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

Michael Kruse via Openmp-commits openmp-commits at lists.llvm.org
Wed Jul 17 04:40:37 PDT 2024


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

>From d6057c46719b39943a402f074a59465405ca6120 Mon Sep 17 00:00:00 2001
From: Michael Kruse <llvm-project at meinersbur.de>
Date: Tue, 21 May 2024 15:22:21 +0200
Subject: [PATCH 01/10] [Clang][OpenMP] Fix tile/unroll on iterator- and
 foreach-loops

---
 clang/include/clang/Sema/SemaOpenMP.h         |   4 +-
 clang/lib/CodeGen/CGStmtOpenMP.cpp            |  29 +-
 clang/lib/Sema/SemaOpenMP.cpp                 | 190 ++--
 clang/test/OpenMP/tile_codegen.cpp            | 887 +++++++++++++-----
 .../OpenMP/tile_codegen_for_dependent.cpp     | 130 +--
 clang/test/OpenMP/tile_codegen_tile_for.cpp   | 218 ++---
 openmp/runtime/test/lit.cfg                   |   4 +
 .../runtime/test/transform/tile/foreach.cpp   | 228 +++++
 .../runtime/test/transform/tile/iterfor.cpp   | 233 +++++
 .../tile/parallel-wsloop-collapse-foreach.cpp | 366 ++++++++
 .../test/transform/unroll/factor_foreach.cpp  | 162 ++++
 .../test/transform/unroll/factor_intfor.c     |  25 +
 .../test/transform/unroll/factor_iterfor.cpp  | 169 ++++
 ...actor_parallel-wsloop-collapse-foreach.cpp | 199 ++++
 ...factor_parallel-wsloop-collapse-intfor.cpp |  32 +
 .../test/transform/unroll/full_intfor.c       |  25 +
 .../test/transform/unroll/heuristic_intfor.c  |  25 +
 .../test/transform/unroll/partial_intfor.c    |  25 +
 18 files changed, 2506 insertions(+), 445 deletions(-)
 create mode 100644 openmp/runtime/test/transform/tile/foreach.cpp
 create mode 100644 openmp/runtime/test/transform/tile/iterfor.cpp
 create mode 100644 openmp/runtime/test/transform/tile/parallel-wsloop-collapse-foreach.cpp
 create mode 100644 openmp/runtime/test/transform/unroll/factor_foreach.cpp
 create mode 100644 openmp/runtime/test/transform/unroll/factor_intfor.c
 create mode 100644 openmp/runtime/test/transform/unroll/factor_iterfor.cpp
 create mode 100644 openmp/runtime/test/transform/unroll/factor_parallel-wsloop-collapse-foreach.cpp
 create mode 100644 openmp/runtime/test/transform/unroll/factor_parallel-wsloop-collapse-intfor.cpp
 create mode 100644 openmp/runtime/test/transform/unroll/full_intfor.c
 create mode 100644 openmp/runtime/test/transform/unroll/heuristic_intfor.c
 create mode 100644 openmp/runtime/test/transform/unroll/partial_intfor.c

diff --git a/clang/include/clang/Sema/SemaOpenMP.h b/clang/include/clang/Sema/SemaOpenMP.h
index 9927459bbc594..51981e1c9a8b9 100644
--- a/clang/include/clang/Sema/SemaOpenMP.h
+++ b/clang/include/clang/Sema/SemaOpenMP.h
@@ -1390,9 +1390,7 @@ class SemaOpenMP : public SemaBase {
   bool checkTransformableLoopNest(
       OpenMPDirectiveKind Kind, Stmt *AStmt, int NumLoops,
       SmallVectorImpl<OMPLoopBasedDirective::HelperExprs> &LoopHelpers,
-      Stmt *&Body,
-      SmallVectorImpl<SmallVector<llvm::PointerUnion<Stmt *, Decl *>, 0>>
-          &OriginalInits);
+      Stmt *&Body, SmallVectorImpl<SmallVector<Stmt *, 0>> &OriginalInits);
 
   /// Helper to keep information about the current `omp begin/end declare
   /// variant` nesting.
diff --git a/clang/lib/CodeGen/CGStmtOpenMP.cpp b/clang/lib/CodeGen/CGStmtOpenMP.cpp
index eac5ef3262937..6410f9e102c90 100644
--- a/clang/lib/CodeGen/CGStmtOpenMP.cpp
+++ b/clang/lib/CodeGen/CGStmtOpenMP.cpp
@@ -142,7 +142,7 @@ class OMPTeamsScope final : public OMPLexicalScope {
 /// of used expression from loop statement.
 class OMPLoopScope : public CodeGenFunction::RunCleanupsScope {
   void emitPreInitStmt(CodeGenFunction &CGF, const OMPLoopBasedDirective &S) {
-    const DeclStmt *PreInits;
+    const Stmt *PreInits;
     CodeGenFunction::OMPMapVars PreCondVars;
     if (auto *LD = dyn_cast<OMPLoopDirective>(&S)) {
       llvm::DenseSet<const VarDecl *> EmittedAsPrivate;
@@ -182,17 +182,34 @@ class OMPLoopScope : public CodeGenFunction::RunCleanupsScope {
             }
             return false;
           });
-      PreInits = cast_or_null<DeclStmt>(LD->getPreInits());
+      PreInits = LD->getPreInits();
     } else if (const auto *Tile = dyn_cast<OMPTileDirective>(&S)) {
-      PreInits = cast_or_null<DeclStmt>(Tile->getPreInits());
+      PreInits = Tile->getPreInits();
     } else if (const auto *Unroll = dyn_cast<OMPUnrollDirective>(&S)) {
-      PreInits = cast_or_null<DeclStmt>(Unroll->getPreInits());
+      PreInits = Unroll->getPreInits();
     } else {
       llvm_unreachable("Unknown loop-based directive kind.");
     }
     if (PreInits) {
-      for (const auto *I : PreInits->decls())
-        CGF.EmitVarDecl(cast<VarDecl>(*I));
+      // CompoundStmts and DeclStmts are used as lists of PreInit statements and
+      // declarations. Since declarations must be visible in the the following
+      // that they initialize, unpack the ComboundStmt they are nested in.
+      SmallVector<const Stmt *> PreInitStmts;
+      if (auto *PreInitCompound = dyn_cast<CompoundStmt>(PreInits))
+        llvm::append_range(PreInitStmts, PreInitCompound->body());
+      else
+        PreInitStmts.push_back(PreInits);
+
+      for (const Stmt *S : PreInitStmts) {
+        // EmitStmt skips any OMPCapturedExprDecls, but needs to be emitted
+        // here.
+        if (auto *PreInitDecl = dyn_cast<DeclStmt>(S)) {
+          for (Decl *I : PreInitDecl->decls())
+            CGF.EmitVarDecl(cast<VarDecl>(*I));
+          continue;
+        }
+        CGF.EmitStmt(S);
+      }
     }
     PreCondVars.restore(CGF);
   }
diff --git a/clang/lib/Sema/SemaOpenMP.cpp b/clang/lib/Sema/SemaOpenMP.cpp
index 6110e5229b076..663dbb927250e 100644
--- a/clang/lib/Sema/SemaOpenMP.cpp
+++ b/clang/lib/Sema/SemaOpenMP.cpp
@@ -9828,6 +9828,23 @@ buildPreInits(ASTContext &Context,
   return nullptr;
 }
 
+/// Build pre-init statement for the given statements.
+static Stmt *buildPreInits(ASTContext &Context, ArrayRef<Stmt *> PreInits) {
+  if (!PreInits.empty()) {
+    SmallVector<Stmt *> Stmts;
+    for (Stmt *S : PreInits) {
+      // Do not nest CompoundStmts.
+      if (auto *CS = dyn_cast<CompoundStmt>(S)) {
+        llvm::append_range(Stmts, CS->body());
+        continue;
+      }
+      Stmts.push_back(S);
+    }
+    return CompoundStmt::Create(Context, PreInits, FPOptionsOverride(), {}, {});
+  }
+  return nullptr;
+}
+
 /// Build postupdate expression for the given list of postupdates expressions.
 static Expr *buildPostUpdate(Sema &S, ArrayRef<Expr *> PostUpdates) {
   Expr *PostUpdate = nullptr;
@@ -9924,11 +9941,24 @@ checkOpenMPLoop(OpenMPDirectiveKind DKind, Expr *CollapseLoopCountExpr,
             Stmt *DependentPreInits = Transform->getPreInits();
             if (!DependentPreInits)
               return;
-            for (Decl *C : cast<DeclStmt>(DependentPreInits)->getDeclGroup()) {
-              auto *D = cast<VarDecl>(C);
-              DeclRefExpr *Ref = buildDeclRefExpr(SemaRef, D, D->getType(),
-                                                  Transform->getBeginLoc());
-              Captures[Ref] = Ref;
+
+            // Search for pre-init declared variables that need to be captured
+            // to be referenceable inside the directive.
+            SmallVector<Stmt *> Constituents;
+            if (auto *CS = dyn_cast<CompoundStmt>(DependentPreInits))
+              llvm::append_range(Constituents, CS->body());
+            else
+              Constituents.push_back(DependentPreInits);
+            for (Stmt *S : Constituents) {
+              if (DeclStmt *DC = dyn_cast<DeclStmt>(S)) {
+                for (Decl *C : DC->decls()) {
+                  auto *D = cast<VarDecl>(C);
+                  DeclRefExpr *Ref = buildDeclRefExpr(
+                      SemaRef, D, D->getType().getNonReferenceType(),
+                      Transform->getBeginLoc());
+                  Captures[Ref] = Ref;
+                }
+              }
             }
           }))
     return 0;
@@ -15059,9 +15089,7 @@ StmtResult SemaOpenMP::ActOnOpenMPTargetTeamsDistributeSimdDirective(
 bool SemaOpenMP::checkTransformableLoopNest(
     OpenMPDirectiveKind Kind, Stmt *AStmt, int NumLoops,
     SmallVectorImpl<OMPLoopBasedDirective::HelperExprs> &LoopHelpers,
-    Stmt *&Body,
-    SmallVectorImpl<SmallVector<llvm::PointerUnion<Stmt *, Decl *>, 0>>
-        &OriginalInits) {
+    Stmt *&Body, SmallVectorImpl<SmallVector<Stmt *, 0>> &OriginalInits) {
   OriginalInits.emplace_back();
   bool Result = OMPLoopBasedDirective::doForAllLoops(
       AStmt->IgnoreContainers(), /*TryImperfectlyNestedLoops=*/false, NumLoops,
@@ -15097,14 +15125,75 @@ bool SemaOpenMP::checkTransformableLoopNest(
           llvm_unreachable("Unhandled loop transformation");
         if (!DependentPreInits)
           return;
-        llvm::append_range(OriginalInits.back(),
-                           cast<DeclStmt>(DependentPreInits)->getDeclGroup());
+        // CompoundStmts are used as lists of other statements, add their
+        // contents, not the lists themselves to avoid nesting. This is
+        // necessary because DeclStmts need to be visible after the pre-init.
+        else if (auto *CS = dyn_cast<CompoundStmt>(DependentPreInits))
+          llvm::append_range(OriginalInits.back(), CS->body());
+        else
+          OriginalInits.back().push_back(DependentPreInits);
       });
   assert(OriginalInits.back().empty() && "No preinit after innermost loop");
   OriginalInits.pop_back();
   return Result;
 }
 
+/// Add preinit statements that need to be propageted from the selected loop.
+static void addLoopPreInits(ASTContext &Context,
+                            OMPLoopBasedDirective::HelperExprs &LoopHelper,
+                            Stmt *LoopStmt, ArrayRef<Stmt *> OriginalInit,
+                            SmallVectorImpl<Stmt *> &PreInits) {
+
+  // For range-based for-statements, ensure that their syntactic sugar is
+  // executed by adding them as pre-init statements.
+  if (auto *CXXRangeFor = dyn_cast<CXXForRangeStmt>(LoopStmt)) {
+    Stmt *RangeInit = CXXRangeFor->getInit();
+    if (RangeInit)
+      PreInits.push_back(RangeInit);
+
+    DeclStmt *RangeStmt = CXXRangeFor->getRangeStmt();
+    PreInits.push_back(new (Context) DeclStmt(RangeStmt->getDeclGroup(),
+                                              RangeStmt->getBeginLoc(),
+                                              RangeStmt->getEndLoc()));
+
+    DeclStmt *RangeEnd = CXXRangeFor->getEndStmt();
+    PreInits.push_back(new (Context) DeclStmt(RangeEnd->getDeclGroup(),
+                                              RangeEnd->getBeginLoc(),
+                                              RangeEnd->getEndLoc()));
+  }
+
+  llvm::append_range(PreInits, OriginalInit);
+
+  // List of OMPCapturedExprDecl, for __begin, __end, and NumIterations
+  if (auto *PI = cast_or_null<DeclStmt>(LoopHelper.PreInits)) {
+    PreInits.push_back(new (Context) DeclStmt(
+        PI->getDeclGroup(), PI->getBeginLoc(), PI->getEndLoc()));
+  }
+
+  // Gather declarations for the data members used as counters.
+  for (Expr *CounterRef : LoopHelper.Counters) {
+    auto *CounterDecl = cast<DeclRefExpr>(CounterRef)->getDecl();
+    if (isa<OMPCapturedExprDecl>(CounterDecl))
+      PreInits.push_back(new (Context) DeclStmt(
+          DeclGroupRef(CounterDecl), SourceLocation(), SourceLocation()));
+  }
+}
+
+/// Collect the loop statements (ForStmt or CXXRangeForStmt) of the affected
+/// loop of a construct.
+static void collectLoopStmts(Stmt *AStmt, MutableArrayRef<Stmt *> LoopStmts) {
+  size_t NumLoops = LoopStmts.size();
+  OMPLoopBasedDirective::doForAllLoops(
+      AStmt, /*TryImperfectlyNestedLoops=*/false, NumLoops,
+      [LoopStmts](unsigned Cnt, Stmt *CurStmt) {
+        assert(!LoopStmts[Cnt] && "Loop statement must not yet be assigned");
+        LoopStmts[Cnt] = CurStmt;
+        return false;
+      });
+  assert(llvm::all_of(LoopStmts, [](Stmt *LoopStmt) { return LoopStmt; }) &&
+         "Expecting a loop statement for each affected loop");
+}
+
 StmtResult SemaOpenMP::ActOnOpenMPTileDirective(ArrayRef<OMPClause *> Clauses,
                                                 Stmt *AStmt,
                                                 SourceLocation StartLoc,
@@ -15126,8 +15215,7 @@ StmtResult SemaOpenMP::ActOnOpenMPTileDirective(ArrayRef<OMPClause *> Clauses,
   // Verify and diagnose loop nest.
   SmallVector<OMPLoopBasedDirective::HelperExprs, 4> LoopHelpers(NumLoops);
   Stmt *Body = nullptr;
-  SmallVector<SmallVector<llvm::PointerUnion<Stmt *, Decl *>, 0>, 4>
-      OriginalInits;
+  SmallVector<SmallVector<Stmt *, 0>, 4> OriginalInits;
   if (!checkTransformableLoopNest(OMPD_tile, AStmt, NumLoops, LoopHelpers, Body,
                                   OriginalInits))
     return StmtError();
@@ -15144,7 +15232,11 @@ StmtResult SemaOpenMP::ActOnOpenMPTileDirective(ArrayRef<OMPClause *> Clauses,
          "Expecting loop iteration space dimensionality to match number of "
          "affected loops");
 
-  SmallVector<Decl *, 4> PreInits;
+  // Collect all affected loop statements.
+  SmallVector<Stmt *> LoopStmts(NumLoops, nullptr);
+  collectLoopStmts(AStmt, LoopStmts);
+
+  SmallVector<Stmt *, 4> PreInits;
   CaptureVars CopyTransformer(SemaRef);
 
   // Create iteration variables for the generated loops.
@@ -15184,20 +15276,9 @@ StmtResult SemaOpenMP::ActOnOpenMPTileDirective(ArrayRef<OMPClause *> Clauses,
           &SemaRef.PP.getIdentifierTable().get(TileCntName));
       TileIndVars[I] = TileCntDecl;
     }
-    for (auto &P : OriginalInits[I]) {
-      if (auto *D = P.dyn_cast<Decl *>())
-        PreInits.push_back(D);
-      else if (auto *PI = dyn_cast_or_null<DeclStmt>(P.dyn_cast<Stmt *>()))
-        PreInits.append(PI->decl_begin(), PI->decl_end());
-    }
-    if (auto *PI = cast_or_null<DeclStmt>(LoopHelper.PreInits))
-      PreInits.append(PI->decl_begin(), PI->decl_end());
-    // Gather declarations for the data members used as counters.
-    for (Expr *CounterRef : LoopHelper.Counters) {
-      auto *CounterDecl = cast<DeclRefExpr>(CounterRef)->getDecl();
-      if (isa<OMPCapturedExprDecl>(CounterDecl))
-        PreInits.push_back(CounterDecl);
-    }
+
+    addLoopPreInits(Context, LoopHelper, LoopStmts[I], OriginalInits[I],
+                    PreInits);
   }
 
   // Once the original iteration values are set, append the innermost body.
@@ -15246,19 +15327,20 @@ StmtResult SemaOpenMP::ActOnOpenMPTileDirective(ArrayRef<OMPClause *> Clauses,
     OMPLoopBasedDirective::HelperExprs &LoopHelper = LoopHelpers[I];
     Expr *NumIterations = LoopHelper.NumIterations;
     auto *OrigCntVar = cast<DeclRefExpr>(LoopHelper.Counters[0]);
-    QualType CntTy = OrigCntVar->getType();
+    QualType IVTy = NumIterations->getType();
+    Stmt *LoopStmt = LoopStmts[I];
 
     // Commonly used variables. One of the constraints of an AST is that every
     // node object must appear at most once, hence we define lamdas that create
     // a new AST node at every use.
-    auto MakeTileIVRef = [&SemaRef = this->SemaRef, &TileIndVars, I, CntTy,
+    auto MakeTileIVRef = [&SemaRef = this->SemaRef, &TileIndVars, I, IVTy,
                           OrigCntVar]() {
-      return buildDeclRefExpr(SemaRef, TileIndVars[I], CntTy,
+      return buildDeclRefExpr(SemaRef, TileIndVars[I], IVTy,
                               OrigCntVar->getExprLoc());
     };
-    auto MakeFloorIVRef = [&SemaRef = this->SemaRef, &FloorIndVars, I, CntTy,
+    auto MakeFloorIVRef = [&SemaRef = this->SemaRef, &FloorIndVars, I, IVTy,
                            OrigCntVar]() {
-      return buildDeclRefExpr(SemaRef, FloorIndVars[I], CntTy,
+      return buildDeclRefExpr(SemaRef, FloorIndVars[I], IVTy,
                               OrigCntVar->getExprLoc());
     };
 
@@ -15320,6 +15402,8 @@ StmtResult SemaOpenMP::ActOnOpenMPTileDirective(ArrayRef<OMPClause *> Clauses,
     // further into the inner loop.
     SmallVector<Stmt *, 4> BodyParts;
     BodyParts.append(LoopHelper.Updates.begin(), LoopHelper.Updates.end());
+    if (auto *SourceCXXFor = dyn_cast<CXXForRangeStmt>(LoopStmt))
+      BodyParts.push_back(SourceCXXFor->getLoopVarStmt());
     BodyParts.push_back(Inner);
     Inner = CompoundStmt::Create(Context, BodyParts, FPOptionsOverride(),
                                  Inner->getBeginLoc(), Inner->getEndLoc());
@@ -15334,12 +15418,14 @@ StmtResult SemaOpenMP::ActOnOpenMPTileDirective(ArrayRef<OMPClause *> Clauses,
     auto &LoopHelper = LoopHelpers[I];
     Expr *NumIterations = LoopHelper.NumIterations;
     DeclRefExpr *OrigCntVar = cast<DeclRefExpr>(LoopHelper.Counters[0]);
-    QualType CntTy = OrigCntVar->getType();
+    QualType IVTy = NumIterations->getType();
 
-    // Commonly used variables.
-    auto MakeFloorIVRef = [&SemaRef = this->SemaRef, &FloorIndVars, I, CntTy,
+    // Commonly used variables. One of the constraints of an AST is that every
+    // node object must appear at most once, hence we define lamdas that create
+    // a new AST node at every use.
+    auto MakeFloorIVRef = [&SemaRef = this->SemaRef, &FloorIndVars, I, IVTy,
                            OrigCntVar]() {
-      return buildDeclRefExpr(SemaRef, FloorIndVars[I], CntTy,
+      return buildDeclRefExpr(SemaRef, FloorIndVars[I], IVTy,
                               OrigCntVar->getExprLoc());
     };
 
@@ -15405,8 +15491,7 @@ StmtResult SemaOpenMP::ActOnOpenMPUnrollDirective(ArrayRef<OMPClause *> Clauses,
   Stmt *Body = nullptr;
   SmallVector<OMPLoopBasedDirective::HelperExprs, NumLoops> LoopHelpers(
       NumLoops);
-  SmallVector<SmallVector<llvm::PointerUnion<Stmt *, Decl *>, 0>, NumLoops + 1>
-      OriginalInits;
+  SmallVector<SmallVector<Stmt *, 0>, NumLoops + 1> OriginalInits;
   if (!checkTransformableLoopNest(OMPD_unroll, AStmt, NumLoops, LoopHelpers,
                                   Body, OriginalInits))
     return StmtError();
@@ -15418,6 +15503,10 @@ StmtResult SemaOpenMP::ActOnOpenMPUnrollDirective(ArrayRef<OMPClause *> Clauses,
     return OMPUnrollDirective::Create(Context, StartLoc, EndLoc, Clauses, AStmt,
                                       NumGeneratedLoops, 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();
 
   if (FullClause) {
@@ -15481,24 +15570,13 @@ StmtResult SemaOpenMP::ActOnOpenMPUnrollDirective(ArrayRef<OMPClause *> Clauses,
   // of a canonical loop nest where these PreInits are emitted before the
   // outermost directive.
 
