
+// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND]]
+// CHECK2:       omp.inner.for.end:
+// CHECK2-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
+// CHECK2:       omp.loop.exit:
+// CHECK2-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP0]])
+// CHECK2-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB3]], i32 [[TMP0]])
+// CHECK2-NEXT:    ret void
+//
+//
+// CHECK2-LABEL: define {{[^@]+}}@foo4
+// CHECK2-SAME: () #[[ATTR1]] {
+// CHECK2-NEXT:  entry:
+// CHECK2-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB2]], i32 0, ptr @foo4.omp_outlined)
+// CHECK2-NEXT:    ret void
+//
+//
+// CHECK2-LABEL: define {{[^@]+}}@foo4.omp_outlined
+// CHECK2-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]]) #[[ATTR4:[0-9]+]] {
+// CHECK2-NEXT:  entry:
+// CHECK2-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
+// CHECK2-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
+// CHECK2-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[TMP:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[I:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[DOTFORWARD_IV_I:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[DOTREVERSED_IV_I:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
+// CHECK2-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
+// CHECK2-NEXT:    store i32 7, ptr [[I]], align 4
+// CHECK2-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
+// CHECK2-NEXT:    store i32 3, ptr [[DOTOMP_UB]], align 4
+// CHECK2-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
+// CHECK2-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
+// CHECK2-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
+// CHECK2-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
+// CHECK2-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP1]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
+// CHECK2-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
+// CHECK2-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP2]], 3
+// CHECK2-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
+// CHECK2:       cond.true:
+// CHECK2-NEXT:    br label [[COND_END:%.*]]
+// CHECK2:       cond.false:
+// CHECK2-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
+// CHECK2-NEXT:    br label [[COND_END]]
+// CHECK2:       cond.end:
+// CHECK2-NEXT:    [[COND:%.*]] = phi i32 [ 3, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
+// CHECK2-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
+// CHECK2-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
+// CHECK2-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_IV]], align 4
+// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
+// CHECK2:       omp.inner.for.cond:
+// CHECK2-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
+// CHECK2-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
+// CHECK2-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
+// CHECK2-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
+// CHECK2:       omp.inner.for.body:
+// CHECK2-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
+// CHECK2-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP7]], 1
+// CHECK2-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
+// CHECK2-NEXT:    store i32 [[ADD]], ptr [[DOTFORWARD_IV_I]], align 4
+// CHECK2-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTFORWARD_IV_I]], align 4
+// CHECK2-NEXT:    [[SUB:%.*]] = sub nsw i32 3, [[TMP8]]
+// CHECK2-NEXT:    store i32 [[SUB]], ptr [[DOTREVERSED_IV_I]], align 4
+// CHECK2-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTREVERSED_IV_I]], align 4
+// CHECK2-NEXT:    [[MUL2:%.*]] = mul nsw i32 [[TMP9]], 3
+// CHECK2-NEXT:    [[ADD3:%.*]] = add nsw i32 7, [[MUL2]]
+// CHECK2-NEXT:    store i32 [[ADD3]], ptr [[I]], align 4
+// CHECK2-NEXT:    [[TMP10:%.*]] = load i32, ptr [[I]], align 4
+// CHECK2-NEXT:    call void (...) @body(i32 noundef [[TMP10]])
+// CHECK2-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
+// CHECK2:       omp.body.continue:
+// CHECK2-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
+// CHECK2:       omp.inner.for.inc:
+// CHECK2-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
+// CHECK2-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP11]], 1
+// CHECK2-NEXT:    store i32 [[ADD4]], ptr [[DOTOMP_IV]], align 4
+// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND]]
+// CHECK2:       omp.inner.for.end:
+// CHECK2-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
+// CHECK2:       omp.loop.exit:
+// CHECK2-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP1]])
+// CHECK2-NEXT:    ret void
+//
+//
+// CHECK2-LABEL: define {{[^@]+}}@foo6
+// CHECK2-SAME: () #[[ATTR1]] {
+// CHECK2-NEXT:  entry:
+// CHECK2-NEXT:    [[ARR:%.*]] = alloca [128 x double], align 16
+// CHECK2-NEXT:    [[C:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[__RANGE2:%.*]] = alloca ptr, align 8
+// CHECK2-NEXT:    [[__END2:%.*]] = alloca ptr, align 8
+// CHECK2-NEXT:    [[__BEGIN2:%.*]] = alloca ptr, align 8
+// CHECK2-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca ptr, align 8
+// CHECK2-NEXT:    [[DOTCAPTURE_EXPR_3:%.*]] = alloca ptr, align 8
+// CHECK2-NEXT:    [[DOTCAPTURE_EXPR_4:%.*]] = alloca i64, align 8
+// CHECK2-NEXT:    [[DOTFORWARD_IV___BEGIN2:%.*]] = alloca i64, align 8
+// CHECK2-NEXT:    [[DOTREVERSED_IV___BEGIN2:%.*]] = alloca i64, align 8
+// CHECK2-NEXT:    [[V:%.*]] = alloca ptr, align 8
+// CHECK2-NEXT:    store i32 42, ptr [[C]], align 4
+// CHECK2-NEXT:    store ptr [[ARR]], ptr [[__RANGE2]], align 8
+// CHECK2-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[__RANGE2]], align 8
+// CHECK2-NEXT:    [[ARRAYDECAY:%.*]] = getelementptr inbounds [128 x double], ptr [[TMP0]], i64 0, i64 0
+// CHECK2-NEXT:    [[ADD_PTR:%.*]] = getelementptr inbounds double, ptr [[ARRAYDECAY]], i64 128
+// CHECK2-NEXT:    store ptr [[ADD_PTR]], ptr [[__END2]], align 8
+// CHECK2-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[__RANGE2]], align 8
+// CHECK2-NEXT:    [[ARRAYDECAY1:%.*]] = getelementptr inbounds [128 x double], ptr [[TMP1]], i64 0, i64 0
+// CHECK2-NEXT:    store ptr [[ARRAYDECAY1]], ptr [[__BEGIN2]], align 8
+// CHECK2-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[__RANGE2]], align 8
+// CHECK2-NEXT:    [[ARRAYDECAY2:%.*]] = getelementptr inbounds [128 x double], ptr [[TMP2]], i64 0, i64 0
+// CHECK2-NEXT:    store ptr [[ARRAYDECAY2]], ptr [[DOTCAPTURE_EXPR_]], align 8
+// CHECK2-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[__END2]], align 8
+// CHECK2-NEXT:    store ptr [[TMP3]], ptr [[DOTCAPTURE_EXPR_3]], align 8
+// CHECK2-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[DOTCAPTURE_EXPR_3]], align 8
+// CHECK2-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[DOTCAPTURE_EXPR_]], align 8
+// CHECK2-NEXT:    [[SUB_PTR_LHS_CAST:%.*]] = ptrtoint ptr [[TMP4]] to i64
+// CHECK2-NEXT:    [[SUB_PTR_RHS_CAST:%.*]] = ptrtoint ptr [[TMP5]] to i64
+// CHECK2-NEXT:    [[SUB_PTR_SUB:%.*]] = sub i64 [[SUB_PTR_LHS_CAST]], [[SUB_PTR_RHS_CAST]]
+// CHECK2-NEXT:    [[SUB_PTR_DIV:%.*]] = sdiv exact i64 [[SUB_PTR_SUB]], 8
+// CHECK2-NEXT:    [[SUB:%.*]] = sub nsw i64 [[SUB_PTR_DIV]], 1
+// CHECK2-NEXT:    [[ADD:%.*]] = add nsw i64 [[SUB]], 1
+// CHECK2-NEXT:    [[DIV:%.*]] = sdiv i64 [[ADD]], 1
+// CHECK2-NEXT:    [[SUB5:%.*]] = sub nsw i64 [[DIV]], 1