+  // Find the loop statement.
+  Stmt *LoopStmt = nullptr;
+  collectLoopStmts(AStmt, {LoopStmt});
+
   // Determine the PreInit declarations.
-  SmallVector<Decl *, 4> PreInits;
-  assert(OriginalInits.size() == 1 &&
-         "Expecting a single-dimensional loop iteration space");
-  for (auto &P : OriginalInits[0]) {
-    if (auto *D = P.dyn_cast<Decl *>())
-      PreInits.push_back(D);
-    else if (auto *PI = dyn_cast_or_null<DeclStmt>(P.dyn_cast<Stmt *>()))
-      PreInits.append(PI->decl_begin(), PI->decl_end());
-  }
-  if (auto *PI = cast_or_null<DeclStmt>(LoopHelper.PreInits))
-    PreInits.append(PI->decl_begin(), PI->decl_end());
-  // Gather declarations for the data members used as counters.
-  for (Expr *CounterRef : LoopHelper.Counters) {
-    auto *CounterDecl = cast<DeclRefExpr>(CounterRef)->getDecl();
-    if (isa<OMPCapturedExprDecl>(CounterDecl))
-      PreInits.push_back(CounterDecl);
-  }
+  SmallVector<Stmt *, 4> PreInits;
+  addLoopPreInits(Context, LoopHelper, LoopStmt, OriginalInits[0], PreInits);
 
   auto *IterationVarRef = cast<DeclRefExpr>(LoopHelper.IterationVarRef);
   QualType IVTy = IterationVarRef->getType();
@@ -15604,6 +15682,8 @@ StmtResult SemaOpenMP::ActOnOpenMPUnrollDirective(ArrayRef<OMPClause *> Clauses,
   // Inner For statement.
   SmallVector<Stmt *> InnerBodyStmts;
   InnerBodyStmts.append(LoopHelper.Updates.begin(), LoopHelper.Updates.end());
+  if (auto *CXXRangeFor = dyn_cast<CXXForRangeStmt>(LoopStmt))
+    InnerBodyStmts.push_back(CXXRangeFor->getLoopVarStmt());
   InnerBodyStmts.push_back(Body);
   CompoundStmt *InnerBody =
       CompoundStmt::Create(getASTContext(), InnerBodyStmts, FPOptionsOverride(),
diff --git a/clang/test/OpenMP/tile_codegen.cpp b/clang/test/OpenMP/tile_codegen.cpp
index 93a3a14133ab5..5fd5609b844cc 100644
--- a/clang/test/OpenMP/tile_codegen.cpp
+++ b/clang/test/OpenMP/tile_codegen.cpp
@@ -1,10 +1,10 @@
-// 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 _
+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --include-generated-funcs --replace-value-regex "__omp_offloading_[0-9a-z]+_[0-9a-z]+" "reduction_size[.].+[.]" "pl_cond[.].+[.|,]" --prefix-filecheck-ir-name _ --version 4
 // Check code generation
-// RUN: %clang_cc1 -verify -triple x86_64-pc-linux-gnu -fclang-abi-compat=latest -fopenmp -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK1
+// RUN: %clang_cc1 -verify -triple x86_64-pc-linux-gnu -fclang-abi-compat=latest -std=c++20 -fopenmp -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 -fclang-abi-compat=latest -fopenmp -emit-pch -o %t %s
-// RUN: %clang_cc1 -verify -triple x86_64-pc-linux-gnu -fclang-abi-compat=latest -fopenmp -include-pch %t -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK2
+// RUN: %clang_cc1 -verify -triple x86_64-pc-linux-gnu -fclang-abi-compat=latest -std=c++20 -fopenmp -emit-pch -o %t %s
+// RUN: %clang_cc1 -verify -triple x86_64-pc-linux-gnu -fclang-abi-compat=latest -std=c++20 -fopenmp -include-pch %t -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK2
 // expected-no-diagnostics
 