+// CHECK2-NEXT:    store i64 [[SUB5]], ptr [[DOTCAPTURE_EXPR_4]], align 8
+// CHECK2-NEXT:    store i64 0, ptr [[DOTFORWARD_IV___BEGIN2]], align 8
+// CHECK2-NEXT:    br label [[FOR_COND:%.*]]
+// CHECK2:       for.cond:
+// CHECK2-NEXT:    [[TMP6:%.*]] = load i64, ptr [[DOTFORWARD_IV___BEGIN2]], align 8
+// CHECK2-NEXT:    [[TMP7:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR_4]], align 8
+// CHECK2-NEXT:    [[ADD6:%.*]] = add nsw i64 [[TMP7]], 1
+// CHECK2-NEXT:    [[CMP:%.*]] = icmp slt i64 [[TMP6]], [[ADD6]]
+// CHECK2-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
+// CHECK2:       for.body:
+// CHECK2-NEXT:    [[TMP8:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR_4]], align 8
+// CHECK2-NEXT:    [[ADD7:%.*]] = add nsw i64 [[TMP8]], 1
+// CHECK2-NEXT:    [[SUB8:%.*]] = sub nsw i64 [[ADD7]], 1
+// CHECK2-NEXT:    [[TMP9:%.*]] = load i64, ptr [[DOTFORWARD_IV___BEGIN2]], align 8
+// CHECK2-NEXT:    [[SUB9:%.*]] = sub nsw i64 [[SUB8]], [[TMP9]]
+// CHECK2-NEXT:    store i64 [[SUB9]], ptr [[DOTREVERSED_IV___BEGIN2]], align 8
+// CHECK2-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DOTCAPTURE_EXPR_]], align 8
+// CHECK2-NEXT:    [[TMP11:%.*]] = load i64, ptr [[DOTREVERSED_IV___BEGIN2]], align 8
+// CHECK2-NEXT:    [[MUL:%.*]] = mul nsw i64 [[TMP11]], 1
+// CHECK2-NEXT:    [[ADD_PTR10:%.*]] = getelementptr inbounds double, ptr [[TMP10]], i64 [[MUL]]
+// CHECK2-NEXT:    store ptr [[ADD_PTR10]], ptr [[__BEGIN2]], align 8
+// CHECK2-NEXT:    [[TMP12:%.*]] = load ptr, ptr [[__BEGIN2]], align 8
+// CHECK2-NEXT:    store ptr [[TMP12]], ptr [[V]], align 8
+// CHECK2-NEXT:    [[TMP13:%.*]] = load ptr, ptr [[V]], align 8
+// CHECK2-NEXT:    [[TMP14:%.*]] = load double, ptr [[TMP13]], align 8
+// CHECK2-NEXT:    [[TMP15:%.*]] = load i32, ptr [[C]], align 4
+// CHECK2-NEXT:    call void (...) @body(double noundef [[TMP14]], i32 noundef [[TMP15]])
+// CHECK2-NEXT:    br label [[FOR_INC:%.*]]
+// CHECK2:       for.inc:
+// CHECK2-NEXT:    [[TMP16:%.*]] = load i64, ptr [[DOTFORWARD_IV___BEGIN2]], align 8
+// CHECK2-NEXT:    [[INC:%.*]] = add nsw i64 [[TMP16]], 1
+// CHECK2-NEXT:    store i64 [[INC]], ptr [[DOTFORWARD_IV___BEGIN2]], align 8
+// CHECK2-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP8:![0-9]+]]
+// CHECK2:       for.end:
+// CHECK2-NEXT:    ret void
+//
+//
+// CHECK2-LABEL: define {{[^@]+}}@foo7
+// CHECK2-SAME: () #[[ATTR1]] {
+// CHECK2-NEXT:  entry:
+// CHECK2-NEXT:    [[A:%.*]] = alloca [128 x double], align 16
+// CHECK2-NEXT:    [[DOTOMP_IV:%.*]] = alloca i64, align 8
+// CHECK2-NEXT:    [[TMP:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[_TMP1:%.*]] = alloca i64, align 8
+// CHECK2-NEXT:    [[_TMP2:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[C:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[__RANGE3:%.*]] = alloca ptr, align 8
+// CHECK2-NEXT:    [[__END3:%.*]] = alloca ptr, align 8
+// CHECK2-NEXT:    [[__BEGIN3:%.*]] = alloca ptr, align 8
+// CHECK2-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca ptr, align 8
+// CHECK2-NEXT:    [[DOTCAPTURE_EXPR_5:%.*]] = alloca ptr, align 8
+// CHECK2-NEXT:    [[DOTCAPTURE_EXPR_6:%.*]] = alloca i64, align 8
+// CHECK2-NEXT:    [[DOTCAPTURE_EXPR_8:%.*]] = alloca i64, align 8
+// CHECK2-NEXT:    [[DOTCAPTURE_EXPR_10:%.*]] = alloca i64, align 8
+// CHECK2-NEXT:    [[K:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[DOTFORWARD_IV___BEGIN3:%.*]] = alloca i64, align 8
+// CHECK2-NEXT:    [[J:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[DOTOMP_LB:%.*]] = alloca i64, align 8
+// CHECK2-NEXT:    [[DOTOMP_UB:%.*]] = alloca i64, align 8
+// CHECK2-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i64, align 8
+// CHECK2-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[K15:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[DOTFORWARD_IV___BEGIN316:%.*]] = alloca i64, align 8
+// CHECK2-NEXT:    [[J17:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[DOTREVERSED_IV___BEGIN3:%.*]] = alloca i64, align 8
+// CHECK2-NEXT:    [[V:%.*]] = alloca ptr, align 8
+// CHECK2-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB2]])
+// CHECK2-NEXT:    store i32 42, ptr [[C]], align 4
+// CHECK2-NEXT:    store ptr [[A]], ptr [[__RANGE3]], align 8
+// CHECK2-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[__RANGE3]], align 8
+// CHECK2-NEXT:    [[ARRAYDECAY:%.*]] = getelementptr inbounds [128 x double], ptr [[TMP1]], i64 0, i64 0
+// CHECK2-NEXT:    [[ADD_PTR:%.*]] = getelementptr inbounds double, ptr [[ARRAYDECAY]], i64 128
+// CHECK2-NEXT:    store ptr [[ADD_PTR]], ptr [[__END3]], align 8
+// CHECK2-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[__RANGE3]], align 8
+// CHECK2-NEXT:    [[ARRAYDECAY3:%.*]] = getelementptr inbounds [128 x double], ptr [[TMP2]], i64 0, i64 0
+// CHECK2-NEXT:    store ptr [[ARRAYDECAY3]], ptr [[__BEGIN3]], align 8
+// CHECK2-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[__RANGE3]], align 8
+// CHECK2-NEXT:    [[ARRAYDECAY4:%.*]] = getelementptr inbounds [128 x double], ptr [[TMP3]], i64 0, i64 0
+// CHECK2-NEXT:    store ptr [[ARRAYDECAY4]], ptr [[DOTCAPTURE_EXPR_]], align 8
+// CHECK2-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[__END3]], align 8
+// CHECK2-NEXT:    store ptr [[TMP4]], ptr [[DOTCAPTURE_EXPR_5]], align 8
+// CHECK2-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[DOTCAPTURE_EXPR_5]], align 8
+// CHECK2-NEXT:    [[TMP6:%.*]] = load ptr, ptr [[DOTCAPTURE_EXPR_]], align 8
+// CHECK2-NEXT:    [[SUB_PTR_LHS_CAST:%.*]] = ptrtoint ptr [[TMP5]] to i64
+// CHECK2-NEXT:    [[SUB_PTR_RHS_CAST:%.*]] = ptrtoint ptr [[TMP6]] to i64
+// CHECK2-NEXT:    [[SUB_PTR_SUB:%.*]] = sub i64 [[SUB_PTR_LHS_CAST]], [[SUB_PTR_RHS_CAST]]
+// CHECK2-NEXT:    [[SUB_PTR_DIV:%.*]] = sdiv exact i64 [[SUB_PTR_SUB]], 8
+// CHECK2-NEXT:    [[SUB:%.*]] = sub nsw i64 [[SUB_PTR_DIV]], 1
+// CHECK2-NEXT:    [[ADD:%.*]] = add nsw i64 [[SUB]], 1
+// CHECK2-NEXT:    [[DIV:%.*]] = sdiv i64 [[ADD]], 1
+// CHECK2-NEXT:    [[SUB7:%.*]] = sub nsw i64 [[DIV]], 1
+// CHECK2-NEXT:    store i64 [[SUB7]], ptr [[DOTCAPTURE_EXPR_6]], align 8
+// CHECK2-NEXT:    [[TMP7:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR_6]], align 8
+// CHECK2-NEXT:    [[ADD9:%.*]] = add nsw i64 [[TMP7]], 1
+// CHECK2-NEXT:    store i64 [[ADD9]], ptr [[DOTCAPTURE_EXPR_8]], align 8
+// CHECK2-NEXT:    [[TMP8:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR_8]], align 8
+// CHECK2-NEXT:    [[SUB11:%.*]] = sub nsw i64 [[TMP8]], 0
+// CHECK2-NEXT:    [[DIV12:%.*]] = sdiv i64 [[SUB11]], 1
+// CHECK2-NEXT:    [[MUL:%.*]] = mul nsw i64 4, [[DIV12]]
+// CHECK2-NEXT:    [[MUL13:%.*]] = mul nsw i64 [[MUL]], 4
+// CHECK2-NEXT:    [[SUB14:%.*]] = sub nsw i64 [[MUL13]], 1
+// CHECK2-NEXT:    store i64 [[SUB14]], ptr [[DOTCAPTURE_EXPR_10]], align 8
+// CHECK2-NEXT:    store i32 7, ptr [[K]], align 4
+// CHECK2-NEXT:    store i64 0, ptr [[DOTFORWARD_IV___BEGIN3]], align 8
+// CHECK2-NEXT:    store i32 7, ptr [[J]], align 4
+// CHECK2-NEXT:    [[TMP9:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR_8]], align 8
+// CHECK2-NEXT:    [[CMP:%.*]] = icmp slt i64 0, [[TMP9]]
+// CHECK2-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
+// CHECK2:       omp.precond.then:
+// CHECK2-NEXT:    store i64 0, ptr [[DOTOMP_LB]], align 8
+// CHECK2-NEXT:    [[TMP10:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR_10]], align 8
+// CHECK2-NEXT:    store i64 [[TMP10]], ptr [[DOTOMP_UB]], align 8
+// CHECK2-NEXT:    store i64 1, ptr [[DOTOMP_STRIDE]], align 8
+// CHECK2-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
+// CHECK2-NEXT:    call void @__kmpc_for_static_init_8(ptr @[[GLOB1]], i32 [[TMP0]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i64 1, i64 1)
+// CHECK2-NEXT:    [[TMP11:%.*]] = load i64, ptr [[DOTOMP_UB]], align 8
+// CHECK2-NEXT:    [[TMP12:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR_10]], align 8
+// CHECK2-NEXT:    [[CMP18:%.*]] = icmp sgt i64 [[TMP11]], [[TMP12]]
+// CHECK2-NEXT:    br i1 [[CMP18]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
+// CHECK2:       cond.true:
+// CHECK2-NEXT:    [[TMP13:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR_10]], align 8
+// CHECK2-NEXT:    br label [[COND_END:%.*]]
+// CHECK2:       cond.false:
+// CHECK2-NEXT:    [[TMP14:%.*]] = load i64, ptr [[DOTOMP_UB]], align 8
+// CHECK2-NEXT:    br label [[COND_END]]
+// CHECK2:       cond.end:
+// CHECK2-NEXT:    [[COND:%.*]] = phi i64 [ [[TMP13]], [[COND_TRUE]] ], [ [[TMP14]], [[COND_FALSE]] ]
+// CHECK2-NEXT:    store i64 [[COND]], ptr [[DOTOMP_UB]], align 8
+// CHECK2-NEXT:    [[TMP15:%.*]] = load i64, ptr [[DOTOMP_LB]], align 8
+// CHECK2-NEXT:    store i64 [[TMP15]], ptr [[DOTOMP_IV]], align 8
+// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
+// CHECK2:       omp.inner.for.cond:
+// CHECK2-NEXT:    [[TMP16:%.*]] = load i64, ptr [[DOTOMP_IV]], align 8
+// CHECK2-NEXT:    [[TMP17:%.*]] = load i64, ptr [[DOTOMP_UB]], align 8
+// CHECK2-NEXT:    [[CMP19:%.*]] = icmp sle i64 [[TMP16]], [[TMP17]]
+// CHECK2-NEXT:    br i1 [[CMP19]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
+// CHECK2:       omp.inner.for.body:
+// CHECK2-NEXT:    [[TMP18:%.*]] = load i64, ptr [[DOTOMP_IV]], align 8
+// CHECK2-NEXT:    [[TMP19:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR_8]], align 8
+// CHECK2-NEXT:    [[SUB20:%.*]] = sub nsw i64 [[TMP19]], 0
+// CHECK2-NEXT:    [[DIV21:%.*]] = sdiv i64 [[SUB20]], 1
+// CHECK2-NEXT:    [[MUL22:%.*]] = mul nsw i64 1, [[DIV21]]
+// CHECK2-NEXT:    [[MUL23:%.*]] = mul nsw i64 [[MUL22]], 4
+// CHECK2-NEXT:    [[DIV24:%.*]] = sdiv i64 [[TMP18]], [[MUL23]]
+// CHECK2-NEXT:    [[MUL25:%.*]] = mul nsw i64 [[DIV24]], 3
+// CHECK2-NEXT:    [[ADD26:%.*]] = add nsw i64 7, [[MUL25]]
+// CHECK2-NEXT:    [[CONV:%.*]] = trunc i64 [[ADD26]] to i32
+// CHECK2-NEXT:    store i32 [[CONV]], ptr [[K15]], align 4
+// CHECK2-NEXT:    [[TMP20:%.*]] = load i64, ptr [[DOTOMP_IV]], align 8
+// CHECK2-NEXT:    [[TMP21:%.*]] = load i64, ptr [[DOTOMP_IV]], align 8
+// CHECK2-NEXT:    [[TMP22:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR_8]], align 8
+// CHECK2-NEXT:    [[SUB27:%.*]] = sub nsw i64 [[TMP22]], 0
+// CHECK2-NEXT:    [[DIV28:%.*]] = sdiv i64 [[SUB27]], 1
+// CHECK2-NEXT:    [[MUL29:%.*]] = mul nsw i64 1, [[DIV28]]
+// CHECK2-NEXT:    [[MUL30:%.*]] = mul nsw i64 [[MUL29]], 4
+// CHECK2-NEXT:    [[DIV31:%.*]] = sdiv i64 [[TMP21]], [[MUL30]]
+// CHECK2-NEXT:    [[TMP23:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR_8]], align 8
+// CHECK2-NEXT:    [[SUB32:%.*]] = sub nsw i64 [[TMP23]], 0
+// CHECK2-NEXT:    [[DIV33:%.*]] = sdiv i64 [[SUB32]], 1
+// CHECK2-NEXT:    [[MUL34:%.*]] = mul nsw i64 1, [[DIV33]]
+// CHECK2-NEXT:    [[MUL35:%.*]] = mul nsw i64 [[MUL34]], 4
+// CHECK2-NEXT:    [[MUL36:%.*]] = mul nsw i64 [[DIV31]], [[MUL35]]
+// CHECK2-NEXT:    [[SUB37:%.*]] = sub nsw i64 [[TMP20]], [[MUL36]]
+// CHECK2-NEXT:    [[DIV38:%.*]] = sdiv i64 [[SUB37]], 4
+// CHECK2-NEXT:    [[MUL39:%.*]] = mul nsw i64 [[DIV38]], 1
+// CHECK2-NEXT:    [[ADD40:%.*]] = add nsw i64 0, [[MUL39]]
+// CHECK2-NEXT:    store i64 [[ADD40]], ptr [[DOTFORWARD_IV___BEGIN316]], align 8
+// CHECK2-NEXT:    [[TMP24:%.*]] = load i64, ptr [[DOTOMP_IV]], align 8
+// CHECK2-NEXT:    [[TMP25:%.*]] = load i64, ptr [[DOTOMP_IV]], align 8
+// CHECK2-NEXT:    [[TMP26:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR_8]], align 8
+// CHECK2-NEXT:    [[SUB41:%.*]] = sub nsw i64 [[TMP26]], 0
+// CHECK2-NEXT:    [[DIV42:%.*]] = sdiv i64 [[SUB41]], 1
+// CHECK2-NEXT:    [[MUL43:%.*]] = mul nsw i64 1, [[DIV42]]
+// CHECK2-NEXT:    [[MUL44:%.*]] = mul nsw i64 [[MUL43]], 4
+// CHECK2-NEXT:    [[DIV45:%.*]] = sdiv i64 [[TMP25]], [[MUL44]]
+// CHECK2-NEXT:    [[TMP27:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR_8]], align 8
+// CHECK2-NEXT:    [[SUB46:%.*]] = sub nsw i64 [[TMP27]], 0
+// CHECK2-NEXT:    [[DIV47:%.*]] = sdiv i64 [[SUB46]], 1
+// CHECK2-NEXT:    [[MUL48:%.*]] = mul nsw i64 1, [[DIV47]]
+// CHECK2-NEXT:    [[MUL49:%.*]] = mul nsw i64 [[MUL48]], 4
+// CHECK2-NEXT:    [[MUL50:%.*]] = mul nsw i64 [[DIV45]], [[MUL49]]
+// CHECK2-NEXT:    [[SUB51:%.*]] = sub nsw i64 [[TMP24]], [[MUL50]]
+// CHECK2-NEXT:    [[TMP28:%.*]] = load i64, ptr [[DOTOMP_IV]], align 8
+// CHECK2-NEXT:    [[TMP29:%.*]] = load i64, ptr [[DOTOMP_IV]], align 8
+// CHECK2-NEXT:    [[TMP30:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR_8]], align 8
+// CHECK2-NEXT:    [[SUB52:%.*]] = sub nsw i64 [[TMP30]], 0
+// CHECK2-NEXT:    [[DIV53:%.*]] = sdiv i64 [[SUB52]], 1
+// CHECK2-NEXT:    [[MUL54:%.*]] = mul nsw i64 1, [[DIV53]]
+// CHECK2-NEXT:    [[MUL55:%.*]] = mul nsw i64 [[MUL54]], 4
+// CHECK2-NEXT:    [[DIV56:%.*]] = sdiv i64 [[TMP29]], [[MUL55]]
+// CHECK2-NEXT:    [[TMP31:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR_8]], align 8
+// CHECK2-NEXT:    [[SUB57:%.*]] = sub nsw i64 [[TMP31]], 0
+// CHECK2-NEXT:    [[DIV58:%.*]] = sdiv i64 [[SUB57]], 1
+// CHECK2-NEXT:    [[MUL59:%.*]] = mul nsw i64 1, [[DIV58]]
+// CHECK2-NEXT:    [[MUL60:%.*]] = mul nsw i64 [[MUL59]], 4
+// CHECK2-NEXT:    [[MUL61:%.*]] = mul nsw i64 [[DIV56]], [[MUL60]]
+// CHECK2-NEXT:    [[SUB62:%.*]] = sub nsw i64 [[TMP28]], [[MUL61]]
+// CHECK2-NEXT:    [[DIV63:%.*]] = sdiv i64 [[SUB62]], 4
+// CHECK2-NEXT:    [[MUL64:%.*]] = mul nsw i64 [[DIV63]], 4
+// CHECK2-NEXT:    [[SUB65:%.*]] = sub nsw i64 [[SUB51]], [[MUL64]]
+// CHECK2-NEXT:    [[MUL66:%.*]] = mul nsw i64 [[SUB65]], 3
+// CHECK2-NEXT:    [[ADD67:%.*]] = add nsw i64 7, [[MUL66]]
+// CHECK2-NEXT:    [[CONV68:%.*]] = trunc i64 [[ADD67]] to i32
+// CHECK2-NEXT:    store i32 [[CONV68]], ptr [[J17]], align 4
+// CHECK2-NEXT:    [[TMP32:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR_6]], align 8
+// CHECK2-NEXT:    [[ADD69:%.*]] = add nsw i64 [[TMP32]], 1
+// CHECK2-NEXT:    [[SUB70:%.*]] = sub nsw i64 [[ADD69]], 1
+// CHECK2-NEXT:    [[TMP33:%.*]] = load i64, ptr [[DOTFORWARD_IV___BEGIN316]], align 8
+// CHECK2-NEXT:    [[SUB71:%.*]] = sub nsw i64 [[SUB70]], [[TMP33]]
+// CHECK2-NEXT:    store i64 [[SUB71]], ptr [[DOTREVERSED_IV___BEGIN3]], align 8
+// CHECK2-NEXT:    [[TMP34:%.*]] = load ptr, ptr [[DOTCAPTURE_EXPR_]], align 8
+// CHECK2-NEXT:    [[TMP35:%.*]] = load i64, ptr [[DOTREVERSED_IV___BEGIN3]], align 8
+// CHECK2-NEXT:    [[MUL72:%.*]] = mul nsw i64 [[TMP35]], 1
+// CHECK2-NEXT:    [[ADD_PTR73:%.*]] = getelementptr inbounds double, ptr [[TMP34]], i64 [[MUL72]]
+// CHECK2-NEXT:    store ptr [[ADD_PTR73]], ptr [[__BEGIN3]], align 8
+// CHECK2-NEXT:    [[TMP36:%.*]] = load ptr, ptr [[__BEGIN3]], align 8
+// CHECK2-NEXT:    store ptr [[TMP36]], ptr [[V]], align 8
+// CHECK2-NEXT:    [[TMP37:%.*]] = load i32, ptr [[K15]], align 4