 #ifndef HEADER
@@ -91,22 +91,38 @@ extern "C" void foo8(int a) {
 }
 
 
+typedef struct { double array[12]; } data_t;
+extern "C" void foo9(data_t data) {
+#pragma omp tile sizes(5)
+  for (double v : data.array)
+    body(v);
+}
+
+
+extern "C" void foo10(data_t data) {
+#pragma omp tile sizes(5)
+  for (double c = 42.0; double v : data.array)
+    body(c, v);
+}
+
+
 #endif /* HEADER */
-// CHECK1-LABEL: define {{[^@]+}}@body
-// CHECK1-SAME: (...) #[[ATTR0:[0-9]+]] {
+
+// CHECK1-LABEL: define dso_local void @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-LABEL: define internal void @__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-LABEL: define linkonce_odr void @_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
@@ -115,50 +131,52 @@ extern "C" void foo8(int a) {
 // 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-LABEL: define linkonce_odr void @_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:    [[I:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[I2:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[DOTFLOOR_0_IV_I:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[DOTTILE_0_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:    [[I2:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], ptr [[THIS1]], i32 0, i32 0
-// CHECK1-NEXT:    store ptr [[I2]], ptr [[I]], 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 [[DOTFLOOR_0_IV_I]], align 4
 // CHECK1-NEXT:    br label [[FOR_COND:%.*]]
 // CHECK1:       for.cond:
 // CHECK1-NEXT:    [[TMP0:%.*]] = load i32, ptr [[DOTFLOOR_0_IV_I]], align 4
 // CHECK1-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP0]], 4
-// CHECK1-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END11:%.*]]
+// CHECK1-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END12:%.*]]
 // CHECK1:       for.body:
 // CHECK1-NEXT:    [[TMP1:%.*]] = load i32, ptr [[DOTFLOOR_0_IV_I]], align 4
 // CHECK1-NEXT:    store i32 [[TMP1]], ptr [[DOTTILE_0_IV_I]], align 4
-// CHECK1-NEXT:    br label [[FOR_COND3:%.*]]
-// CHECK1:       for.cond3:
+// CHECK1-NEXT:    br label [[FOR_COND4:%.*]]
+// CHECK1:       for.cond4:
 // CHECK1-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTTILE_0_IV_I]], align 4
 // CHECK1-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTFLOOR_0_IV_I]], align 4
 // CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP3]], 5
-// CHECK1-NEXT:    [[CMP4:%.*]] = icmp slt i32 4, [[ADD]]
-// CHECK1-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
+// CHECK1-NEXT:    [[CMP5:%.*]] = icmp slt i32 4, [[ADD]]
+// CHECK1-NEXT:    br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK1:       cond.true:
 // CHECK1-NEXT:    br label [[COND_END:%.*]]
 // CHECK1:       cond.false:
 // CHECK1-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTFLOOR_0_IV_I]], align 4
-// CHECK1-NEXT:    [[ADD5:%.*]] = add nsw i32 [[TMP4]], 5
+// CHECK1-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP4]], 5
 // CHECK1-NEXT:    br label [[COND_END]]
 // CHECK1:       cond.end:
-// CHECK1-NEXT:    [[COND:%.*]] = phi i32 [ 4, [[COND_TRUE]] ], [ [[ADD5]], [[COND_FALSE]] ]
-// CHECK1-NEXT:    [[CMP6:%.*]] = icmp slt i32 [[TMP2]], [[COND]]
-// CHECK1-NEXT:    br i1 [[CMP6]], label [[FOR_BODY7:%.*]], label [[FOR_END:%.*]]
-// CHECK1:       for.body7:
+// CHECK1-NEXT:    [[COND:%.*]] = phi i32 [ 4, [[COND_TRUE]] ], [ [[ADD6]], [[COND_FALSE]] ]
+// CHECK1-NEXT:    [[CMP7:%.*]] = icmp slt i32 [[TMP2]], [[COND]]
+// CHECK1-NEXT:    br i1 [[CMP7]], label [[FOR_BODY8:%.*]], label [[FOR_END:%.*]]
+// CHECK1:       for.body8:
 // CHECK1-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTTILE_0_IV_I]], align 4
 // CHECK1-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP5]], 3
-// CHECK1-NEXT:    [[ADD8:%.*]] = add nsw i32 7, [[MUL]]
-// CHECK1-NEXT:    [[TMP6:%.*]] = load ptr, ptr [[I]], align 8
-// CHECK1-NEXT:    store i32 [[ADD8]], ptr [[TMP6]], align 4
-// CHECK1-NEXT:    [[TMP7:%.*]] = load ptr, ptr [[I]], align 8
+// CHECK1-NEXT:    [[ADD9:%.*]] = add nsw i32 7, [[MUL]]
+// CHECK1-NEXT:    [[TMP6:%.*]] = load ptr, ptr [[I2]], align 8
+// CHECK1-NEXT:    store i32 [[ADD9]], ptr [[TMP6]], align 4
+// CHECK1-NEXT:    [[TMP7:%.*]] = load ptr, ptr [[I2]], align 8
 // CHECK1-NEXT:    [[TMP8:%.*]] = load i32, ptr [[TMP7]], align 4
 // CHECK1-NEXT:    call void (...) @body(i32 noundef [[TMP8]])
 // CHECK1-NEXT:    br label [[FOR_INC:%.*]]
@@ -166,20 +184,20 @@ extern "C" void foo8(int a) {
 // CHECK1-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTTILE_0_IV_I]], align 4
 // CHECK1-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP9]], 1
 // CHECK1-NEXT:    store i32 [[INC]], ptr [[DOTTILE_0_IV_I]], align 4
-// CHECK1-NEXT:    br label [[FOR_COND3]], !llvm.loop [[LOOP3:![0-9]+]]
+// CHECK1-NEXT:    br label [[FOR_COND4]], !llvm.loop [[LOOP3:![0-9]+]]
 // CHECK1:       for.end:
-// CHECK1-NEXT:    br label [[FOR_INC9:%.*]]
-// CHECK1:       for.inc9:
+// CHECK1-NEXT:    br label [[FOR_INC10:%.*]]
+// CHECK1:       for.inc10:
 // CHECK1-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTFLOOR_0_IV_I]], align 4
-// CHECK1-NEXT:    [[ADD10:%.*]] = add nsw i32 [[TMP10]], 5
-// CHECK1-NEXT:    store i32 [[ADD10]], ptr [[DOTFLOOR_0_IV_I]], align 4
+// CHECK1-NEXT:    [[ADD11:%.*]] = add nsw i32 [[TMP10]], 5
+// CHECK1-NEXT:    store i32 [[ADD11]], ptr [[DOTFLOOR_0_IV_I]], align 4
 // CHECK1-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP5:![0-9]+]]
-// CHECK1:       for.end11:
+// CHECK1:       for.end12:
 // CHECK1-NEXT:    ret void
 //
 //
-// CHECK1-LABEL: define {{[^@]+}}@foo1
-// CHECK1-SAME: (i32 noundef [[START:%.*]], i32 noundef [[END:%.*]], i32 noundef [[STEP:%.*]]) #[[ATTR0]] {
+// CHECK1-LABEL: define dso_local void @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
@@ -195,81 +213,83 @@ extern "C" void foo8(int a) {
 // 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 [[DOTCAPTURE_EXPR_]], align 4
-// CHECK1-NEXT:    [[TMP1:%.*]] = load i32, ptr [[END_ADDR]], align 4
-// CHECK1-NEXT:    store i32 [[TMP1]], ptr [[DOTCAPTURE_EXPR_1]], align 4
-// CHECK1-NEXT:    [[TMP2:%.*]] = load i32, ptr [[STEP_ADDR]], align 4
-// CHECK1-NEXT:    store i32 [[TMP2]], ptr [[DOTNEW_STEP]], 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:    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:    [[TMP5:%.*]] = load i32, ptr [[DOTNEW_STEP]], align 4
-// CHECK1-NEXT:    [[ADD:%.*]] = add i32 [[SUB3]], [[TMP5]]
 // CHECK1-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTNEW_STEP]], align 4
-// CHECK1-NEXT:    [[DIV:%.*]] = udiv i32 [[ADD]], [[TMP6]]
+// 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 [[DOTFLOOR_0_IV_I]], align 4
 // CHECK1-NEXT:    br label [[FOR_COND:%.*]]
 // CHECK1:       for.cond:
-// CHECK1-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTFLOOR_0_IV_I]], align 4
-// CHECK1-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
-// CHECK1-NEXT:    [[ADD5:%.*]] = add i32 [[TMP8]], 1
-// CHECK1-NEXT:    [[CMP:%.*]] = icmp ult i32 [[TMP7]], [[ADD5]]
+// CHECK1-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTFLOOR_0_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_END17:%.*]]
 // CHECK1:       for.body:
-// CHECK1-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTFLOOR_0_IV_I]], align 4
-// CHECK1-NEXT:    store i32 [[TMP9]], ptr [[DOTTILE_0_IV_I]], align 4
+// CHECK1-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTFLOOR_0_IV_I]], align 4
+// CHECK1-NEXT:    store i32 [[TMP10]], ptr [[DOTTILE_0_IV_I]], align 4
 // CHECK1-NEXT:    br label [[FOR_COND6:%.*]]
 // CHECK1:       for.cond6:
-// CHECK1-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTTILE_0_IV_I]], align 4
-// CHECK1-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
-// CHECK1-NEXT:    [[ADD7:%.*]] = add i32 [[TMP11]], 1
-// CHECK1-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTFLOOR_0_IV_I]], align 4
-// CHECK1-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP12]], 5
+// CHECK1-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTTILE_0_IV_I]], align 4
+// CHECK1-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
+// CHECK1-NEXT:    [[ADD7:%.*]] = add i32 [[TMP12]], 1
+// CHECK1-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTFLOOR_0_IV_I]], align 4
+// CHECK1-NEXT:    [[ADD8:%.*]] = add i32 [[TMP13]], 5
 // CHECK1-NEXT:    [[CMP9:%.*]] = icmp ult i32 [[ADD7]], [[ADD8]]
 // CHECK1-NEXT:    br i1 [[CMP9]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK1:       cond.true:
-// CHECK1-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
-// CHECK1-NEXT:    [[ADD10:%.*]] = add i32 [[TMP13]], 1
+// CHECK1-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
+// CHECK1-NEXT:    [[ADD10:%.*]] = add i32 [[TMP14]], 1
 // CHECK1-NEXT:    br label [[COND_END:%.*]]
 // CHECK1:       cond.false:
-// CHECK1-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTFLOOR_0_IV_I]], align 4
-// CHECK1-NEXT:    [[ADD11:%.*]] = add nsw i32 [[TMP14]], 5
+// CHECK1-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTFLOOR_0_IV_I]], align 4
+// CHECK1-NEXT:    [[ADD11:%.*]] = add i32 [[TMP15]], 5
 // CHECK1-NEXT:    br label [[COND_END]]
 // CHECK1:       cond.end:
 // CHECK1-NEXT:    [[COND:%.*]] = phi i32 [ [[ADD10]], [[COND_TRUE]] ], [ [[ADD11]], [[COND_FALSE]] ]
-// CHECK1-NEXT:    [[CMP12:%.*]] = icmp ult i32 [[TMP10]], [[COND]]
+// CHECK1-NEXT:    [[CMP12:%.*]] = icmp ult i32 [[TMP11]], [[COND]]
 // CHECK1-NEXT:    br i1 [[CMP12]], label [[FOR_BODY13:%.*]], label [[FOR_END:%.*]]
 // CHECK1:       for.body13:
-// CHECK1-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
-// CHECK1-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTTILE_0_IV_I]], align 4
-// CHECK1-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTNEW_STEP]], align 4
-// CHECK1-NEXT:    [[MUL:%.*]] = mul i32 [[TMP16]], [[TMP17]]
-// CHECK1-NEXT:    [[ADD14:%.*]] = add i32 [[TMP15]], [[MUL]]
+// CHECK1-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK1-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTTILE_0_IV_I]], align 4
+// CHECK1-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTNEW_STEP]], align 4
+// CHECK1-NEXT:    [[MUL:%.*]] = mul i32 [[TMP17]], [[TMP18]]
+// CHECK1-NEXT:    [[ADD14:%.*]] = add i32 [[TMP16]], [[MUL]]
 // CHECK1-NEXT:    store i32 [[ADD14]], ptr [[I]], align 4
-// CHECK1-NEXT:    [[TMP18:%.*]] = load i32, ptr [[I]], align 4
-// CHECK1-NEXT:    call void (...) @body(i32 noundef [[TMP18]])
+// CHECK1-NEXT:    [[TMP19:%.*]] = load i32, ptr [[I]], align 4
+// CHECK1-NEXT:    call void (...) @body(i32 noundef [[TMP19]])
 // CHECK1-NEXT:    br label [[FOR_INC:%.*]]
 // CHECK1:       for.inc:
-// CHECK1-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTTILE_0_IV_I]], align 4
-// CHECK1-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP19]], 1
+// CHECK1-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTTILE_0_IV_I]], align 4
+// CHECK1-NEXT:    [[INC:%.*]] = add i32 [[TMP20]], 1
 // CHECK1-NEXT:    store i32 [[INC]], ptr [[DOTTILE_0_IV_I]], align 4
 // CHECK1-NEXT:    br label [[FOR_COND6]], !llvm.loop [[LOOP6:![0-9]+]]
 // CHECK1:       for.end:
 // CHECK1-NEXT:    br label [[FOR_INC15:%.*]]
 // CHECK1:       for.inc15:
-// CHECK1-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTFLOOR_0_IV_I]], align 4
-// CHECK1-NEXT:    [[ADD16:%.*]] = add nsw i32 [[TMP20]], 5
+// CHECK1-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTFLOOR_0_IV_I]], align 4
+// CHECK1-NEXT:    [[ADD16:%.*]] = add i32 [[TMP21]], 5
 // CHECK1-NEXT:    store i32 [[ADD16]], ptr [[DOTFLOOR_0_IV_I]], align 4
 // CHECK1-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP7:![0-9]+]]
 // CHECK1:       for.end17:
 // CHECK1-NEXT:    ret void
 //
 //
-// CHECK1-LABEL: define {{[^@]+}}@foo2
-// CHECK1-SAME: (i32 noundef [[START:%.*]], i32 noundef [[END:%.*]], i32 noundef [[STEP:%.*]]) #[[ATTR0]] {
+// CHECK1-LABEL: define dso_local void @foo2(
+// 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
@@ -381,8 +401,8 @@ extern "C" void foo8(int a) {
 // CHECK1-NEXT:    ret void
 //
 //
-// CHECK1-LABEL: define {{[^@]+}}@foo3
-// CHECK1-SAME: () #[[ATTR0]] {
+// CHECK1-LABEL: define dso_local void @foo3(
+// CHECK1-SAME: ) #[[ATTR0]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[TMP:%.*]] = alloca i32, align 4
@@ -523,8 +543,8 @@ extern "C" void foo8(int a) {
 // CHECK1-NEXT:    ret void
 //
 //
-// CHECK1-LABEL: define {{[^@]+}}@foo4
-// CHECK1-SAME: () #[[ATTR0]] {
+// CHECK1-LABEL: define dso_local void @foo4(
+// CHECK1-SAME: ) #[[ATTR0]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[TMP:%.*]] = alloca i32, align 4
@@ -676,8 +696,8 @@ extern "C" void foo8(int a) {
 // CHECK1-NEXT:    ret void
 //
 //
-// CHECK1-LABEL: define {{[^@]+}}@foo5
-// CHECK1-SAME: () #[[ATTR0]] {
+// CHECK1-LABEL: define dso_local void @foo5(
+// CHECK1-SAME: ) #[[ATTR0]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTOMP_IV:%.*]] = alloca i64, align 8
 // CHECK1-NEXT:    [[TMP:%.*]] = alloca i32, align 4
@@ -885,15 +905,15 @@ extern "C" void foo8(int a) {
 // CHECK1-NEXT:    ret void
 //
 //
-// CHECK1-LABEL: define {{[^@]+}}@foo6
-// CHECK1-SAME: () #[[ATTR0]] {
+// CHECK1-LABEL: define dso_local void @foo6(
+// CHECK1-SAME: ) #[[ATTR0]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB2]], i32 0, ptr @foo6.omp_outlined)
 // CHECK1-NEXT:    ret void
 //
 //
-// CHECK1-LABEL: define {{[^@]+}}@foo6.omp_outlined
-// CHECK1-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]]) #[[ATTR4:[0-9]+]] {
+// CHECK1-LABEL: define internal void @foo6.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
@@ -988,15 +1008,15 @@ extern "C" void foo8(int a) {
 // CHECK1-NEXT:    ret void
 //
 //
-// CHECK1-LABEL: define {{[^@]+}}@tfoo7
-// CHECK1-SAME: () #[[ATTR0]] {
+// CHECK1-LABEL: define dso_local void @tfoo7(
+// CHECK1-SAME: ) #[[ATTR0]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    call void @_Z4foo7IiTnT_Li3ETnS0_Li5EEvS0_S0_(i32 noundef 0, i32 noundef 42)
 // CHECK1-NEXT:    ret void
 //
 //
-// CHECK1-LABEL: define {{[^@]+}}@_Z4foo7IiTnT_Li3ETnS0_Li5EEvS0_S0_
-// CHECK1-SAME: (i32 noundef [[START:%.*]], i32 noundef [[END:%.*]]) #[[ATTR0]] comdat {
+// CHECK1-LABEL: define linkonce_odr void @_Z4foo7IiTnT_Li3ETnS0_Li5EEvS0_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
@@ -1039,7 +1059,7 @@ extern "C" void foo8(int a) {
 // CHECK1-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK1-NEXT:    [[ADD7:%.*]] = add i32 [[TMP9]], 1
 // CHECK1-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTFLOOR_0_IV_I]], align 4
-// CHECK1-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP10]], 5
+// CHECK1-NEXT:    [[ADD8:%.*]] = add i32 [[TMP10]], 5
 // CHECK1-NEXT:    [[CMP9:%.*]] = icmp ult i32 [[ADD7]], [[ADD8]]
 // CHECK1-NEXT:    br i1 [[CMP9]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK1:       cond.true:
@@ -1048,7 +1068,7 @@ extern "C" void foo8(int a) {
 // CHECK1-NEXT:    br label [[COND_END:%.*]]
 // CHECK1:       cond.false:
 // CHECK1-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTFLOOR_0_IV_I]], align 4
-// CHECK1-NEXT:    [[ADD11:%.*]] = add nsw i32 [[TMP12]], 5
+// CHECK1-NEXT:    [[ADD11:%.*]] = add i32 [[TMP12]], 5
 // CHECK1-NEXT:    br label [[COND_END]]
 // CHECK1:       cond.end:
 // CHECK1-NEXT:    [[COND:%.*]] = phi i32 [ [[ADD10]], [[COND_TRUE]] ], [ [[ADD11]], [[COND_FALSE]] ]
@@ -1065,22 +1085,22 @@ extern "C" void foo8(int a) {
 // CHECK1-NEXT:    br label [[FOR_INC:%.*]]
 // CHECK1:       for.inc:
 // CHECK1-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTTILE_0_IV_I]], align 4
-// CHECK1-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP16]], 1
+// CHECK1-NEXT:    [[INC:%.*]] = add i32 [[TMP16]], 1
 // CHECK1-NEXT:    store i32 [[INC]], ptr [[DOTTILE_0_IV_I]], align 4
 // CHECK1-NEXT:    br label [[FOR_COND6]], !llvm.loop [[LOOP21:![0-9]+]]
 // CHECK1:       for.end:
 // CHECK1-NEXT:    br label [[FOR_INC15:%.*]]
 // CHECK1:       for.inc15:
 // CHECK1-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTFLOOR_0_IV_I]], align 4
-// CHECK1-NEXT:    [[ADD16:%.*]] = add nsw i32 [[TMP17]], 5
+// CHECK1-NEXT:    [[ADD16:%.*]] = add i32 [[TMP17]], 5
 // CHECK1-NEXT:    store i32 [[ADD16]], ptr [[DOTFLOOR_0_IV_I]], align 4
 // CHECK1-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP22:![0-9]+]]
 // CHECK1:       for.end17:
 // CHECK1-NEXT:    ret void
 //
 //
-// CHECK1-LABEL: define {{[^@]+}}@foo8
-// CHECK1-SAME: (i32 noundef [[A:%.*]]) #[[ATTR0]] {
+// CHECK1-LABEL: define dso_local void @foo8(
+// CHECK1-SAME: i32 noundef [[A:%.*]]) #[[ATTR0]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[I:%.*]] = alloca i32, align 4
@@ -1168,22 +1188,219 @@ extern "C" void foo8(int a) {
 // CHECK1-NEXT:    ret void
 //
 //
-// CHECK1-LABEL: define {{[^@]+}}@_GLOBAL__sub_I_tile_codegen.cpp
-// CHECK1-SAME: () #[[ATTR1]] section ".text.startup" {
+// CHECK1-LABEL: define dso_local void @foo9(
+// CHECK1-SAME: ptr noundef byval([[STRUCT_DATA_T:%.*]]) align 8 [[DATA:%.*]]) #[[ATTR0]] {
+// CHECK1-NEXT:  entry:
+// 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:    [[DOTFLOOR_0_IV___BEGIN2:%.*]] = alloca i64, align 8
+// CHECK1-NEXT:    [[DOTTILE_0_IV___BEGIN2:%.*]] = alloca i64, align 8
+// CHECK1-NEXT:    [[V:%.*]] = alloca double, align 8
+// CHECK1-NEXT:    [[ARRAY:%.*]] = getelementptr inbounds [[STRUCT_DATA_T]], ptr [[DATA]], i32 0, i32 0
+// CHECK1-NEXT:    store ptr [[ARRAY]], ptr [[__RANGE2]], align 8
+// CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[__RANGE2]], align 8
+// CHECK1-NEXT:    [[ARRAYDECAY:%.*]] = getelementptr inbounds [12 x double], ptr [[TMP0]], i64 0, i64 0
+// CHECK1-NEXT:    [[ADD_PTR:%.*]] = getelementptr inbounds double, ptr [[ARRAYDECAY]], i64 12
+// CHECK1-NEXT:    store ptr [[ADD_PTR]], ptr [[__END2]], align 8
+// CHECK1-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[__RANGE2]], align 8
+// CHECK1-NEXT:    [[ARRAYDECAY1:%.*]] = getelementptr inbounds [12 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 [12 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 [[DOTFLOOR_0_IV___BEGIN2]], align 8
+// CHECK1-NEXT:    br label [[FOR_COND:%.*]]
+// CHECK1:       for.cond:
+// CHECK1-NEXT:    [[TMP6:%.*]] = load i64, ptr [[DOTFLOOR_0_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_END18:%.*]]
+// CHECK1:       for.body:
+// CHECK1-NEXT:    [[TMP8:%.*]] = load i64, ptr [[DOTFLOOR_0_IV___BEGIN2]], align 8
+// CHECK1-NEXT:    store i64 [[TMP8]], ptr [[DOTTILE_0_IV___BEGIN2]], align 8
+// CHECK1-NEXT:    br label [[FOR_COND7:%.*]]
+// CHECK1:       for.cond7:
+// CHECK1-NEXT:    [[TMP9:%.*]] = load i64, ptr [[DOTTILE_0_IV___BEGIN2]], align 8
+// CHECK1-NEXT:    [[TMP10:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR_4]], align 8
+// CHECK1-NEXT:    [[ADD8:%.*]] = add nsw i64 [[TMP10]], 1
+// CHECK1-NEXT:    [[TMP11:%.*]] = load i64, ptr [[DOTFLOOR_0_IV___BEGIN2]], align 8
+// CHECK1-NEXT:    [[ADD9:%.*]] = add nsw i64 [[TMP11]], 5
+// CHECK1-NEXT:    [[CMP10:%.*]] = icmp slt i64 [[ADD8]], [[ADD9]]
+// CHECK1-NEXT:    br i1 [[CMP10]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
+// CHECK1:       cond.true:
+// CHECK1-NEXT:    [[TMP12:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR_4]], align 8
+// CHECK1-NEXT:    [[ADD11:%.*]] = add nsw i64 [[TMP12]], 1
+// CHECK1-NEXT:    br label [[COND_END:%.*]]
+// CHECK1:       cond.false:
+// CHECK1-NEXT:    [[TMP13:%.*]] = load i64, ptr [[DOTFLOOR_0_IV___BEGIN2]], align 8
+// CHECK1-NEXT:    [[ADD12:%.*]] = add nsw i64 [[TMP13]], 5
+// CHECK1-NEXT:    br label [[COND_END]]
+// CHECK1:       cond.end:
+// CHECK1-NEXT:    [[COND:%.*]] = phi i64 [ [[ADD11]], [[COND_TRUE]] ], [ [[ADD12]], [[COND_FALSE]] ]
+// CHECK1-NEXT:    [[CMP13:%.*]] = icmp slt i64 [[TMP9]], [[COND]]
+// CHECK1-NEXT:    br i1 [[CMP13]], label [[FOR_BODY14:%.*]], label [[FOR_END:%.*]]
+// CHECK1:       for.body14:
+// CHECK1-NEXT:    [[TMP14:%.*]] = load ptr, ptr [[DOTCAPTURE_EXPR_]], align 8
+// CHECK1-NEXT:    [[TMP15:%.*]] = load i64, ptr [[DOTTILE_0_IV___BEGIN2]], align 8
+// CHECK1-NEXT:    [[MUL:%.*]] = mul nsw i64 [[TMP15]], 1
+// CHECK1-NEXT:    [[ADD_PTR15:%.*]] = getelementptr inbounds double, ptr [[TMP14]], i64 [[MUL]]
+// CHECK1-NEXT:    store ptr [[ADD_PTR15]], ptr [[__BEGIN2]], align 8
+// CHECK1-NEXT:    [[TMP16:%.*]] = load ptr, ptr [[__BEGIN2]], align 8
+// CHECK1-NEXT:    [[TMP17:%.*]] = load double, ptr [[TMP16]], align 8
+// CHECK1-NEXT:    store double [[TMP17]], ptr [[V]], align 8
+// CHECK1-NEXT:    [[TMP18:%.*]] = load double, ptr [[V]], align 8
+// CHECK1-NEXT:    call void (...) @body(double noundef [[TMP18]])
+// CHECK1-NEXT:    br label [[FOR_INC:%.*]]
+// CHECK1:       for.inc:
+// CHECK1-NEXT:    [[TMP19:%.*]] = load i64, ptr [[DOTTILE_0_IV___BEGIN2]], align 8