+// CHECK2-NEXT:    [[TMP38:%.*]] = load i32, ptr [[C]], align 4
+// CHECK2-NEXT:    [[TMP39:%.*]] = load ptr, ptr [[V]], align 8
+// CHECK2-NEXT:    [[TMP40:%.*]] = load double, ptr [[TMP39]], align 8
+// CHECK2-NEXT:    [[TMP41:%.*]] = load i32, ptr [[J17]], align 4
+// CHECK2-NEXT:    call void (...) @body(i32 noundef [[TMP37]], i32 noundef [[TMP38]], double noundef [[TMP40]], i32 noundef [[TMP41]])
+// CHECK2-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
+// CHECK2:       omp.body.continue:
+// CHECK2-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
+// CHECK2:       omp.inner.for.inc:
+// CHECK2-NEXT:    [[TMP42:%.*]] = load i64, ptr [[DOTOMP_IV]], align 8
+// CHECK2-NEXT:    [[ADD74:%.*]] = add nsw i64 [[TMP42]], 1
+// CHECK2-NEXT:    store i64 [[ADD74]], ptr [[DOTOMP_IV]], align 8
+// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND]]
+// CHECK2:       omp.inner.for.end:
+// CHECK2-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
+// CHECK2:       omp.loop.exit:
+// CHECK2-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP0]])
+// CHECK2-NEXT:    br label [[OMP_PRECOND_END]]
+// CHECK2:       omp.precond.end:
+// CHECK2-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB3]], i32 [[TMP0]])
+// CHECK2-NEXT:    ret void
+//
+//
+// CHECK2-LABEL: define {{[^@]+}}@tfoo5
+// CHECK2-SAME: () #[[ATTR1]] {
+// CHECK2-NEXT:  entry:
+// CHECK2-NEXT:    call void @_Z4foo5IiTnT_Li3EEvS0_S0_(i32 noundef 0, i32 noundef 42)
+// CHECK2-NEXT:    ret void
+//
+//
+// CHECK2-LABEL: define {{[^@]+}}@_Z4foo5IiTnT_Li3EEvS0_S0_
+// CHECK2-SAME: (i32 noundef [[START:%.*]], i32 noundef [[END:%.*]]) #[[ATTR1]] comdat {
+// CHECK2-NEXT:  entry:
+// CHECK2-NEXT:    [[START_ADDR:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[END_ADDR:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[I:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[DOTFORWARD_IV_I:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    [[DOTREVERSED_IV_I:%.*]] = alloca i32, align 4
+// CHECK2-NEXT:    store i32 [[START]], ptr [[START_ADDR]], align 4
+// CHECK2-NEXT:    store i32 [[END]], ptr [[END_ADDR]], align 4
+// CHECK2-NEXT:    [[TMP0:%.*]] = load i32, ptr [[START_ADDR]], align 4
+// CHECK2-NEXT:    store i32 [[TMP0]], ptr [[I]], align 4
+// CHECK2-NEXT:    [[TMP1:%.*]] = load i32, ptr [[START_ADDR]], align 4
+// CHECK2-NEXT:    store i32 [[TMP1]], ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK2-NEXT:    [[TMP2:%.*]] = load i32, ptr [[END_ADDR]], align 4
+// CHECK2-NEXT:    store i32 [[TMP2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
+// CHECK2-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
+// CHECK2-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK2-NEXT:    [[SUB:%.*]] = sub i32 [[TMP3]], [[TMP4]]
+// CHECK2-NEXT:    [[SUB3:%.*]] = sub i32 [[SUB]], 1
+// CHECK2-NEXT:    [[ADD:%.*]] = add i32 [[SUB3]], 3
+// CHECK2-NEXT:    [[DIV:%.*]] = udiv i32 [[ADD]], 3
+// CHECK2-NEXT:    [[SUB4:%.*]] = sub i32 [[DIV]], 1
+// CHECK2-NEXT:    store i32 [[SUB4]], ptr [[DOTCAPTURE_EXPR_2]], align 4
+// CHECK2-NEXT:    store i32 0, ptr [[DOTFORWARD_IV_I]], align 4
+// CHECK2-NEXT:    br label [[FOR_COND:%.*]]
+// CHECK2:       for.cond:
+// CHECK2-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTFORWARD_IV_I]], align 4
+// CHECK2-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
+// CHECK2-NEXT:    [[ADD5:%.*]] = add i32 [[TMP6]], 1
+// CHECK2-NEXT:    [[CMP:%.*]] = icmp ult i32 [[TMP5]], [[ADD5]]
+// CHECK2-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
+// CHECK2:       for.body:
+// CHECK2-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
+// CHECK2-NEXT:    [[ADD6:%.*]] = add i32 [[TMP7]], 1
+// CHECK2-NEXT:    [[SUB7:%.*]] = sub i32 [[ADD6]], 1
+// CHECK2-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTFORWARD_IV_I]], align 4
+// CHECK2-NEXT:    [[SUB8:%.*]] = sub i32 [[SUB7]], [[TMP8]]
+// CHECK2-NEXT:    store i32 [[SUB8]], ptr [[DOTREVERSED_IV_I]], align 4
+// CHECK2-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK2-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTREVERSED_IV_I]], align 4
+// CHECK2-NEXT:    [[MUL:%.*]] = mul i32 [[TMP10]], 3
+// CHECK2-NEXT:    [[ADD9:%.*]] = add i32 [[TMP9]], [[MUL]]
+// CHECK2-NEXT:    store i32 [[ADD9]], ptr [[I]], align 4
+// CHECK2-NEXT:    [[TMP11:%.*]] = load i32, ptr [[I]], align 4
+// CHECK2-NEXT:    call void (...) @body(i32 noundef [[TMP11]])
+// CHECK2-NEXT:    br label [[FOR_INC:%.*]]
+// CHECK2:       for.inc:
+// CHECK2-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTFORWARD_IV_I]], align 4
+// CHECK2-NEXT:    [[INC:%.*]] = add i32 [[TMP12]], 1
+// CHECK2-NEXT:    store i32 [[INC]], ptr [[DOTFORWARD_IV_I]], align 4
+// CHECK2-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP9:![0-9]+]]
+// CHECK2:       for.end:
+// CHECK2-NEXT:    ret void
+//
+//
+// CHECK2-LABEL: define {{[^@]+}}@_GLOBAL__sub_I_reverse_codegen.cpp
+// CHECK2-SAME: () #[[ATTR0]] section ".text.startup" {
+// CHECK2-NEXT:  entry:
+// CHECK2-NEXT:    call void @__cxx_global_var_init()
+// CHECK2-NEXT:    ret void
+
diff --git a/clang/test/OpenMP/reverse_messages.cpp b/clang/test/OpenMP/reverse_messages.cpp
new file mode 100644
index 0000000000000..9636a70bf2753
--- /dev/null
+++ b/clang/test/OpenMP/reverse_messages.cpp
@@ -0,0 +1,40 @@
+// RUN: %clang_cc1 -triple x86_64-pc-linux-gnu -std=c++20 -fopenmp -fopenmp-version=60 -fsyntax-only -Wuninitialized -verify %s
+
+void func() {
+
+  // expected-error at +2 {{statement after '#pragma omp reverse' must be a for loop}}
+  #pragma omp reverse
+    ;
+
+  // expected-error at +2 {{statement after '#pragma omp reverse' must be a for loop}}
+  #pragma omp reverse
+  int b = 0;
+
+  // expected-error at +2 {{statement after '#pragma omp reverse' must be a for loop}}
+  #pragma omp reverse
+  #pragma omp for
+  for (int i = 0; i < 7; ++i)
+    ;
+
+  {
+    // expected-error at +2 {{expected statement}}
+    #pragma omp reverse
+  }
+
+  // expected-error at +2 {{condition of OpenMP for loop must be a relational comparison ('<', '<=', '>', '>=', or '!=') of loop variable 'i'}}
+  #pragma omp reverse
+  for (int i = 0; i/3<7; ++i)
+    ;
+
+  // expected-error at +1 {{unexpected OpenMP clause 'sizes' in directive '#pragma omp reverse'}}
+  #pragma omp reverse sizes(5)
+  for (int i = 0; i < 7; ++i)
+    ;
+
+  // expected-warning at +1 {{extra tokens at the end of '#pragma omp reverse' are ignored}}
+  #pragma omp reverse foo
+  for (int i = 0; i < 7; ++i)
+    ;
+
+}
+
diff --git a/clang/tools/libclang/CIndex.cpp b/clang/tools/libclang/CIndex.cpp
index bfbdb5be9ff2f..82dcee56244bf 100644
--- a/clang/tools/libclang/CIndex.cpp
+++ b/clang/tools/libclang/CIndex.cpp
@@ -2179,6 +2179,7 @@ class EnqueueVisitor : public ConstStmtVisitor<EnqueueVisitor, void>,
   VisitOMPLoopTransformationDirective(const OMPLoopTransformationDirective *D);
   void VisitOMPTileDirective(const OMPTileDirective *D);
   void VisitOMPUnrollDirective(const OMPUnrollDirective *D);
+  void VisitOMPReverseDirective(const OMPReverseDirective *D);
   void VisitOMPForDirective(const OMPForDirective *D);
   void VisitOMPForSimdDirective(const OMPForSimdDirective *D);
   void VisitOMPSectionsDirective(const OMPSectionsDirective *D);
@@ -3217,6 +3218,10 @@ void EnqueueVisitor::VisitOMPUnrollDirective(const OMPUnrollDirective *D) {
   VisitOMPLoopTransformationDirective(D);
 }
 
+void EnqueueVisitor::VisitOMPReverseDirective(const OMPReverseDirective *D) {
+  VisitOMPLoopTransformationDirective(D);
+}
+
 void EnqueueVisitor::VisitOMPForDirective(const OMPForDirective *D) {
   VisitOMPLoopDirective(D);
 }
@@ -6075,6 +6080,8 @@ CXString clang_getCursorKindSpelling(enum CXCursorKind Kind) {
     return cxstring::createRef("OMPTileDirective");
   case CXCursor_OMPUnrollDirective:
     return cxstring::createRef("OMPUnrollDirective");
+  case CXCursor_OMPReverseDirective:
+    return cxstring::createRef("OMPReverseDirective");
   case CXCursor_OMPForDirective:
     return cxstring::createRef("OMPForDirective");
   case CXCursor_OMPForSimdDirective:
diff --git a/clang/tools/libclang/CXCursor.cpp b/clang/tools/libclang/CXCursor.cpp
index 9325a16d2a848..275714d8fbcdd 100644
--- a/clang/tools/libclang/CXCursor.cpp
+++ b/clang/tools/libclang/CXCursor.cpp
@@ -672,6 +672,9 @@ CXCursor cxcursor::MakeCXCursor(const Stmt *S, const Decl *Parent,
   case Stmt::OMPUnrollDirectiveClass:
     K = CXCursor_OMPUnrollDirective;
     break;
+  case Stmt::OMPReverseDirectiveClass:
+    K = CXCursor_OMPReverseDirective;
+    break;
   case Stmt::OMPForDirectiveClass:
     K = CXCursor_OMPForDirective;
     break;
diff --git a/llvm/include/llvm/Frontend/OpenMP/OMP.td b/llvm/include/llvm/Frontend/OpenMP/OMP.td
index e91169e8da1aa..0ad3d919103fe 100644
--- a/llvm/include/llvm/Frontend/OpenMP/OMP.td
+++ b/llvm/include/llvm/Frontend/OpenMP/OMP.td
@@ -537,6 +537,9 @@ def OMP_Unroll : Directive<"unroll"> {
   ];
   let association = AS_Loop;
 }
+def OMP_Reverse : Directive<"reverse"> {
+  let association = AS_Loop;
+}
 def OMP_For : Directive<"for"> {
   let allowedClauses = [
     VersionedClause<OMPC_Private>,
diff --git a/openmp/runtime/test/transform/reverse/foreach.cpp b/openmp/runtime/test/transform/reverse/foreach.cpp
new file mode 100644
index 0000000000000..0784e3c0057c9
--- /dev/null
+++ b/openmp/runtime/test/transform/reverse/foreach.cpp
@@ -0,0 +1,162 @@
+// RUN: %libomp-cxx20-compile-and-run | FileCheck %s --match-full-lines
+
+#ifndef HEADER
+#define HEADER
+
+#include <cstdlib>
+#include <cstdarg>
+#include <cstdio>
+#include <vector>
+
+struct Reporter {
+  const char *name;
+
+  Reporter(const char *name) : name(name) { print("ctor"); }
+
+  Reporter() : name("<anon>") { print("ctor"); }
+
+  Reporter(const Reporter &that) : name(that.name) { print("copy ctor"); }
+
+  Reporter(Reporter &&that) : name(that.name) { print("move ctor"); }
+
+  ~Reporter() { print("dtor"); }
+
+  const Reporter &operator=(const Reporter &that) {
+    print("copy assign");
+    this->name = that.name;
+    return *this;
+  }
+
+  const Reporter &operator=(Reporter &&that) {
+    print("move assign");
+    this->name = that.name;
+    return *this;
+  }
+
+  struct Iterator {
+    const Reporter *owner;
+    int pos;
+
+    Iterator(const Reporter *owner, int pos) : owner(owner), pos(pos) {}
+
+    Iterator(const Iterator &that) : owner(that.owner), pos(that.pos) {
+      owner->print("iterator copy ctor");
+    }
+
+    Iterator(Iterator &&that) : owner(that.owner), pos(that.pos) {
+      owner->print("iterator move ctor");
+    }
+
+    ~Iterator() { owner->print("iterator dtor"); }
+
+    const Iterator &operator=(const Iterator &that) {
+      owner->print("iterator copy assign");
+      this->owner = that.owner;
+      this->pos = that.pos;
+      return *this;
+    }
+
+    const Iterator &operator=(Iterator &&that) {
+      owner->print("iterator move assign");
+      this->owner = that.owner;
+      this->pos = that.pos;
+      return *this;
+    }
+
+    bool operator==(const Iterator &that) const {
+      owner->print("iterator %d == %d", 2 - this->pos, 2 - that.pos);
+      return this->pos == that.pos;
+    }
+
+    Iterator &operator++() {
+      owner->print("iterator prefix ++");
+      pos -= 1;
+      return *this;
+    }
+
+    Iterator operator++(int) {
+      owner->print("iterator postfix ++");
+      auto result = *this;
+      pos -= 1;
+      return result;
+    }
+
+    int operator*() const {
+      int result = 2 - pos;
+      owner->print("iterator deref: %i", result);
+      return result;
+    }
+
+    size_t operator-(const Iterator &that) const {
+      int result = (2 - this->pos) - (2 - that.pos);
+      owner->print("iterator distance: %d", result);
+      return result;
+    }
+
+    Iterator operator+(int steps) const {
+      owner->print("iterator advance: %i += %i", 2 - this->pos, steps);
+      return Iterator(owner, pos - steps);
+    }
+
+    void print(const char *msg) const { owner->print(msg); }
+  };
+
+  Iterator begin() const {
+    print("begin()");
+    return Iterator(this, 2);
+  }
+
+  Iterator end() const {
+    print("end()");
+    return Iterator(this, -1);
+  }
+
+  void print(const char *msg, ...) const {
+    va_list args;
+    va_start(args, msg);
+    printf("[%s] ", name);
+    vprintf(msg, args);
+    printf("\n");
+    va_end(args);
+  }
+};
+
+int main() {
+  printf("do\n");
+#pragma omp reverse
+  for (Reporter c{"init-stmt"}; auto &&v : Reporter("range"))
+    printf("v=%d\n", v);
+  printf("done\n");
+  return EXIT_SUCCESS;
+}
+
+#endif /* HEADER */
+
+// CHECK:      do
+// CHECK-NEXT: [init-stmt] ctor
+// CHECK-NEXT: [range] ctor
+// CHECK-NEXT: [range] end()
+// CHECK-NEXT: [range] begin()
+// CHECK-NEXT: [range] begin()
+// CHECK-NEXT: [range] iterator distance: 3
+// CHECK-NEXT: [range] iterator advance: 0 += 2
+// CHECK-NEXT: [range] iterator move assign
+// CHECK-NEXT: [range] iterator deref: 2
+// CHECK-NEXT: v=2
+// CHECK-NEXT: [range] iterator dtor
+// CHECK-NEXT: [range] iterator advance: 0 += 1
+// CHECK-NEXT: [range] iterator move assign
+// CHECK-NEXT: [range] iterator deref: 1
+// CHECK-NEXT: v=1
+// CHECK-NEXT: [range] iterator dtor
+// CHECK-NEXT: [range] iterator advance: 0 += 0
+// CHECK-NEXT: [range] iterator move assign
+// CHECK-NEXT: [range] iterator deref: 0
+// CHECK-NEXT: v=0
+// CHECK-NEXT: [range] iterator dtor
+// CHECK-NEXT: [range] iterator dtor
+// CHECK-NEXT: [range] iterator dtor
+// CHECK-NEXT: [range] iterator dtor
+// CHECK-NEXT: [range] dtor
+// CHECK-NEXT: [init-stmt] dtor
+// CHECK-NEXT: done
diff --git a/openmp/runtime/test/transform/reverse/intfor.c b/openmp/runtime/test/transform/reverse/intfor.c
new file mode 100644
index 0000000000000..a526a8d493b3d
--- /dev/null
+++ b/openmp/runtime/test/transform/reverse/intfor.c
@@ -0,0 +1,25 @@
+// RUN: %libomp-compile-and-run | FileCheck %s --match-full-lines
+
+#ifndef HEADER
+#define HEADER
+
+#include <stdlib.h>
+#include <stdio.h>
+
+int main() {
+  printf("do\n");
+#pragma omp reverse
+  for (int i = 7; i < 19; i += 3)
+    printf("i=%d\n", i);
+  printf("done\n");
+  return EXIT_SUCCESS;
+}
+
+#endif /* HEADER */
+
+// CHECK:      do
+// CHECK-NEXT: i=16
+// CHECK-NEXT: i=13
+// CHECK-NEXT: i=10
+// CHECK-NEXT: i=7
+// CHECK-NEXT: done
diff --git a/openmp/runtime/test/transform/reverse/iterfor.cpp b/openmp/runtime/test/transform/reverse/iterfor.cpp
new file mode 100644
index 0000000000000..ba1086dbd76a5
--- /dev/null
+++ b/openmp/runtime/test/transform/reverse/iterfor.cpp
@@ -0,0 +1,164 @@
+// RUN: %libomp-cxx20-compile-and-run | FileCheck %s --match-full-lines
+
+#ifndef HEADER