+// CHECK1-NEXT:    [[INC:%.*]] = add nsw i64 [[TMP19]], 1
+// CHECK1-NEXT:    store i64 [[INC]], ptr [[DOTTILE_0_IV___BEGIN2]], align 8
+// CHECK1-NEXT:    br label [[FOR_COND7]], !llvm.loop [[LOOP25:![0-9]+]]
+// CHECK1:       for.end:
+// CHECK1-NEXT:    br label [[FOR_INC16:%.*]]
+// CHECK1:       for.inc16:
+// CHECK1-NEXT:    [[TMP20:%.*]] = load i64, ptr [[DOTFLOOR_0_IV___BEGIN2]], align 8
+// CHECK1-NEXT:    [[ADD17:%.*]] = add nsw i64 [[TMP20]], 5
+// CHECK1-NEXT:    store i64 [[ADD17]], ptr [[DOTFLOOR_0_IV___BEGIN2]], align 8
+// CHECK1-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP26:![0-9]+]]
+// CHECK1:       for.end18:
+// CHECK1-NEXT:    ret void
+//
+//
+// CHECK1-LABEL: define dso_local void @foo10(
+// CHECK1-SAME: ptr noundef byval([[STRUCT_DATA_T:%.*]]) align 8 [[DATA:%.*]]) #[[ATTR0]] {
+// CHECK1-NEXT:  entry:
+// CHECK1-NEXT:    [[C:%.*]] = alloca double, align 8
+// 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:    [[DOTFLOOR_0_IV___BEGIN2:%.*]] = alloca i64, align 8
+// CHECK1-NEXT:    [[DOTTILE_0_IV___BEGIN2:%.*]] = alloca i64, align 8
+// CHECK1-NEXT:    [[V:%.*]] = alloca double, align 8
+// CHECK1-NEXT:    store double 4.200000e+01, ptr [[C]], align 8
+// CHECK1-NEXT:    [[ARRAY:%.*]] = getelementptr inbounds [[STRUCT_DATA_T]], ptr [[DATA]], i32 0, i32 0
+// CHECK1-NEXT:    store ptr [[ARRAY]], ptr [[__RANGE2]], align 8
+// CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[__RANGE2]], align 8
+// CHECK1-NEXT:    [[ARRAYDECAY:%.*]] = getelementptr inbounds [12 x double], ptr [[TMP0]], i64 0, i64 0
+// CHECK1-NEXT:    [[ADD_PTR:%.*]] = getelementptr inbounds double, ptr [[ARRAYDECAY]], i64 12
+// CHECK1-NEXT:    store ptr [[ADD_PTR]], ptr [[__END2]], align 8
+// CHECK1-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[__RANGE2]], align 8
+// CHECK1-NEXT:    [[ARRAYDECAY1:%.*]] = getelementptr inbounds [12 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 [12 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 [[DOTFLOOR_0_IV___BEGIN2]], align 8
+// CHECK1-NEXT:    br label [[FOR_COND:%.*]]
+// CHECK1:       for.cond:
+// CHECK1-NEXT:    [[TMP6:%.*]] = load i64, ptr [[DOTFLOOR_0_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_END18:%.*]]
+// CHECK1:       for.body:
+// CHECK1-NEXT:    [[TMP8:%.*]] = load i64, ptr [[DOTFLOOR_0_IV___BEGIN2]], align 8
+// CHECK1-NEXT:    store i64 [[TMP8]], ptr [[DOTTILE_0_IV___BEGIN2]], align 8
+// CHECK1-NEXT:    br label [[FOR_COND7:%.*]]
+// CHECK1:       for.cond7:
+// CHECK1-NEXT:    [[TMP9:%.*]] = load i64, ptr [[DOTTILE_0_IV___BEGIN2]], align 8
+// CHECK1-NEXT:    [[TMP10:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR_4]], align 8
+// CHECK1-NEXT:    [[ADD8:%.*]] = add nsw i64 [[TMP10]], 1
+// CHECK1-NEXT:    [[TMP11:%.*]] = load i64, ptr [[DOTFLOOR_0_IV___BEGIN2]], align 8
+// CHECK1-NEXT:    [[ADD9:%.*]] = add nsw i64 [[TMP11]], 5
+// CHECK1-NEXT:    [[CMP10:%.*]] = icmp slt i64 [[ADD8]], [[ADD9]]
+// CHECK1-NEXT:    br i1 [[CMP10]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
+// CHECK1:       cond.true:
+// CHECK1-NEXT:    [[TMP12:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR_4]], align 8
+// CHECK1-NEXT:    [[ADD11:%.*]] = add nsw i64 [[TMP12]], 1
+// CHECK1-NEXT:    br label [[COND_END:%.*]]
+// CHECK1:       cond.false:
+// CHECK1-NEXT:    [[TMP13:%.*]] = load i64, ptr [[DOTFLOOR_0_IV___BEGIN2]], align 8
+// CHECK1-NEXT:    [[ADD12:%.*]] = add nsw i64 [[TMP13]], 5
+// CHECK1-NEXT:    br label [[COND_END]]
+// CHECK1:       cond.end:
+// CHECK1-NEXT:    [[COND:%.*]] = phi i64 [ [[ADD11]], [[COND_TRUE]] ], [ [[ADD12]], [[COND_FALSE]] ]
+// CHECK1-NEXT:    [[CMP13:%.*]] = icmp slt i64 [[TMP9]], [[COND]]
+// CHECK1-NEXT:    br i1 [[CMP13]], label [[FOR_BODY14:%.*]], label [[FOR_END:%.*]]
+// CHECK1:       for.body14:
+// CHECK1-NEXT:    [[TMP14:%.*]] = load ptr, ptr [[DOTCAPTURE_EXPR_]], align 8
+// CHECK1-NEXT:    [[TMP15:%.*]] = load i64, ptr [[DOTTILE_0_IV___BEGIN2]], align 8
+// CHECK1-NEXT:    [[MUL:%.*]] = mul nsw i64 [[TMP15]], 1
+// CHECK1-NEXT:    [[ADD_PTR15:%.*]] = getelementptr inbounds double, ptr [[TMP14]], i64 [[MUL]]
+// CHECK1-NEXT:    store ptr [[ADD_PTR15]], ptr [[__BEGIN2]], align 8
+// CHECK1-NEXT:    [[TMP16:%.*]] = load ptr, ptr [[__BEGIN2]], align 8
+// CHECK1-NEXT:    [[TMP17:%.*]] = load double, ptr [[TMP16]], align 8
+// CHECK1-NEXT:    store double [[TMP17]], ptr [[V]], align 8
+// CHECK1-NEXT:    [[TMP18:%.*]] = load double, ptr [[C]], align 8
+// CHECK1-NEXT:    [[TMP19:%.*]] = load double, ptr [[V]], align 8
+// CHECK1-NEXT:    call void (...) @body(double noundef [[TMP18]], double noundef [[TMP19]])
+// CHECK1-NEXT:    br label [[FOR_INC:%.*]]
+// CHECK1:       for.inc:
+// CHECK1-NEXT:    [[TMP20:%.*]] = load i64, ptr [[DOTTILE_0_IV___BEGIN2]], align 8
+// CHECK1-NEXT:    [[INC:%.*]] = add nsw i64 [[TMP20]], 1
+// CHECK1-NEXT:    store i64 [[INC]], ptr [[DOTTILE_0_IV___BEGIN2]], align 8
+// CHECK1-NEXT:    br label [[FOR_COND7]], !llvm.loop [[LOOP27:![0-9]+]]
+// CHECK1:       for.end:
+// CHECK1-NEXT:    br label [[FOR_INC16:%.*]]
+// CHECK1:       for.inc16:
+// CHECK1-NEXT:    [[TMP21:%.*]] = load i64, ptr [[DOTFLOOR_0_IV___BEGIN2]], align 8
+// CHECK1-NEXT:    [[ADD17:%.*]] = add nsw i64 [[TMP21]], 5
+// CHECK1-NEXT:    store i64 [[ADD17]], ptr [[DOTFLOOR_0_IV___BEGIN2]], align 8
+// CHECK1-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP28:![0-9]+]]
+// CHECK1:       for.end18:
+// CHECK1-NEXT:    ret void
+//
+//
+// CHECK1-LABEL: define internal void @_GLOBAL__sub_I_tile_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-LABEL: define internal void @__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-LABEL: define linkonce_odr void @_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
@@ -1192,50 +1409,52 @@ extern "C" void foo8(int a) {
 // 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-LABEL: define linkonce_odr void @_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:    [[I:%.*]] = alloca ptr, align 8
+// CHECK2-NEXT:    [[I2:%.*]] = alloca ptr, align 8
 // CHECK2-NEXT:    [[DOTFLOOR_0_IV_I:%.*]] = alloca i32, align 4
 // CHECK2-NEXT:    [[DOTTILE_0_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:    [[I2:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], ptr [[THIS1]], i32 0, i32 0
-// CHECK2-NEXT:    store ptr [[I2]], ptr [[I]], 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 [[DOTFLOOR_0_IV_I]], align 4
 // CHECK2-NEXT:    br label [[FOR_COND:%.*]]
 // CHECK2:       for.cond:
 // CHECK2-NEXT:    [[TMP0:%.*]] = load i32, ptr [[DOTFLOOR_0_IV_I]], align 4
 // CHECK2-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP0]], 4
-// CHECK2-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END11:%.*]]
+// CHECK2-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END12:%.*]]
 // CHECK2:       for.body:
 // CHECK2-NEXT:    [[TMP1:%.*]] = load i32, ptr [[DOTFLOOR_0_IV_I]], align 4
 // CHECK2-NEXT:    store i32 [[TMP1]], ptr [[DOTTILE_0_IV_I]], align 4
-// CHECK2-NEXT:    br label [[FOR_COND3:%.*]]
-// CHECK2:       for.cond3:
+// CHECK2-NEXT:    br label [[FOR_COND4:%.*]]
+// CHECK2:       for.cond4:
 // CHECK2-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTTILE_0_IV_I]], align 4
 // CHECK2-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTFLOOR_0_IV_I]], align 4
 // CHECK2-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP3]], 5
-// CHECK2-NEXT:    [[CMP4:%.*]] = icmp slt i32 4, [[ADD]]
-// CHECK2-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
+// CHECK2-NEXT:    [[CMP5:%.*]] = icmp slt i32 4, [[ADD]]
+// CHECK2-NEXT:    br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK2:       cond.true:
 // CHECK2-NEXT:    br label [[COND_END:%.*]]
 // CHECK2:       cond.false:
 // CHECK2-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTFLOOR_0_IV_I]], align 4
-// CHECK2-NEXT:    [[ADD5:%.*]] = add nsw i32 [[TMP4]], 5
+// CHECK2-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP4]], 5
 // CHECK2-NEXT:    br label [[COND_END]]
 // CHECK2:       cond.end:
-// CHECK2-NEXT:    [[COND:%.*]] = phi i32 [ 4, [[COND_TRUE]] ], [ [[ADD5]], [[COND_FALSE]] ]
-// CHECK2-NEXT:    [[CMP6:%.*]] = icmp slt i32 [[TMP2]], [[COND]]
-// CHECK2-NEXT:    br i1 [[CMP6]], label [[FOR_BODY7:%.*]], label [[FOR_END:%.*]]
-// CHECK2:       for.body7:
+// CHECK2-NEXT:    [[COND:%.*]] = phi i32 [ 4, [[COND_TRUE]] ], [ [[ADD6]], [[COND_FALSE]] ]
+// CHECK2-NEXT:    [[CMP7:%.*]] = icmp slt i32 [[TMP2]], [[COND]]
+// CHECK2-NEXT:    br i1 [[CMP7]], label [[FOR_BODY8:%.*]], label [[FOR_END:%.*]]
+// CHECK2:       for.body8:
 // CHECK2-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTTILE_0_IV_I]], align 4
 // CHECK2-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP5]], 3
-// CHECK2-NEXT:    [[ADD8:%.*]] = add nsw i32 7, [[MUL]]
-// CHECK2-NEXT:    [[TMP6:%.*]] = load ptr, ptr [[I]], align 8
-// CHECK2-NEXT:    store i32 [[ADD8]], ptr [[TMP6]], align 4
-// CHECK2-NEXT:    [[TMP7:%.*]] = load ptr, ptr [[I]], align 8
+// CHECK2-NEXT:    [[ADD9:%.*]] = add nsw i32 7, [[MUL]]
+// CHECK2-NEXT:    [[TMP6:%.*]] = load ptr, ptr [[I2]], align 8
+// CHECK2-NEXT:    store i32 [[ADD9]], ptr [[TMP6]], align 4
+// CHECK2-NEXT:    [[TMP7:%.*]] = load ptr, ptr [[I2]], align 8
 // CHECK2-NEXT:    [[TMP8:%.*]] = load i32, ptr [[TMP7]], align 4
 // CHECK2-NEXT:    call void (...) @body(i32 noundef [[TMP8]])
 // CHECK2-NEXT:    br label [[FOR_INC:%.*]]
@@ -1243,26 +1462,26 @@ extern "C" void foo8(int a) {
 // CHECK2-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTTILE_0_IV_I]], align 4
 // CHECK2-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP9]], 1
 // CHECK2-NEXT:    store i32 [[INC]], ptr [[DOTTILE_0_IV_I]], align 4
-// CHECK2-NEXT:    br label [[FOR_COND3]], !llvm.loop [[LOOP3:![0-9]+]]
+// CHECK2-NEXT:    br label [[FOR_COND4]], !llvm.loop [[LOOP3:![0-9]+]]
 // CHECK2:       for.end:
-// CHECK2-NEXT:    br label [[FOR_INC9:%.*]]
-// CHECK2:       for.inc9:
+// CHECK2-NEXT:    br label [[FOR_INC10:%.*]]
+// CHECK2:       for.inc10:
 // CHECK2-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTFLOOR_0_IV_I]], align 4
-// CHECK2-NEXT:    [[ADD10:%.*]] = add nsw i32 [[TMP10]], 5
-// CHECK2-NEXT:    store i32 [[ADD10]], ptr [[DOTFLOOR_0_IV_I]], align 4
+// CHECK2-NEXT:    [[ADD11:%.*]] = add nsw i32 [[TMP10]], 5
+// CHECK2-NEXT:    store i32 [[ADD11]], ptr [[DOTFLOOR_0_IV_I]], align 4
 // CHECK2-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP5:![0-9]+]]
-// CHECK2:       for.end11:
+// CHECK2:       for.end12:
 // CHECK2-NEXT:    ret void
 //
 //
-// CHECK2-LABEL: define {{[^@]+}}@body
-// CHECK2-SAME: (...) #[[ATTR1]] {
+// CHECK2-LABEL: define dso_local void @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-LABEL: define dso_local void @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
@@ -1278,81 +1497,183 @@ extern "C" void foo8(int a) {
 // 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 [[DOTCAPTURE_EXPR_]], align 4
-// CHECK2-NEXT:    [[TMP1:%.*]] = load i32, ptr [[END_ADDR]], align 4
-// CHECK2-NEXT:    store i32 [[TMP1]], ptr [[DOTCAPTURE_EXPR_1]], align 4
-// CHECK2-NEXT:    [[TMP2:%.*]] = load i32, ptr [[STEP_ADDR]], align 4
-// CHECK2-NEXT:    store i32 [[TMP2]], ptr [[DOTNEW_STEP]], 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:    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:    [[TMP5:%.*]] = load i32, ptr [[DOTNEW_STEP]], align 4
-// CHECK2-NEXT:    [[ADD:%.*]] = add i32 [[SUB3]], [[TMP5]]
 // CHECK2-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTNEW_STEP]], align 4
-// CHECK2-NEXT:    [[DIV:%.*]] = udiv i32 [[ADD]], [[TMP6]]
+// 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 [[DOTFLOOR_0_IV_I]], align 4
 // CHECK2-NEXT:    br label [[FOR_COND:%.*]]
 // CHECK2:       for.cond:
-// CHECK2-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTFLOOR_0_IV_I]], align 4
-// CHECK2-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
-// CHECK2-NEXT:    [[ADD5:%.*]] = add i32 [[TMP8]], 1
-// CHECK2-NEXT:    [[CMP:%.*]] = icmp ult i32 [[TMP7]], [[ADD5]]
+// CHECK2-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTFLOOR_0_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_END17:%.*]]
 // CHECK2:       for.body:
-// CHECK2-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTFLOOR_0_IV_I]], align 4
-// CHECK2-NEXT:    store i32 [[TMP9]], ptr [[DOTTILE_0_IV_I]], align 4
+// CHECK2-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTFLOOR_0_IV_I]], align 4
+// CHECK2-NEXT:    store i32 [[TMP10]], ptr [[DOTTILE_0_IV_I]], align 4
 // CHECK2-NEXT:    br label [[FOR_COND6:%.*]]
 // CHECK2:       for.cond6:
-// CHECK2-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTTILE_0_IV_I]], align 4
-// CHECK2-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
-// CHECK2-NEXT:    [[ADD7:%.*]] = add i32 [[TMP11]], 1
-// CHECK2-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTFLOOR_0_IV_I]], align 4
-// CHECK2-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP12]], 5
+// CHECK2-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTTILE_0_IV_I]], align 4
+// CHECK2-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
+// CHECK2-NEXT:    [[ADD7:%.*]] = add i32 [[TMP12]], 1
+// CHECK2-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTFLOOR_0_IV_I]], align 4
+// CHECK2-NEXT:    [[ADD8:%.*]] = add i32 [[TMP13]], 5
 // CHECK2-NEXT:    [[CMP9:%.*]] = icmp ult i32 [[ADD7]], [[ADD8]]
 // CHECK2-NEXT:    br i1 [[CMP9]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK2:       cond.true:
-// CHECK2-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
-// CHECK2-NEXT:    [[ADD10:%.*]] = add i32 [[TMP13]], 1
+// CHECK2-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
+// CHECK2-NEXT:    [[ADD10:%.*]] = add i32 [[TMP14]], 1
 // CHECK2-NEXT:    br label [[COND_END:%.*]]
 // CHECK2:       cond.false:
-// CHECK2-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTFLOOR_0_IV_I]], align 4
-// CHECK2-NEXT:    [[ADD11:%.*]] = add nsw i32 [[TMP14]], 5
+// CHECK2-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTFLOOR_0_IV_I]], align 4
+// CHECK2-NEXT:    [[ADD11:%.*]] = add i32 [[TMP15]], 5
 // CHECK2-NEXT:    br label [[COND_END]]
 // CHECK2:       cond.end:
 // CHECK2-NEXT:    [[COND:%.*]] = phi i32 [ [[ADD10]], [[COND_TRUE]] ], [ [[ADD11]], [[COND_FALSE]] ]
-// CHECK2-NEXT:    [[CMP12:%.*]] = icmp ult i32 [[TMP10]], [[COND]]
+// CHECK2-NEXT:    [[CMP12:%.*]] = icmp ult i32 [[TMP11]], [[COND]]
 // CHECK2-NEXT:    br i1 [[CMP12]], label [[FOR_BODY13:%.*]], label [[FOR_END:%.*]]
 // CHECK2:       for.body13:
-// CHECK2-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
-// CHECK2-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTTILE_0_IV_I]], align 4
-// CHECK2-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTNEW_STEP]], align 4
-// CHECK2-NEXT:    [[MUL:%.*]] = mul i32 [[TMP16]], [[TMP17]]
-// CHECK2-NEXT:    [[ADD14:%.*]] = add i32 [[TMP15]], [[MUL]]
+// CHECK2-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
+// CHECK2-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTTILE_0_IV_I]], align 4
+// CHECK2-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTNEW_STEP]], align 4
+// CHECK2-NEXT:    [[MUL:%.*]] = mul i32 [[TMP17]], [[TMP18]]
+// CHECK2-NEXT:    [[ADD14:%.*]] = add i32 [[TMP16]], [[MUL]]
 // CHECK2-NEXT:    store i32 [[ADD14]], ptr [[I]], align 4
-// CHECK2-NEXT:    [[TMP18:%.*]] = load i32, ptr [[I]], align 4
-// CHECK2-NEXT:    call void (...) @body(i32 noundef [[TMP18]])
+// CHECK2-NEXT:    [[TMP19:%.*]] = load i32, ptr [[I]], align 4
+// CHECK2-NEXT:    call void (...) @body(i32 noundef [[TMP19]])
 // CHECK2-NEXT:    br label [[FOR_INC:%.*]]
 // CHECK2:       for.inc:
-// CHECK2-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTTILE_0_IV_I]], align 4
-// CHECK2-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP19]], 1
+// CHECK2-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTTILE_0_IV_I]], align 4
+// CHECK2-NEXT:    [[INC:%.*]] = add i32 [[TMP20]], 1
 // CHECK2-NEXT:    store i32 [[INC]], ptr [[DOTTILE_0_IV_I]], align 4
 // CHECK2-NEXT:    br label [[FOR_COND6]], !llvm.loop [[LOOP6:![0-9]+]]
 // CHECK2:       for.end:
 // CHECK2-NEXT:    br label [[FOR_INC15:%.*]]
 // CHECK2:       for.inc15:
-// CHECK2-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTFLOOR_0_IV_I]], align 4
-// CHECK2-NEXT:    [[ADD16:%.*]] = add nsw i32 [[TMP20]], 5
+// CHECK2-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTFLOOR_0_IV_I]], align 4
+// CHECK2-NEXT:    [[ADD16:%.*]] = add i32 [[TMP21]], 5
 // CHECK2-NEXT:    store i32 [[ADD16]], ptr [[DOTFLOOR_0_IV_I]], align 4
 // CHECK2-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP7:![0-9]+]]
 // CHECK2:       for.end17:
 // CHECK2-NEXT:    ret void
 //
 //
-// CHECK2-LABEL: define {{[^@]+}}@foo2
-// CHECK2-SAME: (i32 noundef [[START:%.*]], i32 noundef [[END:%.*]], i32 noundef [[STEP:%.*]]) #[[ATTR1]] {
+// CHECK2-LABEL: define dso_local void @foo10(
+// CHECK2-SAME: ptr noundef byval([[STRUCT_DATA_T:%.*]]) align 8 [[DATA:%.*]]) #[[ATTR1]] {
+// CHECK2-NEXT:  entry:
+// CHECK2-NEXT:    [[C:%.*]] = alloca double, align 8
+// 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:    [[DOTFLOOR_0_IV___BEGIN2:%.*]] = alloca i64, align 8
+// CHECK2-NEXT:    [[DOTTILE_0_IV___BEGIN2:%.*]] = alloca i64, align 8
+// CHECK2-NEXT:    [[V:%.*]] = alloca double, align 8
+// CHECK2-NEXT:    store double 4.200000e+01, ptr [[C]], align 8
+// CHECK2-NEXT:    [[ARRAY:%.*]] = getelementptr inbounds [[STRUCT_DATA_T]], ptr [[DATA]], i32 0, i32 0
+// CHECK2-NEXT:    store ptr [[ARRAY]], ptr [[__RANGE2]], align 8
+// CHECK2-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[__RANGE2]], align 8
+// CHECK2-NEXT:    [[ARRAYDECAY:%.*]] = getelementptr inbounds [12 x double], ptr [[TMP0]], i64 0, i64 0
+// CHECK2-NEXT:    [[ADD_PTR:%.*]] = getelementptr inbounds double, ptr [[ARRAYDECAY]], i64 12
+// CHECK2-NEXT:    store ptr [[ADD_PTR]], ptr [[__END2]], align 8
+// CHECK2-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[__RANGE2]], align 8
+// CHECK2-NEXT:    [[ARRAYDECAY1:%.*]] = getelementptr inbounds [12 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 [12 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 [[DOTFLOOR_0_IV___BEGIN2]], align 8
+// CHECK2-NEXT:    br label [[FOR_COND:%.*]]
+// CHECK2:       for.cond:
+// CHECK2-NEXT:    [[TMP6:%.*]] = load i64, ptr [[DOTFLOOR_0_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_END18:%.*]]
+// CHECK2:       for.body:
+// CHECK2-NEXT:    [[TMP8:%.*]] = load i64, ptr [[DOTFLOOR_0_IV___BEGIN2]], align 8
+// CHECK2-NEXT:    store i64 [[TMP8]], ptr [[DOTTILE_0_IV___BEGIN2]], align 8
+// CHECK2-NEXT:    br label [[FOR_COND7:%.*]]
+// CHECK2:       for.cond7:
+// CHECK2-NEXT:    [[TMP9:%.*]] = load i64, ptr [[DOTTILE_0_IV___BEGIN2]], align 8
+// CHECK2-NEXT:    [[TMP10:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR_4]], align 8
+// CHECK2-NEXT:    [[ADD8:%.*]] = add nsw i64 [[TMP10]], 1
+// CHECK2-NEXT:    [[TMP11:%.*]] = load i64, ptr [[DOTFLOOR_0_IV___BEGIN2]], align 8
+// CHECK2-NEXT:    [[ADD9:%.*]] = add nsw i64 [[TMP11]], 5
+// CHECK2-NEXT:    [[CMP10:%.*]] = icmp slt i64 [[ADD8]], [[ADD9]]
+// CHECK2-NEXT:    br i1 [[CMP10]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
+// CHECK2:       cond.true:
+// CHECK2-NEXT:    [[TMP12:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR_4]], align 8
+// CHECK2-NEXT:    [[ADD11:%.*]] = add nsw i64 [[TMP12]], 1
+// CHECK2-NEXT:    br label [[COND_END:%.*]]
+// CHECK2:       cond.false:
+// CHECK2-NEXT:    [[TMP13:%.*]] = load i64, ptr [[DOTFLOOR_0_IV___BEGIN2]], align 8
+// CHECK2-NEXT:    [[ADD12:%.*]] = add nsw i64 [[TMP13]], 5
+// CHECK2-NEXT:    br label [[COND_END]]
+// CHECK2:       cond.end:
+// CHECK2-NEXT:    [[COND:%.*]] = phi i64 [ [[ADD11]], [[COND_TRUE]] ], [ [[ADD12]], [[COND_FALSE]] ]
+// CHECK2-NEXT:    [[CMP13:%.*]] = icmp slt i64 [[TMP9]], [[COND]]
+// CHECK2-NEXT:    br i1 [[CMP13]], label [[FOR_BODY14:%.*]], label [[FOR_END:%.*]]
+// CHECK2:       for.body14:
+// CHECK2-NEXT:    [[TMP14:%.*]] = load ptr, ptr [[DOTCAPTURE_EXPR_]], align 8
+// CHECK2-NEXT:    [[TMP15:%.*]] = load i64, ptr [[DOTTILE_0_IV___BEGIN2]], align 8
+// CHECK2-NEXT:    [[MUL:%.*]] = mul nsw i64 [[TMP15]], 1
+// CHECK2-NEXT:    [[ADD_PTR15:%.*]] = getelementptr inbounds double, ptr [[TMP14]], i64 [[MUL]]
+// CHECK2-NEXT:    store ptr [[ADD_PTR15]], ptr [[__BEGIN2]], align 8
+// CHECK2-NEXT:    [[TMP16:%.*]] = load ptr, ptr [[__BEGIN2]], align 8
+// CHECK2-NEXT:    [[TMP17:%.*]] = load double, ptr [[TMP16]], align 8
+// CHECK2-NEXT:    store double [[TMP17]], ptr [[V]], align 8
+// CHECK2-NEXT:    [[TMP18:%.*]] = load double, ptr [[C]], align 8