+#define HEADER
+
+#include <cstdlib>
+#include <cstdarg>
+#include <cstdio>
+#include <vector>
+
+struct Reporter {
+  const char *name;
+
+  Reporter(const char *name) : name(name) { print("ctor"); }
+
+  Reporter() : name("<anon>") { print("ctor"); }
+
+  Reporter(const Reporter &that) : name(that.name) { print("copy ctor"); }
+
+  Reporter(Reporter &&that) : name(that.name) { print("move ctor"); }
+
+  ~Reporter() { print("dtor"); }
+
+  const Reporter &operator=(const Reporter &that) {
+    print("copy assign");
+    this->name = that.name;
+    return *this;
+  }
+
+  const Reporter &operator=(Reporter &&that) {
+    print("move assign");
+    this->name = that.name;
+    return *this;
+  }
+
+  struct Iterator {
+    const Reporter *owner;
+    int pos;
+
+    Iterator(const Reporter *owner, int pos) : owner(owner), pos(pos) {}
+
+    Iterator(const Iterator &that) : owner(that.owner), pos(that.pos) {
+      owner->print("iterator copy ctor");
+    }
+
+    Iterator(Iterator &&that) : owner(that.owner), pos(that.pos) {
+      owner->print("iterator move ctor");
+    }
+
+    ~Iterator() { owner->print("iterator dtor"); }
+
+    const Iterator &operator=(const Iterator &that) {
+      owner->print("iterator copy assign");
+      this->owner = that.owner;
+      this->pos = that.pos;
+      return *this;
+    }
+
+    const Iterator &operator=(Iterator &&that) {
+      owner->print("iterator move assign");
+      this->owner = that.owner;
+      this->pos = that.pos;
+      return *this;
+    }
+
+    bool operator==(const Iterator &that) const {
+      owner->print("iterator %d == %d", 2 - this->pos, 2 - that.pos);
+      return this->pos == that.pos;
+    }
+
+    bool operator!=(const Iterator &that) const {
+      owner->print("iterator %d != %d", 2 - this->pos, 2 - that.pos);
+      return this->pos != that.pos;
+    }
+
+    Iterator &operator++() {
+      owner->print("iterator prefix ++");
+      pos -= 1;
+      return *this;
+    }
+
+    Iterator operator++(int) {
+      owner->print("iterator postfix ++");
+      auto result = *this;
+      pos -= 1;
+      return result;
+    }
+
+    int operator*() const {
+      int result = 2 - pos;
+      owner->print("iterator deref: %i", result);
+      return result;
+    }
+
+    size_t operator-(const Iterator &that) const {
+      int result = (2 - this->pos) - (2 - that.pos);
+      owner->print("iterator distance: %d", result);
+      return result;
+    }
+
+    Iterator operator+(int steps) const {
+      owner->print("iterator advance: %i += %i", 2 - this->pos, steps);
+      return Iterator(owner, pos - steps);
+    }
+  };
+
+  Iterator begin() const {
+    print("begin()");
+    return Iterator(this, 2);
+  }
+
+  Iterator end() const {
+    print("end()");
+    return Iterator(this, -1);
+  }
+
+  void print(const char *msg, ...) const {
+    va_list args;
+    va_start(args, msg);
+    printf("[%s] ", name);
+    vprintf(msg, args);
+    printf("\n");
+    va_end(args);
+  }
+};
+
+int main() {
+  printf("do\n");
+  Reporter range("range");
+#pragma omp reverse
+  for (auto it = range.begin(); it != range.end(); ++it)
+    printf("v=%d\n", *it);
+  printf("done\n");
+  return EXIT_SUCCESS;
+}
+
+#endif /* HEADER */
+
+// CHECK:      do
+// CHECK-NEXT: [range] ctor
+// CHECK-NEXT: [range] begin()
+// CHECK-NEXT: [range] begin()
+// CHECK-NEXT: [range] end()
+// CHECK-NEXT: [range] iterator distance: 3
+// CHECK-NEXT: [range] iterator advance: 0 += 2
+// CHECK-NEXT: [range] iterator move assign
+// CHECK-NEXT: [range] iterator deref: 2
+// CHECK-NEXT: v=2
+// CHECK-NEXT: [range] iterator dtor
+// CHECK-NEXT: [range] iterator advance: 0 += 1
+// CHECK-NEXT: [range] iterator move assign
+// CHECK-NEXT: [range] iterator deref: 1
+// CHECK-NEXT: v=1
+// CHECK-NEXT: [range] iterator dtor
+// CHECK-NEXT: [range] iterator advance: 0 += 0
+// CHECK-NEXT: [range] iterator move assign
+// CHECK-NEXT: [range] iterator deref: 0
+// CHECK-NEXT: v=0
+// CHECK-NEXT: [range] iterator dtor
+// CHECK-NEXT: [range] iterator dtor
+// CHECK-NEXT: [range] iterator dtor
+// CHECK-NEXT: done
+// CHECK-NEXT: [range] iterator dtor
+// CHECK-NEXT: [range] dtor
diff --git a/openmp/runtime/test/transform/reverse/parallel-wsloop-collapse-foreach.cpp b/openmp/runtime/test/transform/reverse/parallel-wsloop-collapse-foreach.cpp
new file mode 100644
index 0000000000000..240ef59bd6b4b
--- /dev/null
+++ b/openmp/runtime/test/transform/reverse/parallel-wsloop-collapse-foreach.cpp
@@ -0,0 +1,285 @@
+// RUN: %libomp-cxx20-compile-and-run | FileCheck %s --match-full-lines
+
+#ifndef HEADER
+#define HEADER
+
+#include <cstdlib>
+#include <cstdarg>
+#include <cstdio>
+#include <vector>
+
+struct Reporter {
+  const char *name;
+
+  Reporter(const char *name) : name(name) { print("ctor"); }
+
+  Reporter() : name("<anon>") { print("ctor"); }
+
+  Reporter(const Reporter &that) : name(that.name) { print("copy ctor"); }
+
+  Reporter(Reporter &&that) : name(that.name) { print("move ctor"); }
+
+  ~Reporter() { print("dtor"); }
+
+  const Reporter &operator=(const Reporter &that) {
+    print("copy assign");
+    this->name = that.name;
+    return *this;
+  }
+
+  const Reporter &operator=(Reporter &&that) {
+    print("move assign");
+    this->name = that.name;
+    return *this;
+  }
+
+  struct Iterator {
+    const Reporter *owner;
+    int pos;
+
+    Iterator(const Reporter *owner, int pos) : owner(owner), pos(pos) {}
+
+    Iterator(const Iterator &that) : owner(that.owner), pos(that.pos) {
+      owner->print("iterator copy ctor");
+    }
+
+    Iterator(Iterator &&that) : owner(that.owner), pos(that.pos) {
+      owner->print("iterator move ctor");
+    }
+
+    ~Iterator() { owner->print("iterator dtor"); }
+
+    const Iterator &operator=(const Iterator &that) {
+      owner->print("iterator copy assign");
+      this->owner = that.owner;
+      this->pos = that.pos;
+      return *this;
+    }
+
+    const Iterator &operator=(Iterator &&that) {
+      owner->print("iterator move assign");
+      this->owner = that.owner;
+      this->pos = that.pos;
+      return *this;
+    }
+
+    bool operator==(const Iterator &that) const {
+      owner->print("iterator %d == %d", 2 - this->pos, 2 - that.pos);
+      return this->pos == that.pos;
+    }
+
+    Iterator &operator++() {
+      owner->print("iterator prefix ++");
+      pos -= 1;
+      return *this;
+    }
+
+    Iterator operator++(int) {
+      owner->print("iterator postfix ++");
+      auto result = *this;
+      pos -= 1;
+      return result;
+    }
+
+    int operator*() const {
+      int result = 2 - pos;
+      owner->print("iterator deref: %i", result);
+      return result;
+    }
+
+    size_t operator-(const Iterator &that) const {
+      int result = (2 - this->pos) - (2 - that.pos);
+      owner->print("iterator distance: %d", result);
+      return result;
+    }
+
+    Iterator operator+(int steps) const {
+      owner->print("iterator advance: %i += %i", 2 - this->pos, steps);
+      return Iterator(owner, pos - steps);
+    }
+
+    void print(const char *msg) const { owner->print(msg); }
+  };
+
+  Iterator begin() const {
+    print("begin()");
+    return Iterator(this, 2);
+  }
+
+  Iterator end() const {
+    print("end()");
+    return Iterator(this, -1);
+  }
+
+  void print(const char *msg, ...) const {
+    va_list args;
+    va_start(args, msg);
+    printf("[%s] ", name);
+    vprintf(msg, args);
+    printf("\n");
+    va_end(args);
+  }
+};
+
+int main() {
+  printf("do\n");
+#pragma omp parallel for collapse(3) num_threads(1)
+  for (int i = 0; i < 3; ++i)
+#pragma omp reverse
+    for (Reporter c{"init-stmt"}; auto &&v : Reporter("range"))
+      for (int k = 0; k < 3; ++k)
+        printf("i=%d j=%d k=%d\n", i, v, k);
+  printf("done\n");
+  return EXIT_SUCCESS;
+}
+
+#endif /* HEADER */
+
+// CHECK:      do
+// CHECK-NEXT: [init-stmt] ctor
+// CHECK-NEXT: [range] ctor
+// CHECK-NEXT: [range] end()
+// CHECK-NEXT: [range] begin()
+// CHECK-NEXT: [range] begin()
+// CHECK-NEXT: [range] iterator distance: 3
+// CHECK-NEXT: [range] iterator advance: 0 += 2
+// CHECK-NEXT: [range] iterator move assign
+// CHECK-NEXT: [range] iterator deref: 2
+// CHECK-NEXT: i=0 j=2 k=0
+// CHECK-NEXT: [range] iterator dtor
+// CHECK-NEXT: [range] iterator advance: 0 += 2
+// CHECK-NEXT: [range] iterator move assign
+// CHECK-NEXT: [range] iterator deref: 2
+// CHECK-NEXT: i=0 j=2 k=1
+// CHECK-NEXT: [range] iterator dtor
+// CHECK-NEXT: [range] iterator advance: 0 += 2
+// CHECK-NEXT: [range] iterator move assign
+// CHECK-NEXT: [range] iterator deref: 2
+// CHECK-NEXT: i=0 j=2 k=2
+// CHECK-NEXT: [range] iterator dtor
+// CHECK-NEXT: [range] iterator advance: 0 += 1
+// CHECK-NEXT: [range] iterator move assign
+// CHECK-NEXT: [range] iterator deref: 1
+// CHECK-NEXT: i=0 j=1 k=0
+// CHECK-NEXT: [range] iterator dtor
+// CHECK-NEXT: [range] iterator advance: 0 += 1
+// CHECK-NEXT: [range] iterator move assign
+// CHECK-NEXT: [range] iterator deref: 1
+// CHECK-NEXT: i=0 j=1 k=1
+// CHECK-NEXT: [range] iterator dtor
+// CHECK-NEXT: [range] iterator advance: 0 += 1
+// CHECK-NEXT: [range] iterator move assign
+// CHECK-NEXT: [range] iterator deref: 1
+// CHECK-NEXT: i=0 j=1 k=2
+// CHECK-NEXT: [range] iterator dtor
+// CHECK-NEXT: [range] iterator advance: 0 += 0
+// CHECK-NEXT: [range] iterator move assign
+// CHECK-NEXT: [range] iterator deref: 0
+// CHECK-NEXT: i=0 j=0 k=0
+// CHECK-NEXT: [range] iterator dtor
+// CHECK-NEXT: [range] iterator advance: 0 += 0
+// CHECK-NEXT: [range] iterator move assign
+// CHECK-NEXT: [range] iterator deref: 0
+// CHECK-NEXT: i=0 j=0 k=1
+// CHECK-NEXT: [range] iterator dtor
+// CHECK-NEXT: [range] iterator advance: 0 += 0
+// CHECK-NEXT: [range] iterator move assign
+// CHECK-NEXT: [range] iterator deref: 0
+// CHECK-NEXT: i=0 j=0 k=2
+// CHECK-NEXT: [range] iterator dtor
+// CHECK-NEXT: [range] iterator advance: 0 += 2
+// CHECK-NEXT: [range] iterator move assign
+// CHECK-NEXT: [range] iterator deref: 2
+// CHECK-NEXT: i=1 j=2 k=0
+// CHECK-NEXT: [range] iterator dtor
+// CHECK-NEXT: [range] iterator advance: 0 += 2
+// CHECK-NEXT: [range] iterator move assign
+// CHECK-NEXT: [range] iterator deref: 2
+// CHECK-NEXT: i=1 j=2 k=1
+// CHECK-NEXT: [range] iterator dtor
+// CHECK-NEXT: [range] iterator advance: 0 += 2
+// CHECK-NEXT: [range] iterator move assign
+// CHECK-NEXT: [range] iterator deref: 2
+// CHECK-NEXT: i=1 j=2 k=2
+// CHECK-NEXT: [range] iterator dtor
+// CHECK-NEXT: [range] iterator advance: 0 += 1
+// CHECK-NEXT: [range] iterator move assign
+// CHECK-NEXT: [range] iterator deref: 1
+// CHECK-NEXT: i=1 j=1 k=0
+// CHECK-NEXT: [range] iterator dtor
+// CHECK-NEXT: [range] iterator advance: 0 += 1
+// CHECK-NEXT: [range] iterator move assign
+// CHECK-NEXT: [range] iterator deref: 1
+// CHECK-NEXT: i=1 j=1 k=1
+// CHECK-NEXT: [range] iterator dtor
+// CHECK-NEXT: [range] iterator advance: 0 += 1
+// CHECK-NEXT: [range] iterator move assign
+// CHECK-NEXT: [range] iterator deref: 1
+// CHECK-NEXT: i=1 j=1 k=2
+// CHECK-NEXT: [range] iterator dtor
+// CHECK-NEXT: [range] iterator advance: 0 += 0
+// CHECK-NEXT: [range] iterator move assign
+// CHECK-NEXT: [range] iterator deref: 0
+// CHECK-NEXT: i=1 j=0 k=0
+// CHECK-NEXT: [range] iterator dtor
+// CHECK-NEXT: [range] iterator advance: 0 += 0
+// CHECK-NEXT: [range] iterator move assign
+// CHECK-NEXT: [range] iterator deref: 0
+// CHECK-NEXT: i=1 j=0 k=1
+// CHECK-NEXT: [range] iterator dtor
+// CHECK-NEXT: [range] iterator advance: 0 += 0
+// CHECK-NEXT: [range] iterator move assign
+// CHECK-NEXT: [range] iterator deref: 0
+// CHECK-NEXT: i=1 j=0 k=2
+// CHECK-NEXT: [range] iterator dtor
+// CHECK-NEXT: [range] iterator advance: 0 += 2
+// CHECK-NEXT: [range] iterator move assign
+// CHECK-NEXT: [range] iterator deref: 2
+// CHECK-NEXT: i=2 j=2 k=0
+// CHECK-NEXT: [range] iterator dtor
+// CHECK-NEXT: [range] iterator advance: 0 += 2
+// CHECK-NEXT: [range] iterator move assign
+// CHECK-NEXT: [range] iterator deref: 2
+// CHECK-NEXT: i=2 j=2 k=1
+// CHECK-NEXT: [range] iterator dtor
+// CHECK-NEXT: [range] iterator advance: 0 += 2
+// CHECK-NEXT: [range] iterator move assign
+// CHECK-NEXT: [range] iterator deref: 2
+// CHECK-NEXT: i=2 j=2 k=2
+// CHECK-NEXT: [range] iterator dtor
+// CHECK-NEXT: [range] iterator advance: 0 += 1
+// CHECK-NEXT: [range] iterator move assign
+// CHECK-NEXT: [range] iterator deref: 1
+// CHECK-NEXT: i=2 j=1 k=0
+// CHECK-NEXT: [range] iterator dtor
+// CHECK-NEXT: [range] iterator advance: 0 += 1
+// CHECK-NEXT: [range] iterator move assign
+// CHECK-NEXT: [range] iterator deref: 1
+// CHECK-NEXT: i=2 j=1 k=1
+// CHECK-NEXT: [range] iterator dtor
+// CHECK-NEXT: [range] iterator advance: 0 += 1
+// CHECK-NEXT: [range] iterator move assign
+// CHECK-NEXT: [range] iterator deref: 1
+// CHECK-NEXT: i=2 j=1 k=2
+// CHECK-NEXT: [range] iterator dtor
+// CHECK-NEXT: [range] iterator advance: 0 += 0
+// CHECK-NEXT: [range] iterator move assign
+// CHECK-NEXT: [range] iterator deref: 0
+// CHECK-NEXT: i=2 j=0 k=0
+// CHECK-NEXT: [range] iterator dtor
+// CHECK-NEXT: [range] iterator advance: 0 += 0
+// CHECK-NEXT: [range] iterator move assign
+// CHECK-NEXT: [range] iterator deref: 0
+// CHECK-NEXT: i=2 j=0 k=1
+// CHECK-NEXT: [range] iterator dtor
+// CHECK-NEXT: [range] iterator advance: 0 += 0
+// CHECK-NEXT: [range] iterator move assign
+// CHECK-NEXT: [range] iterator deref: 0
+// CHECK-NEXT: i=2 j=0 k=2
+// CHECK-NEXT: [range] iterator dtor
+// CHECK-NEXT: [range] iterator dtor
+// CHECK-NEXT: [range] iterator dtor
+// CHECK-NEXT: [range] iterator dtor
+// CHECK-NEXT: [range] dtor
+// CHECK-NEXT: [init-stmt] dtor
+// CHECK-NEXT: done
diff --git a/openmp/runtime/test/transform/reverse/parallel-wsloop-collapse-intfor.cpp b/openmp/runtime/test/transform/reverse/parallel-wsloop-collapse-intfor.cpp
new file mode 100644
index 0000000000000..ae545b863d86c
--- /dev/null
+++ b/openmp/runtime/test/transform/reverse/parallel-wsloop-collapse-intfor.cpp
@@ -0,0 +1,51 @@
+// RUN: %libomp-cxx-compile-and-run | FileCheck %s --match-full-lines
+
+#ifndef HEADER
+#define HEADER
+
+#include <cstdlib>
+#include <cstdio>
+
+int main() {
+  printf("do\n");
+#pragma omp parallel for collapse(3) num_threads(1)
+  for (int i = 0; i < 3; ++i)
+#pragma omp reverse
+    for (int j = 0; j < 3; ++j)
+      for (int k = 0; k < 3; ++k)
+        printf("i=%d j=%d k=%d\n", i, j, k);
+  printf("done\n");
+  return EXIT_SUCCESS;
+}
+
+#endif /* HEADER */
+
+// CHECK:      do
+// CHECK-NEXT: i=0 j=2 k=0
+// CHECK-NEXT: i=0 j=2 k=1
+// CHECK-NEXT: i=0 j=2 k=2
+// CHECK-NEXT: i=0 j=1 k=0