+// CHECK2-NEXT:    [[TMP19:%.*]] = load double, ptr [[V]], align 8
+// CHECK2-NEXT:    call void (...) @body(double noundef [[TMP18]], double noundef [[TMP19]])
+// CHECK2-NEXT:    br label [[FOR_INC:%.*]]
+// CHECK2:       for.inc:
+// CHECK2-NEXT:    [[TMP20:%.*]] = load i64, ptr [[DOTTILE_0_IV___BEGIN2]], align 8
+// CHECK2-NEXT:    [[INC:%.*]] = add nsw i64 [[TMP20]], 1
+// CHECK2-NEXT:    store i64 [[INC]], ptr [[DOTTILE_0_IV___BEGIN2]], align 8
+// CHECK2-NEXT:    br label [[FOR_COND7]], !llvm.loop [[LOOP8:![0-9]+]]
+// CHECK2:       for.end:
+// CHECK2-NEXT:    br label [[FOR_INC16:%.*]]
+// CHECK2:       for.inc16:
+// CHECK2-NEXT:    [[TMP21:%.*]] = load i64, ptr [[DOTFLOOR_0_IV___BEGIN2]], align 8
+// CHECK2-NEXT:    [[ADD17:%.*]] = add nsw i64 [[TMP21]], 5
+// CHECK2-NEXT:    store i64 [[ADD17]], ptr [[DOTFLOOR_0_IV___BEGIN2]], align 8
+// CHECK2-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP9:![0-9]+]]
+// CHECK2:       for.end18:
+// CHECK2-NEXT:    ret void
+//
+//
+// CHECK2-LABEL: define dso_local void @foo2(
+// 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
@@ -1438,34 +1759,34 @@ extern "C" void foo8(int a) {
 // CHECK2-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTTILE_1_IV_J]], align 4
 // CHECK2-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP14]], 1
 // CHECK2-NEXT:    store i32 [[INC]], ptr [[DOTTILE_1_IV_J]], align 4
-// CHECK2-NEXT:    br label [[FOR_COND10]], !llvm.loop [[LOOP8:![0-9]+]]
+// CHECK2-NEXT:    br label [[FOR_COND10]], !llvm.loop [[LOOP10:![0-9]+]]
 // CHECK2:       for.end:
 // CHECK2-NEXT:    br label [[FOR_INC22:%.*]]
 // CHECK2:       for.inc22:
 // CHECK2-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTTILE_0_IV_I]], align 4
 // CHECK2-NEXT:    [[INC23:%.*]] = add nsw i32 [[TMP15]], 1
 // CHECK2-NEXT:    store i32 [[INC23]], ptr [[DOTTILE_0_IV_I]], align 4
-// CHECK2-NEXT:    br label [[FOR_COND4]], !llvm.loop [[LOOP9:![0-9]+]]
+// CHECK2-NEXT:    br label [[FOR_COND4]], !llvm.loop [[LOOP11:![0-9]+]]
 // CHECK2:       for.end24:
 // CHECK2-NEXT:    br label [[FOR_INC25:%.*]]
 // CHECK2:       for.inc25:
 // CHECK2-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTFLOOR_1_IV_J]], align 4
 // CHECK2-NEXT:    [[ADD26:%.*]] = add nsw i32 [[TMP16]], 5
 // CHECK2-NEXT:    store i32 [[ADD26]], ptr [[DOTFLOOR_1_IV_J]], align 4
-// CHECK2-NEXT:    br label [[FOR_COND1]], !llvm.loop [[LOOP10:![0-9]+]]
+// CHECK2-NEXT:    br label [[FOR_COND1]], !llvm.loop [[LOOP12:![0-9]+]]
 // CHECK2:       for.end27:
 // CHECK2-NEXT:    br label [[FOR_INC28:%.*]]
 // CHECK2:       for.inc28:
 // CHECK2-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTFLOOR_0_IV_I]], align 4
 // CHECK2-NEXT:    [[ADD29:%.*]] = add nsw i32 [[TMP17]], 5
 // CHECK2-NEXT:    store i32 [[ADD29]], ptr [[DOTFLOOR_0_IV_I]], align 4
-// CHECK2-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP11:![0-9]+]]
+// CHECK2-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP13:![0-9]+]]
 // CHECK2:       for.end30:
 // CHECK2-NEXT:    ret void
 //
 //
-// CHECK2-LABEL: define {{[^@]+}}@foo3
-// CHECK2-SAME: () #[[ATTR1]] {
+// CHECK2-LABEL: define dso_local void @foo3(
+// CHECK2-SAME: ) #[[ATTR1]] {
 // CHECK2-NEXT:  entry:
 // CHECK2-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK2-NEXT:    [[TMP:%.*]] = alloca i32, align 4
@@ -1574,21 +1895,21 @@ extern "C" void foo8(int a) {
 // CHECK2-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTTILE_1_IV_J]], align 4
 // CHECK2-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP20]], 1
 // CHECK2-NEXT:    store i32 [[INC]], ptr [[DOTTILE_1_IV_J]], align 4
-// CHECK2-NEXT:    br label [[FOR_COND15]], !llvm.loop [[LOOP12:![0-9]+]]
+// CHECK2-NEXT:    br label [[FOR_COND15]], !llvm.loop [[LOOP14:![0-9]+]]
 // CHECK2:       for.end:
 // CHECK2-NEXT:    br label [[FOR_INC27:%.*]]
 // CHECK2:       for.inc27:
 // CHECK2-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTTILE_0_IV_I]], align 4
 // CHECK2-NEXT:    [[INC28:%.*]] = add nsw i32 [[TMP21]], 1
 // CHECK2-NEXT:    store i32 [[INC28]], ptr [[DOTTILE_0_IV_I]], align 4
-// CHECK2-NEXT:    br label [[FOR_COND3]], !llvm.loop [[LOOP13:![0-9]+]]
+// CHECK2-NEXT:    br label [[FOR_COND3]], !llvm.loop [[LOOP15:![0-9]+]]
 // CHECK2:       for.end29:
 // CHECK2-NEXT:    br label [[FOR_INC30:%.*]]
 // CHECK2:       for.inc30:
 // CHECK2-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTFLOOR_1_IV_J]], align 4
 // CHECK2-NEXT:    [[ADD31:%.*]] = add nsw i32 [[TMP22]], 5
 // CHECK2-NEXT:    store i32 [[ADD31]], ptr [[DOTFLOOR_1_IV_J]], align 4
-// CHECK2-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP14:![0-9]+]]
+// CHECK2-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP16:![0-9]+]]
 // CHECK2:       for.end32:
 // CHECK2-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
 // CHECK2:       omp.body.continue:
@@ -1606,8 +1927,8 @@ extern "C" void foo8(int a) {
 // CHECK2-NEXT:    ret void
 //
 //
-// CHECK2-LABEL: define {{[^@]+}}@foo4
-// CHECK2-SAME: () #[[ATTR1]] {
+// CHECK2-LABEL: define dso_local void @foo4(
+// CHECK2-SAME: ) #[[ATTR1]] {
 // CHECK2-NEXT:  entry:
 // CHECK2-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK2-NEXT:    [[TMP:%.*]] = alloca i32, align 4
@@ -1727,21 +2048,21 @@ extern "C" void foo8(int a) {
 // CHECK2-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTTILE_1_IV_J]], align 4
 // CHECK2-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP22]], 1
 // CHECK2-NEXT:    store i32 [[INC]], ptr [[DOTTILE_1_IV_J]], align 4
-// CHECK2-NEXT:    br label [[FOR_COND20]], !llvm.loop [[LOOP15:![0-9]+]]
+// CHECK2-NEXT:    br label [[FOR_COND20]], !llvm.loop [[LOOP17:![0-9]+]]
 // CHECK2:       for.end:
 // CHECK2-NEXT:    br label [[FOR_INC32:%.*]]
 // CHECK2:       for.inc32:
 // CHECK2-NEXT:    [[TMP23:%.*]] = load i32, ptr [[DOTTILE_0_IV_I]], align 4
 // CHECK2-NEXT:    [[INC33:%.*]] = add nsw i32 [[TMP23]], 1
 // CHECK2-NEXT:    store i32 [[INC33]], ptr [[DOTTILE_0_IV_I]], align 4
-// CHECK2-NEXT:    br label [[FOR_COND8]], !llvm.loop [[LOOP16:![0-9]+]]
+// CHECK2-NEXT:    br label [[FOR_COND8]], !llvm.loop [[LOOP18:![0-9]+]]
 // CHECK2:       for.end34:
 // CHECK2-NEXT:    br label [[FOR_INC35:%.*]]
 // CHECK2:       for.inc35:
 // CHECK2-NEXT:    [[TMP24:%.*]] = load i32, ptr [[DOTFLOOR_1_IV_J]], align 4
 // CHECK2-NEXT:    [[ADD36:%.*]] = add nsw i32 [[TMP24]], 5
 // CHECK2-NEXT:    store i32 [[ADD36]], ptr [[DOTFLOOR_1_IV_J]], align 4
-// CHECK2-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP17:![0-9]+]]
+// CHECK2-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP19:![0-9]+]]
 // CHECK2:       for.end37:
 // CHECK2-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
 // CHECK2:       omp.body.continue:
@@ -1759,8 +2080,8 @@ extern "C" void foo8(int a) {
 // CHECK2-NEXT:    ret void
 //
 //
-// CHECK2-LABEL: define {{[^@]+}}@foo5
-// CHECK2-SAME: () #[[ATTR1]] {
+// CHECK2-LABEL: define dso_local void @foo5(
+// CHECK2-SAME: ) #[[ATTR1]] {
 // CHECK2-NEXT:  entry:
 // CHECK2-NEXT:    [[DOTOMP_IV:%.*]] = alloca i64, align 8
 // CHECK2-NEXT:    [[TMP:%.*]] = alloca i32, align 4
@@ -1968,15 +2289,15 @@ extern "C" void foo8(int a) {
 // CHECK2-NEXT:    ret void
 //
 //
-// CHECK2-LABEL: define {{[^@]+}}@foo6
-// CHECK2-SAME: () #[[ATTR1]] {
+// CHECK2-LABEL: define dso_local void @foo6(
+// CHECK2-SAME: ) #[[ATTR1]] {
 // CHECK2-NEXT:  entry:
 // CHECK2-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB2]], i32 0, ptr @foo6.omp_outlined)
 // CHECK2-NEXT:    ret void
 //
 //
-// CHECK2-LABEL: define {{[^@]+}}@foo6.omp_outlined
-// CHECK2-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]]) #[[ATTR4:[0-9]+]] {
+// CHECK2-LABEL: define internal void @foo6.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
@@ -2054,7 +2375,7 @@ extern "C" void foo8(int a) {
 // CHECK2-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTTILE_0_IV_I]], align 4
 // CHECK2-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP14]], 1
 // CHECK2-NEXT:    store i32 [[INC]], ptr [[DOTTILE_0_IV_I]], align 4
-// CHECK2-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP18:![0-9]+]]
+// CHECK2-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP20:![0-9]+]]
 // CHECK2:       for.end:
 // CHECK2-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
 // CHECK2:       omp.body.continue:
@@ -2071,8 +2392,8 @@ extern "C" void foo8(int a) {
 // CHECK2-NEXT:    ret void
 //
 //
-// CHECK2-LABEL: define {{[^@]+}}@foo8
-// CHECK2-SAME: (i32 noundef [[A:%.*]]) #[[ATTR1]] {
+// CHECK2-LABEL: define dso_local void @foo8(
+// CHECK2-SAME: i32 noundef [[A:%.*]]) #[[ATTR1]] {
 // CHECK2-NEXT:  entry:
 // CHECK2-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
 // CHECK2-NEXT:    [[I:%.*]] = alloca i32, align 4
@@ -2138,7 +2459,7 @@ extern "C" void foo8(int a) {
 // CHECK2-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTTILE_0_IV_I]], align 4
 // CHECK2-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP11]], 1
 // CHECK2-NEXT:    store i32 [[INC]], ptr [[DOTTILE_0_IV_I]], align 4
-// CHECK2-NEXT:    br label [[FOR_COND1]], !llvm.loop [[LOOP21:![0-9]+]]
+// CHECK2-NEXT:    br label [[FOR_COND1]], !llvm.loop [[LOOP23:![0-9]+]]
 // CHECK2:       for.end:
 // CHECK2-NEXT:    br label [[FOR_INC17:%.*]]
 // CHECK2:       for.inc17:
@@ -2155,20 +2476,117 @@ extern "C" void foo8(int a) {
 // CHECK2-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTFLOOR_0_IV_I]], align 4
 // CHECK2-NEXT:    [[ADD23:%.*]] = add nsw i32 [[TMP14]], [[COND22]]
 // CHECK2-NEXT:    store i32 [[ADD23]], ptr [[DOTFLOOR_0_IV_I]], align 4
-// CHECK2-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP22:![0-9]+]]
+// CHECK2-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP24:![0-9]+]]
 // CHECK2:       for.end24:
 // CHECK2-NEXT:    ret void
 //
 //
-// CHECK2-LABEL: define {{[^@]+}}@tfoo7
-// CHECK2-SAME: () #[[ATTR1]] {
+// CHECK2-LABEL: define dso_local void @foo9(
+// CHECK2-SAME: ptr noundef byval([[STRUCT_DATA_T:%.*]]) align 8 [[DATA:%.*]]) #[[ATTR1]] {
+// CHECK2-NEXT:  entry:
+// 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:    [[DOTFLOOR_0_IV___BEGIN2:%.*]] = alloca i64, align 8
+// CHECK2-NEXT:    [[DOTTILE_0_IV___BEGIN2:%.*]] = alloca i64, align 8
+// CHECK2-NEXT:    [[V:%.*]] = alloca double, align 8
+// CHECK2-NEXT:    [[ARRAY:%.*]] = getelementptr inbounds [[STRUCT_DATA_T]], ptr [[DATA]], i32 0, i32 0
+// CHECK2-NEXT:    store ptr [[ARRAY]], ptr [[__RANGE2]], align 8
+// CHECK2-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[__RANGE2]], align 8
+// CHECK2-NEXT:    [[ARRAYDECAY:%.*]] = getelementptr inbounds [12 x double], ptr [[TMP0]], i64 0, i64 0
+// CHECK2-NEXT:    [[ADD_PTR:%.*]] = getelementptr inbounds double, ptr [[ARRAYDECAY]], i64 12
+// CHECK2-NEXT:    store ptr [[ADD_PTR]], ptr [[__END2]], align 8
+// CHECK2-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[__RANGE2]], align 8
+// CHECK2-NEXT:    [[ARRAYDECAY1:%.*]] = getelementptr inbounds [12 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 [12 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 [[DOTFLOOR_0_IV___BEGIN2]], align 8
+// CHECK2-NEXT:    br label [[FOR_COND:%.*]]
+// CHECK2:       for.cond:
+// CHECK2-NEXT:    [[TMP6:%.*]] = load i64, ptr [[DOTFLOOR_0_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_END18:%.*]]
+// CHECK2:       for.body:
+// CHECK2-NEXT:    [[TMP8:%.*]] = load i64, ptr [[DOTFLOOR_0_IV___BEGIN2]], align 8
+// CHECK2-NEXT:    store i64 [[TMP8]], ptr [[DOTTILE_0_IV___BEGIN2]], align 8
+// CHECK2-NEXT:    br label [[FOR_COND7:%.*]]
+// CHECK2:       for.cond7:
+// CHECK2-NEXT:    [[TMP9:%.*]] = load i64, ptr [[DOTTILE_0_IV___BEGIN2]], align 8
+// CHECK2-NEXT:    [[TMP10:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR_4]], align 8
+// CHECK2-NEXT:    [[ADD8:%.*]] = add nsw i64 [[TMP10]], 1
+// CHECK2-NEXT:    [[TMP11:%.*]] = load i64, ptr [[DOTFLOOR_0_IV___BEGIN2]], align 8
+// CHECK2-NEXT:    [[ADD9:%.*]] = add nsw i64 [[TMP11]], 5
+// CHECK2-NEXT:    [[CMP10:%.*]] = icmp slt i64 [[ADD8]], [[ADD9]]
+// CHECK2-NEXT:    br i1 [[CMP10]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
+// CHECK2:       cond.true:
+// CHECK2-NEXT:    [[TMP12:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR_4]], align 8
+// CHECK2-NEXT:    [[ADD11:%.*]] = add nsw i64 [[TMP12]], 1
+// CHECK2-NEXT:    br label [[COND_END:%.*]]
+// CHECK2:       cond.false:
+// CHECK2-NEXT:    [[TMP13:%.*]] = load i64, ptr [[DOTFLOOR_0_IV___BEGIN2]], align 8
+// CHECK2-NEXT:    [[ADD12:%.*]] = add nsw i64 [[TMP13]], 5
+// CHECK2-NEXT:    br label [[COND_END]]
+// CHECK2:       cond.end:
+// CHECK2-NEXT:    [[COND:%.*]] = phi i64 [ [[ADD11]], [[COND_TRUE]] ], [ [[ADD12]], [[COND_FALSE]] ]
+// CHECK2-NEXT:    [[CMP13:%.*]] = icmp slt i64 [[TMP9]], [[COND]]
+// CHECK2-NEXT:    br i1 [[CMP13]], label [[FOR_BODY14:%.*]], label [[FOR_END:%.*]]
+// CHECK2:       for.body14:
+// CHECK2-NEXT:    [[TMP14:%.*]] = load ptr, ptr [[DOTCAPTURE_EXPR_]], align 8
+// CHECK2-NEXT:    [[TMP15:%.*]] = load i64, ptr [[DOTTILE_0_IV___BEGIN2]], align 8
+// CHECK2-NEXT:    [[MUL:%.*]] = mul nsw i64 [[TMP15]], 1
+// CHECK2-NEXT:    [[ADD_PTR15:%.*]] = getelementptr inbounds double, ptr [[TMP14]], i64 [[MUL]]
+// CHECK2-NEXT:    store ptr [[ADD_PTR15]], ptr [[__BEGIN2]], align 8
+// CHECK2-NEXT:    [[TMP16:%.*]] = load ptr, ptr [[__BEGIN2]], align 8
+// CHECK2-NEXT:    [[TMP17:%.*]] = load double, ptr [[TMP16]], align 8
+// CHECK2-NEXT:    store double [[TMP17]], ptr [[V]], align 8
+// CHECK2-NEXT:    [[TMP18:%.*]] = load double, ptr [[V]], align 8
+// CHECK2-NEXT:    call void (...) @body(double noundef [[TMP18]])
+// CHECK2-NEXT:    br label [[FOR_INC:%.*]]
+// CHECK2:       for.inc:
+// CHECK2-NEXT:    [[TMP19:%.*]] = load i64, ptr [[DOTTILE_0_IV___BEGIN2]], align 8
+// CHECK2-NEXT:    [[INC:%.*]] = add nsw i64 [[TMP19]], 1
+// CHECK2-NEXT:    store i64 [[INC]], ptr [[DOTTILE_0_IV___BEGIN2]], align 8
+// CHECK2-NEXT:    br label [[FOR_COND7]], !llvm.loop [[LOOP25:![0-9]+]]
+// CHECK2:       for.end:
+// CHECK2-NEXT:    br label [[FOR_INC16:%.*]]
+// CHECK2:       for.inc16:
+// CHECK2-NEXT:    [[TMP20:%.*]] = load i64, ptr [[DOTFLOOR_0_IV___BEGIN2]], align 8
+// CHECK2-NEXT:    [[ADD17:%.*]] = add nsw i64 [[TMP20]], 5
+// CHECK2-NEXT:    store i64 [[ADD17]], ptr [[DOTFLOOR_0_IV___BEGIN2]], align 8
+// CHECK2-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP26:![0-9]+]]
+// CHECK2:       for.end18:
+// CHECK2-NEXT:    ret void
+//
+//
+// CHECK2-LABEL: define dso_local void @tfoo7(
+// CHECK2-SAME: ) #[[ATTR1]] {
 // CHECK2-NEXT:  entry:
 // CHECK2-NEXT:    call void @_Z4foo7IiTnT_Li3ETnS0_Li5EEvS0_S0_(i32 noundef 0, i32 noundef 42)
 // CHECK2-NEXT:    ret void
 //
 //
-// CHECK2-LABEL: define {{[^@]+}}@_Z4foo7IiTnT_Li3ETnS0_Li5EEvS0_S0_
-// CHECK2-SAME: (i32 noundef [[START:%.*]], i32 noundef [[END:%.*]]) #[[ATTR1]] comdat {
+// CHECK2-LABEL: define linkonce_odr void @_Z4foo7IiTnT_Li3ETnS0_Li5EEvS0_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
@@ -2211,7 +2629,7 @@ extern "C" void foo8(int a) {
 // CHECK2-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK2-NEXT:    [[ADD7:%.*]] = add i32 [[TMP9]], 1
 // CHECK2-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTFLOOR_0_IV_I]], align 4
-// CHECK2-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP10]], 5
+// CHECK2-NEXT:    [[ADD8:%.*]] = add i32 [[TMP10]], 5
 // CHECK2-NEXT:    [[CMP9:%.*]] = icmp ult i32 [[ADD7]], [[ADD8]]
 // CHECK2-NEXT:    br i1 [[CMP9]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK2:       cond.true:
@@ -2220,7 +2638,7 @@ extern "C" void foo8(int a) {
 // CHECK2-NEXT:    br label [[COND_END:%.*]]
 // CHECK2:       cond.false:
 // CHECK2-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTFLOOR_0_IV_I]], align 4
-// CHECK2-NEXT:    [[ADD11:%.*]] = add nsw i32 [[TMP12]], 5
+// CHECK2-NEXT:    [[ADD11:%.*]] = add i32 [[TMP12]], 5
 // CHECK2-NEXT:    br label [[COND_END]]
 // CHECK2:       cond.end:
 // CHECK2-NEXT:    [[COND:%.*]] = phi i32 [ [[ADD10]], [[COND_TRUE]] ], [ [[ADD11]], [[COND_FALSE]] ]
@@ -2237,23 +2655,74 @@ extern "C" void foo8(int a) {
 // CHECK2-NEXT:    br label [[FOR_INC:%.*]]
 // CHECK2:       for.inc:
 // CHECK2-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTTILE_0_IV_I]], align 4
-// CHECK2-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP16]], 1
+// CHECK2-NEXT:    [[INC:%.*]] = add i32 [[TMP16]], 1
 // CHECK2-NEXT:    store i32 [[INC]], ptr [[DOTTILE_0_IV_I]], align 4
-// CHECK2-NEXT:    br label [[FOR_COND6]], !llvm.loop [[LOOP23:![0-9]+]]
+// CHECK2-NEXT:    br label [[FOR_COND6]], !llvm.loop [[LOOP27:![0-9]+]]
 // CHECK2:       for.end:
 // CHECK2-NEXT:    br label [[FOR_INC15:%.*]]
 // CHECK2:       for.inc15:
 // CHECK2-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTFLOOR_0_IV_I]], align 4
-// CHECK2-NEXT:    [[ADD16:%.*]] = add nsw i32 [[TMP17]], 5
+// CHECK2-NEXT:    [[ADD16:%.*]] = add i32 [[TMP17]], 5
 // CHECK2-NEXT:    store i32 [[ADD16]], ptr [[DOTFLOOR_0_IV_I]], align 4
-// CHECK2-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP24:![0-9]+]]
+// CHECK2-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP28:![0-9]+]]
 // CHECK2:       for.end17:
 // CHECK2-NEXT:    ret void
 //
 //
-// CHECK2-LABEL: define {{[^@]+}}@_GLOBAL__sub_I_tile_codegen.cpp
-// CHECK2-SAME: () #[[ATTR0]] section ".text.startup" {
+// CHECK2-LABEL: define internal void @_GLOBAL__sub_I_tile_codegen.cpp(
+// CHECK2-SAME: ) #[[ATTR0]] section ".text.startup" {
 // CHECK2-NEXT:  entry:
 // CHECK2-NEXT:    call void @__cxx_global_var_init()
 // CHECK2-NEXT:    ret void
 //
+//.
+// CHECK1: [[LOOP3]] = distinct !{[[LOOP3]], [[META4:![0-9]+]]}
+// CHECK1: [[META4]] = !{!"llvm.loop.mustprogress"}
+// CHECK1: [[LOOP5]] = distinct !{[[LOOP5]], [[META4]]}
+// CHECK1: [[LOOP6]] = distinct !{[[LOOP6]], [[META4]]}
+// CHECK1: [[LOOP7]] = distinct !{[[LOOP7]], [[META4]]}
+// CHECK1: [[LOOP8]] = distinct !{[[LOOP8]], [[META4]]}
+// CHECK1: [[LOOP9]] = distinct !{[[LOOP9]], [[META4]]}
+// CHECK1: [[LOOP10]] = distinct !{[[LOOP10]], [[META4]]}
+// CHECK1: [[LOOP11]] = distinct !{[[LOOP11]], [[META4]]}
+// CHECK1: [[LOOP12]] = distinct !{[[LOOP12]], [[META4]]}
+// CHECK1: [[LOOP13]] = distinct !{[[LOOP13]], [[META4]]}
+// CHECK1: [[LOOP14]] = distinct !{[[LOOP14]], [[META4]]}
+// CHECK1: [[LOOP15]] = distinct !{[[LOOP15]], [[META4]]}
+// CHECK1: [[LOOP16]] = distinct !{[[LOOP16]], [[META4]]}
+// CHECK1: [[LOOP17]] = distinct !{[[LOOP17]], [[META4]]}
+// CHECK1: [[LOOP18]] = distinct !{[[LOOP18]], [[META4]]}
+// CHECK1: [[LOOP21]] = distinct !{[[LOOP21]], [[META4]]}
+// CHECK1: [[LOOP22]] = distinct !{[[LOOP22]], [[META4]]}
+// CHECK1: [[LOOP23]] = distinct !{[[LOOP23]], [[META4]]}
+// CHECK1: [[LOOP24]] = distinct !{[[LOOP24]], [[META4]]}
+// CHECK1: [[LOOP25]] = distinct !{[[LOOP25]], [[META4]]}
+// CHECK1: [[LOOP26]] = distinct !{[[LOOP26]], [[META4]]}
+// CHECK1: [[LOOP27]] = distinct !{[[LOOP27]], [[META4]]}
+// CHECK1: [[LOOP28]] = distinct !{[[LOOP28]], [[META4]]}
+//.
+// CHECK2: [[LOOP3]] = distinct !{[[LOOP3]], [[META4:![0-9]+]]}
+// CHECK2: [[META4]] = !{!"llvm.loop.mustprogress"}
+// CHECK2: [[LOOP5]] = distinct !{[[LOOP5]], [[META4]]}
+// CHECK2: [[LOOP6]] = distinct !{[[LOOP6]], [[META4]]}
+// CHECK2: [[LOOP7]] = distinct !{[[LOOP7]], [[META4]]}
+// CHECK2: [[LOOP8]] = distinct !{[[LOOP8]], [[META4]]}
+// CHECK2: [[LOOP9]] = distinct !{[[LOOP9]], [[META4]]}
+// CHECK2: [[LOOP10]] = distinct !{[[LOOP10]], [[META4]]}
+// CHECK2: [[LOOP11]] = distinct !{[[LOOP11]], [[META4]]}
+// CHECK2: [[LOOP12]] = distinct !{[[LOOP12]], [[META4]]}
+// CHECK2: [[LOOP13]] = distinct !{[[LOOP13]], [[META4]]}
+// CHECK2: [[LOOP14]] = distinct !{[[LOOP14]], [[META4]]}
+// CHECK2: [[LOOP15]] = distinct !{[[LOOP15]], [[META4]]}
+// CHECK2: [[LOOP16]] = distinct !{[[LOOP16]], [[META4]]}
+// CHECK2: [[LOOP17]] = distinct !{[[LOOP17]], [[META4]]}
+// CHECK2: [[LOOP18]] = distinct !{[[LOOP18]], [[META4]]}
+// CHECK2: [[LOOP19]] = distinct !{[[LOOP19]], [[META4]]}
+// CHECK2: [[LOOP20]] = distinct !{[[LOOP20]], [[META4]]}
+// CHECK2: [[LOOP23]] = distinct !{[[LOOP23]], [[META4]]}
+// CHECK2: [[LOOP24]] = distinct !{[[LOOP24]], [[META4]]}
+// CHECK2: [[LOOP25]] = distinct !{[[LOOP25]], [[META4]]}
+// CHECK2: [[LOOP26]] = distinct !{[[LOOP26]], [[META4]]}
+// CHECK2: [[LOOP27]] = distinct !{[[LOOP27]], [[META4]]}
+// CHECK2: [[LOOP28]] = distinct !{[[LOOP28]], [[META4]]}
+//.
diff --git a/clang/test/OpenMP/tile_codegen_for_dependent.cpp b/clang/test/OpenMP/tile_codegen_for_dependent.cpp
index 93c51c9165a47..820d33d15287b 100644
--- a/clang/test/OpenMP/tile_codegen_for_dependent.cpp
+++ b/clang/test/OpenMP/tile_codegen_for_dependent.cpp
@@ -17,7 +17,7 @@
 extern "C" void body(...) {}
 
 
-// IR-LABEL: @func(
+// IR-LABEL: define {{.*}}@func(
 // IR-NEXT:  [[ENTRY:.*]]:
 // IR-NEXT:    %[[START_ADDR:.+]] = alloca i32, align 4
 // IR-NEXT:    %[[END_ADDR:.+]] = alloca i32, align 4
@@ -27,18 +27,18 @@ extern "C" void body(...) {}
 // IR-NEXT:    %[[I:.+]] = alloca i32, align 4
 // IR-NEXT:    %[[DOTCAPTURE_EXPR_:.+]] = alloca i32, align 4
 // IR-NEXT:    %[[DOTCAPTURE_EXPR_1:.+]] = alloca i32, align 4
+// IR-NEXT:    %[[DOTNEW_STEP:.+]] = alloca i32, align 4
 // IR-NEXT:    %[[DOTCAPTURE_EXPR_2:.+]] = alloca i32, align 4
-// IR-NEXT:    %[[DOTCAPTURE_EXPR_3:.+]] = alloca i32, align 4
-// IR-NEXT:    %[[DOTCAPTURE_EXPR_6:.+]] = alloca i32, align 4
-// IR-NEXT:    %[[DOTCAPTURE_EXPR_8:.+]] = alloca i32, align 4
+// IR-NEXT:    %[[DOTCAPTURE_EXPR_5:.+]] = alloca i32, align 4
+// IR-NEXT:    %[[DOTCAPTURE_EXPR_7:.+]] = alloca i32, align 4
 // IR-NEXT:    %[[DOTFLOOR_0_IV_I:.+]] = alloca i32, align 4
 // IR-NEXT:    %[[DOTOMP_LB:.+]] = alloca i32, align 4
 // IR-NEXT:    %[[DOTOMP_UB:.+]] = alloca i32, align 4
 // IR-NEXT:    %[[DOTOMP_STRIDE:.+]] = alloca i32, align 4
 // IR-NEXT:    %[[DOTOMP_IS_LAST:.+]] = alloca i32, align 4
-// IR-NEXT:    %[[DOTFLOOR_0_IV_I12:.+]] = alloca i32, align 4
+// IR-NEXT:    %[[DOTFLOOR_0_IV_I11:.+]] = alloca i32, align 4
 // IR-NEXT:    %[[DOTTILE_0_IV_I:.+]] = alloca i32, align 4
-// IR-NEXT:    %[[TMP0:.+]] = call i32 @__kmpc_global_thread_num(ptr @2)
+// IR-NEXT:    %[[TMP0:.+]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB2:.+]])
 // IR-NEXT:    store i32 %[[START:.+]], ptr %[[START_ADDR]], align 4
 // IR-NEXT:    store i32 %[[END:.+]], ptr %[[END_ADDR]], align 4
 // IR-NEXT:    store i32 %[[STEP:.+]], ptr %[[STEP_ADDR]], align 4
@@ -49,44 +49,44 @@ extern "C" void body(...) {}
 // IR-NEXT:    %[[TMP3:.+]] = load i32, ptr %[[END_ADDR]], align 4
 // IR-NEXT:    store i32 %[[TMP3]], ptr %[[DOTCAPTURE_EXPR_1]], align 4
 // IR-NEXT:    %[[TMP4:.+]] = load i32, ptr %[[STEP_ADDR]], align 4
-// IR-NEXT:    store i32 %[[TMP4]], ptr %[[DOTCAPTURE_EXPR_2]], align 4
+// IR-NEXT:    store i32 %[[TMP4]], ptr %[[DOTNEW_STEP]], align 4
 // IR-NEXT:    %[[TMP5:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_1]], align 4
 // IR-NEXT:    %[[TMP6:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_]], align 4
 // IR-NEXT:    %[[SUB:.+]] = sub i32 %[[TMP5]], %[[TMP6]]
-// IR-NEXT:    %[[SUB4:.+]] = sub i32 %[[SUB]], 1
-// IR-NEXT:    %[[TMP7:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_2]], align 4
-// IR-NEXT:    %[[ADD:.+]] = add i32 %[[SUB4]], %[[TMP7]]
-// IR-NEXT:    %[[TMP8:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_2]], align 4
+// IR-NEXT:    %[[SUB3:.+]] = sub i32 %[[SUB]], 1
+// IR-NEXT:    %[[TMP7:.+]] = load i32, ptr %[[DOTNEW_STEP]], align 4
+// IR-NEXT:    %[[ADD:.+]] = add i32 %[[SUB3]], %[[TMP7]]
+// IR-NEXT:    %[[TMP8:.+]] = load i32, ptr %[[DOTNEW_STEP]], align 4
 // IR-NEXT:    %[[DIV:.+]] = udiv i32 %[[ADD]], %[[TMP8]]
-// IR-NEXT:    %[[SUB5:.+]] = sub i32 %[[DIV]], 1
-// IR-NEXT:    store i32 %[[SUB5]], ptr %[[DOTCAPTURE_EXPR_3]], align 4
-// IR-NEXT:    %[[TMP9:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_3]], align 4
-// IR-NEXT:    %[[ADD7:.+]] = add i32 %[[TMP9]], 1
-// IR-NEXT:    store i32 %[[ADD7]], ptr %[[DOTCAPTURE_EXPR_6]], align 4
-// IR-NEXT:    %[[TMP10:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_6]], align 4
-// IR-NEXT:    %[[SUB9:.+]] = sub i32 %[[TMP10]], -3
-// IR-NEXT:    %[[DIV10:.+]] = udiv i32 %[[SUB9]], 4
-// IR-NEXT:    %[[SUB11:.+]] = sub i32 %[[DIV10]], 1
-// IR-NEXT:    store i32 %[[SUB11]], ptr %[[DOTCAPTURE_EXPR_8]], align 4
+// IR-NEXT:    %[[SUB4:.+]] = sub i32 %[[DIV]], 1
+// IR-NEXT:    store i32 %[[SUB4]], ptr %[[DOTCAPTURE_EXPR_2]], align 4
+// IR-NEXT:    %[[TMP9:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_2]], align 4
+// IR-NEXT:    %[[ADD6:.+]] = add i32 %[[TMP9]], 1
+// IR-NEXT:    store i32 %[[ADD6]], ptr %[[DOTCAPTURE_EXPR_5]], align 4
+// IR-NEXT:    %[[TMP10:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_5]], align 4
+// IR-NEXT:    %[[SUB8:.+]] = sub i32 %[[TMP10]], -3
+// IR-NEXT:    %[[DIV9:.+]] = udiv i32 %[[SUB8]], 4
+// IR-NEXT:    %[[SUB10:.+]] = sub i32 %[[DIV9]], 1
+// IR-NEXT:    store i32 %[[SUB10]], ptr %[[DOTCAPTURE_EXPR_7]], align 4
 // IR-NEXT:    store i32 0, ptr %[[DOTFLOOR_0_IV_I]], align 4
-// IR-NEXT:    %[[TMP11:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_6]], align 4
+// IR-NEXT:    %[[TMP11:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_5]], align 4
 // IR-NEXT:    %[[CMP:.+]] = icmp ult i32 0, %[[TMP11]]
 // IR-NEXT:    br i1 %[[CMP]], label %[[OMP_PRECOND_THEN:.+]], label %[[OMP_PRECOND_END:.+]]
 // IR-EMPTY:
 // IR-NEXT:  [[OMP_PRECOND_THEN]]:
 // IR-NEXT:    store i32 0, ptr %[[DOTOMP_LB]], align 4
-// IR-NEXT:    %[[TMP12:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_8]], align 4
+// IR-NEXT:    %[[TMP12:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_7]], align 4
 // IR-NEXT:    store i32 %[[TMP12]], ptr %[[DOTOMP_UB]], align 4
 // IR-NEXT:    store i32 1, ptr %[[DOTOMP_STRIDE]], align 4
 // IR-NEXT:    store i32 0, ptr %[[DOTOMP_IS_LAST]], align 4
-// IR-NEXT:    call void @__kmpc_for_static_init_4u(ptr @1, i32 %[[TMP0]], i32 34, ptr %[[DOTOMP_IS_LAST]], ptr %[[DOTOMP_LB]], ptr %[[DOTOMP_UB]], ptr %[[DOTOMP_STRIDE]], i32 1, i32 1)
+// IR-NEXT:    call void @__kmpc_for_static_init_4u(ptr @[[GLOB1:.+]], i32 %[[TMP0]], i32 34, ptr %[[DOTOMP_IS_LAST]], ptr %[[DOTOMP_LB]], ptr %[[DOTOMP_UB]], ptr %[[DOTOMP_STRIDE]], i32 1, i32 1)
 // IR-NEXT:    %[[TMP13:.+]] = load i32, ptr %[[DOTOMP_UB]], align 4
-// IR-NEXT:    %[[TMP14:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_8]], align 4
-// IR-NEXT:    %[[CMP13:.+]] = icmp ugt i32 %[[TMP13]], %[[TMP14]]
-// IR-NEXT:    br i1 %[[CMP13]], label %[[COND_TRUE:.+]], label %[[COND_FALSE:.+]]
+// IR-NEXT:    %[[TMP14:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_7]], align 4
+// IR-NEXT:    %[[CMP12:.+]] = icmp ugt i32 %[[TMP13]], %[[TMP14]]
+// IR-NEXT:    br i1 %[[CMP12]], label %[[COND_TRUE:.+]], label %[[COND_FALSE:.+]]
 // IR-EMPTY:
 // IR-NEXT:  [[COND_TRUE]]:
-// IR-NEXT:    %[[TMP15:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_8]], align 4
+// IR-NEXT:    %[[TMP15:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_7]], align 4
 // IR-NEXT:    br label %[[COND_END:.+]]
 // IR-EMPTY:
 // IR-NEXT:  [[COND_FALSE]]:
@@ -103,50 +103,50 @@ extern "C" void body(...) {}
 // IR-NEXT:  [[OMP_INNER_FOR_COND]]:
 // IR-NEXT:    %[[TMP18:.+]] = load i32, ptr %[[DOTOMP_IV]], align 4
 // IR-NEXT:    %[[TMP19:.+]] = load i32, ptr %[[DOTOMP_UB]], align 4
-// IR-NEXT:    %[[ADD14:.+]] = add i32 %[[TMP19]], 1
-// IR-NEXT:    %[[CMP15:.+]] = icmp ult i32 %[[TMP18]], %[[ADD14]]
-// IR-NEXT:    br i1 %[[CMP15]], label %[[OMP_INNER_FOR_BODY:.+]], label %[[OMP_INNER_FOR_END:.+]]
+// IR-NEXT:    %[[ADD13:.+]] = add i32 %[[TMP19]], 1
+// IR-NEXT:    %[[CMP14:.+]] = icmp ult i32 %[[TMP18]], %[[ADD13]]
+// IR-NEXT:    br i1 %[[CMP14]], label %[[OMP_INNER_FOR_BODY:.+]], label %[[OMP_INNER_FOR_END:.+]]
 // IR-EMPTY:
 // IR-NEXT:  [[OMP_INNER_FOR_BODY]]:
 // IR-NEXT:    %[[TMP20:.+]] = load i32, ptr %[[DOTOMP_IV]], align 4
 // IR-NEXT:    %[[MUL:.+]] = mul i32 %[[TMP20]], 4
-// IR-NEXT:    %[[ADD16:.+]] = add i32 0, %[[MUL]]
-// IR-NEXT:    store i32 %[[ADD16]], ptr %[[DOTFLOOR_0_IV_I12]], align 4
-// IR-NEXT:    %[[TMP21:.+]] = load i32, ptr %[[DOTFLOOR_0_IV_I12]], align 4
+// IR-NEXT:    %[[ADD15:.+]] = add i32 0, %[[MUL]]
+// IR-NEXT:    store i32 %[[ADD15]], ptr %[[DOTFLOOR_0_IV_I11]], align 4
+// IR-NEXT:    %[[TMP21:.+]] = load i32, ptr %[[DOTFLOOR_0_IV_I11]], align 4
 // IR-NEXT:    store i32 %[[TMP21]], ptr %[[DOTTILE_0_IV_I]], align 4
 // IR-NEXT:    br label %[[FOR_COND:.+]]
 // IR-EMPTY:
 // IR-NEXT:  [[FOR_COND]]:
 // IR-NEXT:    %[[TMP22:.+]] = load i32, ptr %[[DOTTILE_0_IV_I]], align 4
-// IR-NEXT:    %[[TMP23:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_3]], align 4
-// IR-NEXT:    %[[ADD17:.+]] = add i32 %[[TMP23]], 1
-// IR-NEXT:    %[[TMP24:.+]] = load i32, ptr %[[DOTFLOOR_0_IV_I12]], align 4
-// IR-NEXT:    %[[ADD18:.+]] = add nsw i32 %[[TMP24]], 4
-// IR-NEXT:    %[[CMP19:.+]] = icmp ult i32 %[[ADD17]], %[[ADD18]]
-// IR-NEXT:    br i1 %[[CMP19]], label %[[COND_TRUE20:.+]], label %[[COND_FALSE22:.+]]
-// IR-EMPTY:
-// IR-NEXT:  [[COND_TRUE20]]:
-// IR-NEXT:    %[[TMP25:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_3]], align 4
-// IR-NEXT:    %[[ADD21:.+]] = add i32 %[[TMP25]], 1
-// IR-NEXT:    br label %[[COND_END24:.+]]
-// IR-EMPTY:
-// IR-NEXT:  [[COND_FALSE22]]:
-// IR-NEXT:    %[[TMP26:.+]] = load i32, ptr %[[DOTFLOOR_0_IV_I12]], align 4
-// IR-NEXT:    %[[ADD23:.+]] = add nsw i32 %[[TMP26]], 4
-// IR-NEXT:    br label %[[COND_END24]]
-// IR-EMPTY:
-// IR-NEXT:  [[COND_END24]]:
-// IR-NEXT:    %[[COND25:.+]] = phi i32 [ %[[ADD21]], %[[COND_TRUE20]] ], [ %[[ADD23]], %[[COND_FALSE22]] ]
-// IR-NEXT:    %[[CMP26:.+]] = icmp ult i32 %[[TMP22]], %[[COND25]]
-// IR-NEXT:    br i1 %[[CMP26]], label %[[FOR_BODY:.+]], label %[[FOR_END:.+]]
+// IR-NEXT:    %[[TMP23:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_2]], align 4
+// IR-NEXT:    %[[ADD16:.+]] = add i32 %[[TMP23]], 1
+// IR-NEXT:    %[[TMP24:.+]] = load i32, ptr %[[DOTFLOOR_0_IV_I11]], align 4
+// IR-NEXT:    %[[ADD17:.+]] = add i32 %[[TMP24]], 4
+// IR-NEXT:    %[[CMP18:.+]] = icmp ult i32 %[[ADD16]], %[[ADD17]]
+// IR-NEXT:    br i1 %[[CMP18]], label %[[COND_TRUE19:.+]], label %[[COND_FALSE21:.+]]
+// IR-EMPTY:
+// IR-NEXT:  [[COND_TRUE19]]:
+// IR-NEXT:    %[[TMP25:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_2]], align 4
+// IR-NEXT:    %[[ADD20:.+]] = add i32 %[[TMP25]], 1
+// IR-NEXT:    br label %[[COND_END23:.+]]
+// IR-EMPTY:
+// IR-NEXT:  [[COND_FALSE21]]:
+// IR-NEXT:    %[[TMP26:.+]] = load i32, ptr %[[DOTFLOOR_0_IV_I11]], align 4
+// IR-NEXT:    %[[ADD22:.+]] = add i32 %[[TMP26]], 4
+// IR-NEXT:    br label %[[COND_END23]]
+// IR-EMPTY:
+// IR-NEXT:  [[COND_END23]]:
+// IR-NEXT:    %[[COND24:.+]] = phi i32 [ %[[ADD20]], %[[COND_TRUE19]] ], [ %[[ADD22]], %[[COND_FALSE21]] ]
+// IR-NEXT:    %[[CMP25:.+]] = icmp ult i32 %[[TMP22]], %[[COND24]]
+// IR-NEXT:    br i1 %[[CMP25]], label %[[FOR_BODY:.+]], label %[[FOR_END:.+]]
 // IR-EMPTY:
 // IR-NEXT:  [[FOR_BODY]]:
 // IR-NEXT:    %[[TMP27:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_]], align 4
 // IR-NEXT:    %[[TMP28:.+]] = load i32, ptr %[[DOTTILE_0_IV_I]], align 4
-// IR-NEXT:    %[[TMP29:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_2]], align 4
-// IR-NEXT:    %[[MUL27:.+]] = mul i32 %[[TMP28]], %[[TMP29]]
-// IR-NEXT:    %[[ADD28:.+]] = add i32 %[[TMP27]], %[[MUL27]]
-// IR-NEXT:    store i32 %[[ADD28]], ptr %[[I]], align 4
+// IR-NEXT:    %[[TMP29:.+]] = load i32, ptr %[[DOTNEW_STEP]], align 4
+// IR-NEXT:    %[[MUL26:.+]] = mul i32 %[[TMP28]], %[[TMP29]]
+// IR-NEXT:    %[[ADD27:.+]] = add i32 %[[TMP27]], %[[MUL26]]
+// IR-NEXT:    store i32 %[[ADD27]], ptr %[[I]], align 4
 // IR-NEXT:    %[[TMP30:.+]] = load i32, ptr %[[START_ADDR]], align 4
 // IR-NEXT:    %[[TMP31:.+]] = load i32, ptr %[[END_ADDR]], align 4
 // IR-NEXT:    %[[TMP32:.+]] = load i32, ptr %[[STEP_ADDR]], align 4
@@ -156,9 +156,9 @@ extern "C" void body(...) {}
 // IR-EMPTY:
 // IR-NEXT:  [[FOR_INC]]:
 // IR-NEXT:    %[[TMP34:.+]] = load i32, ptr %[[DOTTILE_0_IV_I]], align 4
-// IR-NEXT:    %[[INC:.+]] = add nsw i32 %[[TMP34]], 1
+// IR-NEXT:    %[[INC:.+]] = add i32 %[[TMP34]], 1
 // IR-NEXT:    store i32 %[[INC]], ptr %[[DOTTILE_0_IV_I]], align 4
-// IR-NEXT:    br label %[[FOR_COND]], !llvm.loop ![[LOOP2:[0-9]+]]
+// IR-NEXT:    br label %[[FOR_COND]], !llvm.loop ![[LOOP3:[0-9]+]]
 // IR-EMPTY:
 // IR-NEXT:  [[FOR_END]]:
 // IR-NEXT:    br label %[[OMP_BODY_CONTINUE:.+]]
@@ -168,19 +168,19 @@ extern "C" void body(...) {}
 // IR-EMPTY:
 // IR-NEXT:  [[OMP_INNER_FOR_INC]]:
 // IR-NEXT:    %[[TMP35:.+]] = load i32, ptr %[[DOTOMP_IV]], align 4
-// IR-NEXT:    %[[ADD29:.+]] = add i32 %[[TMP35]], 1
-// IR-NEXT:    store i32 %[[ADD29]], ptr %[[DOTOMP_IV]], align 4
+// IR-NEXT:    %[[ADD28:.+]] = add i32 %[[TMP35]], 1
+// IR-NEXT:    store i32 %[[ADD28]], ptr %[[DOTOMP_IV]], align 4
 // IR-NEXT:    br label %[[OMP_INNER_FOR_COND]]
 // IR-EMPTY:
 // IR-NEXT:  [[OMP_INNER_FOR_END]]:
 // IR-NEXT:    br label %[[OMP_LOOP_EXIT:.+]]
 // IR-EMPTY:
 // IR-NEXT:  [[OMP_LOOP_EXIT]]:
-// IR-NEXT:    call void @__kmpc_for_static_fini(ptr @1, i32 %[[TMP0]])
+// IR-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 %[[TMP0]])
 // IR-NEXT:    br label %[[OMP_PRECOND_END]]
 // IR-EMPTY:
 // IR-NEXT:  [[OMP_PRECOND_END]]:
-// IR-NEXT:    call void @__kmpc_barrier(ptr @3, i32 %[[TMP0]])
+// IR-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB3:.+]], i32 %[[TMP0]])
 // IR-NEXT:    ret void
 // IR-NEXT:  }
 extern "C" void func(int start, int end, int step) {
diff --git a/clang/test/OpenMP/tile_codegen_tile_for.cpp b/clang/test/OpenMP/tile_codegen_tile_for.cpp
index d0fb89398c241..91536c406368b 100644
--- a/clang/test/OpenMP/tile_codegen_tile_for.cpp
+++ b/clang/test/OpenMP/tile_codegen_tile_for.cpp
@@ -16,7 +16,7 @@
 extern "C" void body(...) {}
 