+// CHECK-NEXT: i=0 j=1 k=1
+// CHECK-NEXT: i=0 j=1 k=2
+// CHECK-NEXT: i=0 j=0 k=0
+// CHECK-NEXT: i=0 j=0 k=1
+// CHECK-NEXT: i=0 j=0 k=2
+// CHECK-NEXT: i=1 j=2 k=0
+// CHECK-NEXT: i=1 j=2 k=1
+// CHECK-NEXT: i=1 j=2 k=2
+// CHECK-NEXT: i=1 j=1 k=0
+// CHECK-NEXT: i=1 j=1 k=1
+// CHECK-NEXT: i=1 j=1 k=2
+// CHECK-NEXT: i=1 j=0 k=0
+// CHECK-NEXT: i=1 j=0 k=1
+// CHECK-NEXT: i=1 j=0 k=2
+// CHECK-NEXT: i=2 j=2 k=0
+// CHECK-NEXT: i=2 j=2 k=1
+// CHECK-NEXT: i=2 j=2 k=2
+// CHECK-NEXT: i=2 j=1 k=0
+// CHECK-NEXT: i=2 j=1 k=1
+// CHECK-NEXT: i=2 j=1 k=2
+// CHECK-NEXT: i=2 j=0 k=0
+// CHECK-NEXT: i=2 j=0 k=1
+// CHECK-NEXT: i=2 j=0 k=2
+// CHECK-NEXT: done

>From 0a38da39e5520dcd5e4da07275d79d35430dd846 Mon Sep 17 00:00:00 2001
From: Michael Kruse <llvm-project at meinersbur.de>
Date: Tue, 21 May 2024 17:32:00 +0200
Subject: [PATCH 2/2] Address review

---
 clang/include/clang/AST/StmtOpenMP.h  |  4 +---
 clang/include/clang/Sema/SemaOpenMP.h |  6 ++----
 clang/lib/AST/StmtOpenMP.cpp          |  5 ++---
 clang/lib/Sema/SemaOpenMP.cpp         | 20 +++++++++-----------
 4 files changed, 14 insertions(+), 21 deletions(-)

diff --git a/clang/include/clang/AST/StmtOpenMP.h b/clang/include/clang/AST/StmtOpenMP.h
index 4be2e2d3a4605..fb7f413162fad 100644
--- a/clang/include/clang/AST/StmtOpenMP.h
+++ b/clang/include/clang/AST/StmtOpenMP.h
@@ -5748,15 +5748,13 @@ class OMPReverseDirective final : public OMPLoopTransformationDirective {
   /// \param C         Context of the AST.
   /// \param StartLoc  Location of the introducer (e.g. the 'omp' token).
   /// \param EndLoc    Location of the directive's end (e.g. the tok::eod).
-  /// \param Clauses   The directive's clauses.
   /// \param AssociatedStmt  The outermost associated loop.
   /// \param TransformedStmt The loop nest after tiling, or nullptr in
   ///                        dependent contexts.
   /// \param PreInits   Helper preinits statements for the loop nest.
   static OMPReverseDirective *
   Create(const ASTContext &C, SourceLocation StartLoc, SourceLocation EndLoc,
-         ArrayRef<OMPClause *> Clauses, Stmt *AssociatedStmt,
-         Stmt *TransformedStmt, Stmt *PreInits);
+         Stmt *AssociatedStmt, Stmt *TransformedStmt, Stmt *PreInits);
 
   /// Build an empty '#pragma omp reverse' AST node for deserialization.
   ///
diff --git a/clang/include/clang/Sema/SemaOpenMP.h b/clang/include/clang/Sema/SemaOpenMP.h
index e36a90ba4e1b9..ca91bffe24f6f 100644
--- a/clang/include/clang/Sema/SemaOpenMP.h
+++ b/clang/include/clang/Sema/SemaOpenMP.h
@@ -422,10 +422,8 @@ class SemaOpenMP : public SemaBase {
   StmtResult ActOnOpenMPUnrollDirective(ArrayRef<OMPClause *> Clauses,
                                         Stmt *AStmt, SourceLocation StartLoc,
                                         SourceLocation EndLoc);
-  /// Called on well-formed '#pragma omp reverse' after parsing of its clauses
-  /// and the associated statement.
-  StmtResult ActOnOpenMPReverseDirective(ArrayRef<OMPClause *> Clauses,
-                                         Stmt *AStmt, SourceLocation StartLoc,
+  /// Called on well-formed '#pragma omp reverse'.
+  StmtResult ActOnOpenMPReverseDirective(Stmt *AStmt, SourceLocation StartLoc,
                                          SourceLocation EndLoc);
   /// Called on well-formed '\#pragma omp for' after parsing
   /// of the associated statement.
diff --git a/clang/lib/AST/StmtOpenMP.cpp b/clang/lib/AST/StmtOpenMP.cpp
index 0be0d9d2cfa94..83b8a08e9af73 100644
--- a/clang/lib/AST/StmtOpenMP.cpp
+++ b/clang/lib/AST/StmtOpenMP.cpp
@@ -451,11 +451,10 @@ OMPUnrollDirective *OMPUnrollDirective::CreateEmpty(const ASTContext &C,
 
 OMPReverseDirective *
 OMPReverseDirective::Create(const ASTContext &C, SourceLocation StartLoc,
-                            SourceLocation EndLoc,
-                            ArrayRef<OMPClause *> Clauses, Stmt *AssociatedStmt,
+                            SourceLocation EndLoc, Stmt *AssociatedStmt,
                             Stmt *TransformedStmt, Stmt *PreInits) {
   OMPReverseDirective *Dir = createDirective<OMPReverseDirective>(
-      C, Clauses, AssociatedStmt, TransformedStmtOffset + 1, StartLoc, EndLoc);
+      C, {}, AssociatedStmt, TransformedStmtOffset + 1, StartLoc, EndLoc);
   Dir->setTransformedStmt(TransformedStmt);
   Dir->setPreInits(PreInits);
   return Dir;
diff --git a/clang/lib/Sema/SemaOpenMP.cpp b/clang/lib/Sema/SemaOpenMP.cpp
index 7b9898704eb1c..f3fbc462aa4ab 100644
--- a/clang/lib/Sema/SemaOpenMP.cpp
+++ b/clang/lib/Sema/SemaOpenMP.cpp
@@ -6548,8 +6548,9 @@ StmtResult SemaOpenMP::ActOnOpenMPExecutableDirective(
                                      EndLoc);
     break;
   case OMPD_reverse:
-    Res = ActOnOpenMPReverseDirective(ClausesWithImplicit, AStmt, StartLoc,
-                                      EndLoc);
+    assert(ClausesWithImplicit.empty() &&
+           "reverse directive does not support any clauses");
+    Res = ActOnOpenMPReverseDirective(AStmt, StartLoc, EndLoc);
     break;
   case OMPD_for:
     Res = ActOnOpenMPForDirective(ClausesWithImplicit, AStmt, StartLoc, EndLoc,
@@ -15753,14 +15754,11 @@ StmtResult SemaOpenMP::ActOnOpenMPUnrollDirective(ArrayRef<OMPClause *> Clauses,
                                     buildPreInits(Context, PreInits));
 }
 
-StmtResult
-SemaOpenMP::ActOnOpenMPReverseDirective(ArrayRef<OMPClause *> Clauses,
-                                        Stmt *AStmt, SourceLocation StartLoc,
-                                        SourceLocation EndLoc) {
+StmtResult SemaOpenMP::ActOnOpenMPReverseDirective(Stmt *AStmt,
+                                                   SourceLocation StartLoc,
+                                                   SourceLocation EndLoc) {
   ASTContext &Context = getASTContext();
   Scope *CurScope = SemaRef.getCurScope();
-  assert(Clauses.empty() && "reverse directive does not accept any clauses; "
-                            "must have beed checked before");
 
   // Empty statement should only be possible if there already was an error.
   if (!AStmt)
@@ -15778,8 +15776,8 @@ SemaOpenMP::ActOnOpenMPReverseDirective(ArrayRef<OMPClause *> Clauses,
   // Delay applying the transformation to when template is completely
   // instantiated.
   if (SemaRef.CurContext->isDependentContext())
-    return OMPReverseDirective::Create(Context, StartLoc, EndLoc, Clauses,
-                                       AStmt, nullptr, nullptr);
+    return OMPReverseDirective::Create(Context, StartLoc, EndLoc, AStmt,
+                                       nullptr, nullptr);
 
   assert(LoopHelpers.size() == NumLoops &&
          "Expecting a single-dimensional loop iteration space");
@@ -15931,7 +15929,7 @@ SemaOpenMP::ActOnOpenMPReverseDirective(ArrayRef<OMPClause *> Clauses,
       ForStmt(Context, Init.get(), Cond.get(), nullptr, Incr.get(),
               ReversedBody, LoopHelper.Init->getBeginLoc(),
               LoopHelper.Init->getBeginLoc(), LoopHelper.Inc->getEndLoc());
-  return OMPReverseDirective::Create(Context, StartLoc, EndLoc, Clauses, AStmt,
+  return OMPReverseDirective::Create(Context, StartLoc, EndLoc, AStmt,
                                      ReversedFor,
                                      buildPreInits(Context, PreInits));
 }



More information about the llvm-branch-commits mailing list