 
-// IR-LABEL: @func(
+// IR-LABEL: define {{.*}}@func(
 // IR-NEXT:  [[ENTRY:.*]]:
 // IR-NEXT:    %[[START_ADDR:.+]] = alloca i32, align 4
 // IR-NEXT:    %[[END_ADDR:.+]] = alloca i32, align 4
@@ -26,22 +26,22 @@ extern "C" void body(...) {}
 // IR-NEXT:    %[[I:.+]] = alloca i32, align 4
 // IR-NEXT:    %[[DOTCAPTURE_EXPR_:.+]] = alloca i32, align 4
 // IR-NEXT:    %[[DOTCAPTURE_EXPR_1:.+]] = alloca i32, align 4
+// IR-NEXT:    %[[DOTNEW_STEP:.+]] = alloca i32, align 4
 // IR-NEXT:    %[[DOTCAPTURE_EXPR_2:.+]] = alloca i32, align 4
-// IR-NEXT:    %[[DOTCAPTURE_EXPR_3:.+]] = alloca i32, align 4
 // IR-NEXT:    %[[DOTFLOOR_0_IV_I:.+]] = alloca i32, align 4
-// IR-NEXT:    %[[DOTCAPTURE_EXPR_6:.+]] = alloca i32, align 4
-// IR-NEXT:    %[[DOTCAPTURE_EXPR_8:.+]] = alloca i32, align 4
-// IR-NEXT:    %[[DOTCAPTURE_EXPR_12:.+]] = alloca i32, align 4
-// IR-NEXT:    %[[DOTCAPTURE_EXPR_14:.+]] = alloca i32, align 4
+// IR-NEXT:    %[[DOTCAPTURE_EXPR_5:.+]] = alloca i32, align 4
+// IR-NEXT:    %[[DOTCAPTURE_EXPR_7:.+]] = alloca i32, align 4
+// IR-NEXT:    %[[DOTCAPTURE_EXPR_11:.+]] = alloca i32, align 4
+// IR-NEXT:    %[[DOTCAPTURE_EXPR_13:.+]] = alloca i32, align 4
 // IR-NEXT:    %[[DOTFLOOR_0_IV__FLOOR_0_IV_I:.+]] = alloca i32, align 4
 // IR-NEXT:    %[[DOTOMP_LB:.+]] = alloca i32, align 4
 // IR-NEXT:    %[[DOTOMP_UB:.+]] = alloca i32, align 4
 // IR-NEXT:    %[[DOTOMP_STRIDE:.+]] = alloca i32, align 4
 // IR-NEXT:    %[[DOTOMP_IS_LAST:.+]] = alloca i32, align 4
-// IR-NEXT:    %[[DOTFLOOR_0_IV__FLOOR_0_IV_I18:.+]] = alloca i32, align 4
+// IR-NEXT:    %[[DOTFLOOR_0_IV__FLOOR_0_IV_I17:.+]] = alloca i32, align 4
 // IR-NEXT:    %[[DOTTILE_0_IV__FLOOR_0_IV_I:.+]] = alloca i32, align 4
 // IR-NEXT:    %[[DOTTILE_0_IV_I:.+]] = alloca i32, align 4
-// IR-NEXT:    %[[TMP0:.+]] = call i32 @__kmpc_global_thread_num(ptr @2)
+// IR-NEXT:    %[[TMP0:.+]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB2:.+]])
 // IR-NEXT:    store i32 %[[START:.+]], ptr %[[START_ADDR]], align 4
 // IR-NEXT:    store i32 %[[END:.+]], ptr %[[END_ADDR]], align 4
 // IR-NEXT:    store i32 %[[STEP:.+]], ptr %[[STEP_ADDR]], align 4
@@ -52,53 +52,53 @@ extern "C" void body(...) {}
 // IR-NEXT:    %[[TMP3:.+]] = load i32, ptr %[[END_ADDR]], align 4
 // IR-NEXT:    store i32 %[[TMP3]], ptr %[[DOTCAPTURE_EXPR_1]], align 4
 // IR-NEXT:    %[[TMP4:.+]] = load i32, ptr %[[STEP_ADDR]], align 4
-// IR-NEXT:    store i32 %[[TMP4]], ptr %[[DOTCAPTURE_EXPR_2]], align 4
+// IR-NEXT:    store i32 %[[TMP4]], ptr %[[DOTNEW_STEP]], align 4
 // IR-NEXT:    %[[TMP5:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_1]], align 4
 // IR-NEXT:    %[[TMP6:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_]], align 4
 // IR-NEXT:    %[[SUB:.+]] = sub i32 %[[TMP5]], %[[TMP6]]
-// IR-NEXT:    %[[SUB4:.+]] = sub i32 %[[SUB]], 1
-// IR-NEXT:    %[[TMP7:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_2]], align 4
-// IR-NEXT:    %[[ADD:.+]] = add i32 %[[SUB4]], %[[TMP7]]
-// IR-NEXT:    %[[TMP8:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_2]], align 4
+// IR-NEXT:    %[[SUB3:.+]] = sub i32 %[[SUB]], 1
+// IR-NEXT:    %[[TMP7:.+]] = load i32, ptr %[[DOTNEW_STEP]], align 4
+// IR-NEXT:    %[[ADD:.+]] = add i32 %[[SUB3]], %[[TMP7]]
+// IR-NEXT:    %[[TMP8:.+]] = load i32, ptr %[[DOTNEW_STEP]], align 4
 // IR-NEXT:    %[[DIV:.+]] = udiv i32 %[[ADD]], %[[TMP8]]
-// IR-NEXT:    %[[SUB5:.+]] = sub i32 %[[DIV]], 1
-// IR-NEXT:    store i32 %[[SUB5]], ptr %[[DOTCAPTURE_EXPR_3]], align 4
+// IR-NEXT:    %[[SUB4:.+]] = sub i32 %[[DIV]], 1
+// IR-NEXT:    store i32 %[[SUB4]], ptr %[[DOTCAPTURE_EXPR_2]], align 4
 // IR-NEXT:    store i32 0, ptr %[[DOTFLOOR_0_IV_I]], align 4
-// IR-NEXT:    %[[TMP9:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_3]], align 4
-// IR-NEXT:    %[[ADD7:.+]] = add i32 %[[TMP9]], 1
-// IR-NEXT:    store i32 %[[ADD7]], ptr %[[DOTCAPTURE_EXPR_6]], align 4
-// IR-NEXT:    %[[TMP10:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_6]], align 4
-// IR-NEXT:    %[[SUB9:.+]] = sub i32 %[[TMP10]], -3
-// IR-NEXT:    %[[DIV10:.+]] = udiv i32 %[[SUB9]], 4
-// IR-NEXT:    %[[SUB11:.+]] = sub i32 %[[DIV10]], 1
-// IR-NEXT:    store i32 %[[SUB11]], ptr %[[DOTCAPTURE_EXPR_8]], align 4
-// IR-NEXT:    %[[TMP11:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_8]], align 4
-// IR-NEXT:    %[[ADD13:.+]] = add i32 %[[TMP11]], 1
-// IR-NEXT:    store i32 %[[ADD13]], ptr %[[DOTCAPTURE_EXPR_12]], align 4
-// IR-NEXT:    %[[TMP12:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_12]], align 4
-// IR-NEXT:    %[[SUB15:.+]] = sub i32 %[[TMP12]], -2
-// IR-NEXT:    %[[DIV16:.+]] = udiv i32 %[[SUB15]], 3
-// IR-NEXT:    %[[SUB17:.+]] = sub i32 %[[DIV16]], 1
-// IR-NEXT:    store i32 %[[SUB17]], ptr %[[DOTCAPTURE_EXPR_14]], align 4
+// IR-NEXT:    %[[TMP9:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_2]], align 4
+// IR-NEXT:    %[[ADD6:.+]] = add i32 %[[TMP9]], 1
+// IR-NEXT:    store i32 %[[ADD6]], ptr %[[DOTCAPTURE_EXPR_5]], align 4
+// IR-NEXT:    %[[TMP10:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_5]], align 4
+// IR-NEXT:    %[[SUB8:.+]] = sub i32 %[[TMP10]], -3
+// IR-NEXT:    %[[DIV9:.+]] = udiv i32 %[[SUB8]], 4
+// IR-NEXT:    %[[SUB10:.+]] = sub i32 %[[DIV9]], 1
+// IR-NEXT:    store i32 %[[SUB10]], ptr %[[DOTCAPTURE_EXPR_7]], align 4
+// IR-NEXT:    %[[TMP11:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_7]], align 4
+// IR-NEXT:    %[[ADD12:.+]] = add i32 %[[TMP11]], 1
+// IR-NEXT:    store i32 %[[ADD12]], ptr %[[DOTCAPTURE_EXPR_11]], align 4
+// IR-NEXT:    %[[TMP12:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_11]], align 4
+// IR-NEXT:    %[[SUB14:.+]] = sub i32 %[[TMP12]], -2
+// IR-NEXT:    %[[DIV15:.+]] = udiv i32 %[[SUB14]], 3
+// IR-NEXT:    %[[SUB16:.+]] = sub i32 %[[DIV15]], 1
+// IR-NEXT:    store i32 %[[SUB16]], ptr %[[DOTCAPTURE_EXPR_13]], align 4
 // IR-NEXT:    store i32 0, ptr %[[DOTFLOOR_0_IV__FLOOR_0_IV_I]], align 4
-// IR-NEXT:    %[[TMP13:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_12]], align 4
+// IR-NEXT:    %[[TMP13:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_11]], align 4
 // IR-NEXT:    %[[CMP:.+]] = icmp ult i32 0, %[[TMP13]]
 // IR-NEXT:    br i1 %[[CMP]], label %[[OMP_PRECOND_THEN:.+]], label %[[OMP_PRECOND_END:.+]]
 // IR-EMPTY:
 // IR-NEXT:  [[OMP_PRECOND_THEN]]:
 // IR-NEXT:    store i32 0, ptr %[[DOTOMP_LB]], align 4
-// IR-NEXT:    %[[TMP14:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_14]], align 4
+// IR-NEXT:    %[[TMP14:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_13]], align 4
 // IR-NEXT:    store i32 %[[TMP14]], ptr %[[DOTOMP_UB]], align 4
 // IR-NEXT:    store i32 1, ptr %[[DOTOMP_STRIDE]], align 4
 // IR-NEXT:    store i32 0, ptr %[[DOTOMP_IS_LAST]], align 4
-// IR-NEXT:    call void @__kmpc_for_static_init_4u(ptr @1, i32 %[[TMP0]], i32 34, ptr %[[DOTOMP_IS_LAST]], ptr %[[DOTOMP_LB]], ptr %[[DOTOMP_UB]], ptr %[[DOTOMP_STRIDE]], i32 1, i32 1)
+// IR-NEXT:    call void @__kmpc_for_static_init_4u(ptr @[[GLOB1:.+]], i32 %[[TMP0]], i32 34, ptr %[[DOTOMP_IS_LAST]], ptr %[[DOTOMP_LB]], ptr %[[DOTOMP_UB]], ptr %[[DOTOMP_STRIDE]], i32 1, i32 1)
 // IR-NEXT:    %[[TMP15:.+]] = load i32, ptr %[[DOTOMP_UB]], align 4
-// IR-NEXT:    %[[TMP16:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_14]], align 4
-// IR-NEXT:    %[[CMP19:.+]] = icmp ugt i32 %[[TMP15]], %[[TMP16]]
-// IR-NEXT:    br i1 %[[CMP19]], label %[[COND_TRUE:.+]], label %[[COND_FALSE:.+]]
+// IR-NEXT:    %[[TMP16:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_13]], align 4
+// IR-NEXT:    %[[CMP18:.+]] = icmp ugt i32 %[[TMP15]], %[[TMP16]]
+// IR-NEXT:    br i1 %[[CMP18]], label %[[COND_TRUE:.+]], label %[[COND_FALSE:.+]]
 // IR-EMPTY:
 // IR-NEXT:  [[COND_TRUE]]:
-// IR-NEXT:    %[[TMP17:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_14]], align 4
+// IR-NEXT:    %[[TMP17:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_13]], align 4
 // IR-NEXT:    br label %[[COND_END:.+]]
 // IR-EMPTY:
 // IR-NEXT:  [[COND_FALSE]]:
@@ -115,83 +115,83 @@ extern "C" void body(...) {}
 // IR-NEXT:  [[OMP_INNER_FOR_COND]]:
 // IR-NEXT:    %[[TMP20:.+]] = load i32, ptr %[[DOTOMP_IV]], align 4
 // IR-NEXT:    %[[TMP21:.+]] = load i32, ptr %[[DOTOMP_UB]], align 4
-// IR-NEXT:    %[[ADD20:.+]] = add i32 %[[TMP21]], 1
-// IR-NEXT:    %[[CMP21:.+]] = icmp ult i32 %[[TMP20]], %[[ADD20]]
-// IR-NEXT:    br i1 %[[CMP21]], label %[[OMP_INNER_FOR_BODY:.+]], label %[[OMP_INNER_FOR_END:.+]]
+// IR-NEXT:    %[[ADD19:.+]] = add i32 %[[TMP21]], 1
+// IR-NEXT:    %[[CMP20:.+]] = icmp ult i32 %[[TMP20]], %[[ADD19]]
+// IR-NEXT:    br i1 %[[CMP20]], label %[[OMP_INNER_FOR_BODY:.+]], label %[[OMP_INNER_FOR_END:.+]]
 // IR-EMPTY:
 // IR-NEXT:  [[OMP_INNER_FOR_BODY]]:
 // IR-NEXT:    %[[TMP22:.+]] = load i32, ptr %[[DOTOMP_IV]], align 4
 // IR-NEXT:    %[[MUL:.+]] = mul i32 %[[TMP22]], 3
-// IR-NEXT:    %[[ADD22:.+]] = add i32 0, %[[MUL]]
-// IR-NEXT:    store i32 %[[ADD22]], ptr %[[DOTFLOOR_0_IV__FLOOR_0_IV_I18]], align 4
-// IR-NEXT:    %[[TMP23:.+]] = load i32, ptr %[[DOTFLOOR_0_IV__FLOOR_0_IV_I18]], align 4
+// IR-NEXT:    %[[ADD21:.+]] = add i32 0, %[[MUL]]
+// IR-NEXT:    store i32 %[[ADD21]], ptr %[[DOTFLOOR_0_IV__FLOOR_0_IV_I17]], align 4
+// IR-NEXT:    %[[TMP23:.+]] = load i32, ptr %[[DOTFLOOR_0_IV__FLOOR_0_IV_I17]], align 4
 // IR-NEXT:    store i32 %[[TMP23]], ptr %[[DOTTILE_0_IV__FLOOR_0_IV_I]], align 4
 // IR-NEXT:    br label %[[FOR_COND:.+]]
 // IR-EMPTY:
 // IR-NEXT:  [[FOR_COND]]:
 // IR-NEXT:    %[[TMP24:.+]] = load i32, ptr %[[DOTTILE_0_IV__FLOOR_0_IV_I]], align 4
-// IR-NEXT:    %[[TMP25:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_8]], align 4
-// IR-NEXT:    %[[ADD23:.+]] = add i32 %[[TMP25]], 1
-// IR-NEXT:    %[[TMP26:.+]] = load i32, ptr %[[DOTFLOOR_0_IV__FLOOR_0_IV_I18]], align 4
-// IR-NEXT:    %[[ADD24:.+]] = add i32 %[[TMP26]], 3
-// IR-NEXT:    %[[CMP25:.+]] = icmp ult i32 %[[ADD23]], %[[ADD24]]
-// IR-NEXT:    br i1 %[[CMP25]], label %[[COND_TRUE26:.+]], label %[[COND_FALSE28:.+]]
-// IR-EMPTY:
-// IR-NEXT:  [[COND_TRUE26]]:
-// IR-NEXT:    %[[TMP27:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_8]], align 4
-// IR-NEXT:    %[[ADD27:.+]] = add i32 %[[TMP27]], 1
-// IR-NEXT:    br label %[[COND_END30:.+]]
-// IR-EMPTY:
-// IR-NEXT:  [[COND_FALSE28]]:
-// IR-NEXT:    %[[TMP28:.+]] = load i32, ptr %[[DOTFLOOR_0_IV__FLOOR_0_IV_I18]], align 4
-// IR-NEXT:    %[[ADD29:.+]] = add i32 %[[TMP28]], 3
-// IR-NEXT:    br label %[[COND_END30]]
-// IR-EMPTY:
-// IR-NEXT:  [[COND_END30]]:
-// IR-NEXT:    %[[COND31:.+]] = phi i32 [ %[[ADD27]], %[[COND_TRUE26]] ], [ %[[ADD29]], %[[COND_FALSE28]] ]
-// IR-NEXT:    %[[CMP32:.+]] = icmp ult i32 %[[TMP24]], %[[COND31]]
-// IR-NEXT:    br i1 %[[CMP32]], label %[[FOR_BODY:.+]], label %[[FOR_END51:.+]]
+// IR-NEXT:    %[[TMP25:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_7]], align 4
+// IR-NEXT:    %[[ADD22:.+]] = add i32 %[[TMP25]], 1
+// IR-NEXT:    %[[TMP26:.+]] = load i32, ptr %[[DOTFLOOR_0_IV__FLOOR_0_IV_I17]], align 4
+// IR-NEXT:    %[[ADD23:.+]] = add i32 %[[TMP26]], 3
+// IR-NEXT:    %[[CMP24:.+]] = icmp ult i32 %[[ADD22]], %[[ADD23]]
+// IR-NEXT:    br i1 %[[CMP24]], label %[[COND_TRUE25:.+]], label %[[COND_FALSE27:.+]]
+// IR-EMPTY:
+// IR-NEXT:  [[COND_TRUE25]]:
+// IR-NEXT:    %[[TMP27:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_7]], align 4
+// IR-NEXT:    %[[ADD26:.+]] = add i32 %[[TMP27]], 1
+// IR-NEXT:    br label %[[COND_END29:.+]]
+// IR-EMPTY:
+// IR-NEXT:  [[COND_FALSE27]]:
+// IR-NEXT:    %[[TMP28:.+]] = load i32, ptr %[[DOTFLOOR_0_IV__FLOOR_0_IV_I17]], align 4
+// IR-NEXT:    %[[ADD28:.+]] = add i32 %[[TMP28]], 3
+// IR-NEXT:    br label %[[COND_END29]]
+// IR-EMPTY:
+// IR-NEXT:  [[COND_END29]]:
+// IR-NEXT:    %[[COND30:.+]] = phi i32 [ %[[ADD26]], %[[COND_TRUE25]] ], [ %[[ADD28]], %[[COND_FALSE27]] ]
+// IR-NEXT:    %[[CMP31:.+]] = icmp ult i32 %[[TMP24]], %[[COND30]]
+// IR-NEXT:    br i1 %[[CMP31]], label %[[FOR_BODY:.+]], label %[[FOR_END50:.+]]
 // IR-EMPTY:
 // IR-NEXT:  [[FOR_BODY]]:
 // IR-NEXT:    %[[TMP29:.+]] = load i32, ptr %[[DOTTILE_0_IV__FLOOR_0_IV_I]], align 4
-// IR-NEXT:    %[[MUL33:.+]] = mul i32 %[[TMP29]], 4
-// IR-NEXT:    %[[ADD34:.+]] = add i32 0, %[[MUL33]]
-// IR-NEXT:    store i32 %[[ADD34]], ptr %[[DOTFLOOR_0_IV_I]], align 4
+// IR-NEXT:    %[[MUL32:.+]] = mul i32 %[[TMP29]], 4
+// IR-NEXT:    %[[ADD33:.+]] = add i32 0, %[[MUL32]]
+// IR-NEXT:    store i32 %[[ADD33]], ptr %[[DOTFLOOR_0_IV_I]], align 4
 // IR-NEXT:    %[[TMP30:.+]] = load i32, ptr %[[DOTFLOOR_0_IV_I]], align 4
 // IR-NEXT:    store i32 %[[TMP30]], ptr %[[DOTTILE_0_IV_I]], align 4
-// IR-NEXT:    br label %[[FOR_COND35:.+]]
+// IR-NEXT:    br label %[[FOR_COND34:.+]]
 // IR-EMPTY:
-// IR-NEXT:  [[FOR_COND35]]:
+// IR-NEXT:  [[FOR_COND34]]:
 // IR-NEXT:    %[[TMP31:.+]] = load i32, ptr %[[DOTTILE_0_IV_I]], align 4
-// IR-NEXT:    %[[TMP32:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_3]], align 4
-// IR-NEXT:    %[[ADD36:.+]] = add i32 %[[TMP32]], 1
+// IR-NEXT:    %[[TMP32:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_2]], align 4
+// IR-NEXT:    %[[ADD35:.+]] = add i32 %[[TMP32]], 1
 // IR-NEXT:    %[[TMP33:.+]] = load i32, ptr %[[DOTFLOOR_0_IV_I]], align 4
-// IR-NEXT:    %[[ADD37:.+]] = add nsw i32 %[[TMP33]], 4
-// IR-NEXT:    %[[CMP38:.+]] = icmp ult i32 %[[ADD36]], %[[ADD37]]
-// IR-NEXT:    br i1 %[[CMP38]], label %[[COND_TRUE39:.+]], label %[[COND_FALSE41:.+]]
+// IR-NEXT:    %[[ADD36:.+]] = add i32 %[[TMP33]], 4
+// IR-NEXT:    %[[CMP37:.+]] = icmp ult i32 %[[ADD35]], %[[ADD36]]
+// IR-NEXT:    br i1 %[[CMP37]], label %[[COND_TRUE38:.+]], label %[[COND_FALSE40:.+]]
 // IR-EMPTY:
-// IR-NEXT:  [[COND_TRUE39]]:
-// IR-NEXT:    %[[TMP34:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_3]], align 4
-// IR-NEXT:    %[[ADD40:.+]] = add i32 %[[TMP34]], 1
-// IR-NEXT:    br label %[[COND_END43:.+]]
+// IR-NEXT:  [[COND_TRUE38]]:
+// IR-NEXT:    %[[TMP34:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_2]], align 4
+// IR-NEXT:    %[[ADD39:.+]] = add i32 %[[TMP34]], 1
+// IR-NEXT:    br label %[[COND_END42:.+]]
 // IR-EMPTY:
-// IR-NEXT:  [[COND_FALSE41]]:
+// IR-NEXT:  [[COND_FALSE40]]:
 // IR-NEXT:    %[[TMP35:.+]] = load i32, ptr %[[DOTFLOOR_0_IV_I]], align 4
-// IR-NEXT:    %[[ADD42:.+]] = add nsw i32 %[[TMP35]], 4
-// IR-NEXT:    br label %[[COND_END43]]
+// IR-NEXT:    %[[ADD41:.+]] = add i32 %[[TMP35]], 4
+// IR-NEXT:    br label %[[COND_END42]]
 // IR-EMPTY:
-// IR-NEXT:  [[COND_END43]]:
-// IR-NEXT:    %[[COND44:.+]] = phi i32 [ %[[ADD40]], %[[COND_TRUE39]] ], [ %[[ADD42]], %[[COND_FALSE41]] ]
-// IR-NEXT:    %[[CMP45:.+]] = icmp ult i32 %[[TMP31]], %[[COND44]]
-// IR-NEXT:    br i1 %[[CMP45]], label %[[FOR_BODY46:.+]], label %[[FOR_END:.+]]
+// IR-NEXT:  [[COND_END42]]:
+// IR-NEXT:    %[[COND43:.+]] = phi i32 [ %[[ADD39]], %[[COND_TRUE38]] ], [ %[[ADD41]], %[[COND_FALSE40]] ]
+// IR-NEXT:    %[[CMP44:.+]] = icmp ult i32 %[[TMP31]], %[[COND43]]
+// IR-NEXT:    br i1 %[[CMP44]], label %[[FOR_BODY45:.+]], label %[[FOR_END:.+]]
 // IR-EMPTY:
-// IR-NEXT:  [[FOR_BODY46]]:
+// IR-NEXT:  [[FOR_BODY45]]:
 // IR-NEXT:    %[[TMP36:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_]], align 4
 // IR-NEXT:    %[[TMP37:.+]] = load i32, ptr %[[DOTTILE_0_IV_I]], align 4
-// IR-NEXT:    %[[TMP38:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_2]], align 4
-// IR-NEXT:    %[[MUL47:.+]] = mul i32 %[[TMP37]], %[[TMP38]]
-// IR-NEXT:    %[[ADD48:.+]] = add i32 %[[TMP36]], %[[MUL47]]
-// IR-NEXT:    store i32 %[[ADD48]], ptr %[[I]], align 4
+// IR-NEXT:    %[[TMP38:.+]] = load i32, ptr %[[DOTNEW_STEP]], align 4
+// IR-NEXT:    %[[MUL46:.+]] = mul i32 %[[TMP37]], %[[TMP38]]
+// IR-NEXT:    %[[ADD47:.+]] = add i32 %[[TMP36]], %[[MUL46]]
+// IR-NEXT:    store i32 %[[ADD47]], ptr %[[I]], align 4
 // IR-NEXT:    %[[TMP39:.+]] = load i32, ptr %[[START_ADDR]], align 4
 // IR-NEXT:    %[[TMP40:.+]] = load i32, ptr %[[END_ADDR]], align 4
 // IR-NEXT:    %[[TMP41:.+]] = load i32, ptr %[[STEP_ADDR]], align 4
@@ -201,20 +201,20 @@ extern "C" void body(...) {}
 // IR-EMPTY:
 // IR-NEXT:  [[FOR_INC]]:
 // IR-NEXT:    %[[TMP43:.+]] = load i32, ptr %[[DOTTILE_0_IV_I]], align 4
-// IR-NEXT:    %[[INC:.+]] = add nsw i32 %[[TMP43]], 1
+// IR-NEXT:    %[[INC:.+]] = add i32 %[[TMP43]], 1
 // IR-NEXT:    store i32 %[[INC]], ptr %[[DOTTILE_0_IV_I]], align 4
-// IR-NEXT:    br label %[[FOR_COND35]], !llvm.loop ![[LOOP2:[0-9]+]]
+// IR-NEXT:    br label %[[FOR_COND34]], !llvm.loop ![[LOOP3:[0-9]+]]
 // IR-EMPTY:
 // IR-NEXT:  [[FOR_END]]:
-// IR-NEXT:    br label %[[FOR_INC49:.+]]
+// IR-NEXT:    br label %[[FOR_INC48:.+]]
 // IR-EMPTY:
-// IR-NEXT:  [[FOR_INC49]]:
+// IR-NEXT:  [[FOR_INC48]]:
 // IR-NEXT:    %[[TMP44:.+]] = load i32, ptr %[[DOTTILE_0_IV__FLOOR_0_IV_I]], align 4
-// IR-NEXT:    %[[INC50:.+]] = add i32 %[[TMP44]], 1
-// IR-NEXT:    store i32 %[[INC50]], ptr %[[DOTTILE_0_IV__FLOOR_0_IV_I]], align 4
-// IR-NEXT:    br label %[[FOR_COND]], !llvm.loop ![[LOOP4:[0-9]+]]
+// IR-NEXT:    %[[INC49:.+]] = add i32 %[[TMP44]], 1
+// IR-NEXT:    store i32 %[[INC49]], ptr %[[DOTTILE_0_IV__FLOOR_0_IV_I]], align 4
+// IR-NEXT:    br label %[[FOR_COND]], !llvm.loop ![[LOOP5:[0-9]+]]
 // IR-EMPTY:
-// IR-NEXT:  [[FOR_END51]]:
+// IR-NEXT:  [[FOR_END50]]:
 // IR-NEXT:    br label %[[OMP_BODY_CONTINUE:.+]]
 // IR-EMPTY:
 // IR-NEXT:  [[OMP_BODY_CONTINUE]]:
@@ -222,21 +222,23 @@ extern "C" void body(...) {}
 // IR-EMPTY:
 // IR-NEXT:  [[OMP_INNER_FOR_INC]]:
 // IR-NEXT:    %[[TMP45:.+]] = load i32, ptr %[[DOTOMP_IV]], align 4
-// IR-NEXT:    %[[ADD52:.+]] = add i32 %[[TMP45]], 1
-// IR-NEXT:    store i32 %[[ADD52]], ptr %[[DOTOMP_IV]], align 4
+// IR-NEXT:    %[[ADD51:.+]] = add i32 %[[TMP45]], 1
+// IR-NEXT:    store i32 %[[ADD51]], ptr %[[DOTOMP_IV]], align 4
 // IR-NEXT:    br label %[[OMP_INNER_FOR_COND]]
 // IR-EMPTY:
 // IR-NEXT:  [[OMP_INNER_FOR_END]]:
 // IR-NEXT:    br label %[[OMP_LOOP_EXIT:.+]]
 // IR-EMPTY:
 // IR-NEXT:  [[OMP_LOOP_EXIT]]:
-// IR-NEXT:    call void @__kmpc_for_static_fini(ptr @1, i32 %[[TMP0]])
+// IR-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 %[[TMP0]])
 // IR-NEXT:    br label %[[OMP_PRECOND_END]]
 // IR-EMPTY:
 // IR-NEXT:  [[OMP_PRECOND_END]]:
-// IR-NEXT:    call void @__kmpc_barrier(ptr @3, i32 %[[TMP0]])
+// IR-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB3:.+]], i32 %[[TMP0]])
 // IR-NEXT:    ret void
 // IR-NEXT:  }
+
+
 extern "C" void func(int start, int end, int step) {
 #pragma omp for
 #pragma omp tile sizes(3)
@@ -246,8 +248,10 @@ extern "C" void func(int start, int end, int step) {
 }
 
 #endif /* HEADER */
+
 // IR: ![[META0:[0-9]+]] = !{i32 1, !"wchar_size", i32 4}
-// IR: ![[META1:[0-9]+]] = !{!"{{[^"]*}}"}
-// IR: ![[LOOP2]] = distinct !{![[LOOP2]], ![[LOOPPROP3:[0-9]+]]}
-// IR: ![[LOOPPROP3]] = !{!"llvm.loop.mustprogress"}
-// IR: ![[LOOP4]] = distinct !{![[LOOP4]], ![[LOOPPROP3]]}
+// IR: ![[META1:[0-9]+]] = !{i32 7, !"openmp", i32 51}
+// IR: ![[META2:[0-9]+]] =
+// IR: ![[LOOP3]] = distinct !{![[LOOP3]], ![[LOOPPROP4:[0-9]+]]}
+// IR: ![[LOOPPROP4]] = !{!"llvm.loop.mustprogress"}
+// IR: ![[LOOP5]] = distinct !{![[LOOP5]], ![[LOOPPROP4]]}
diff --git a/openmp/runtime/test/lit.cfg b/openmp/runtime/test/lit.cfg
index e8f7f3470580e..14c7468982137 100644
--- a/openmp/runtime/test/lit.cfg
+++ b/openmp/runtime/test/lit.cfg
@@ -171,10 +171,14 @@ config.substitutions.append(("%libomp-c99-compile-and-run", \
     "%libomp-c99-compile && %libomp-run"))
 config.substitutions.append(("%libomp-cxx-compile-and-run", \
     "%libomp-cxx-compile && %libomp-run"))
+config.substitutions.append(("%libomp-cxx20-compile-and-run", \
+    "%libomp-cxx20-compile && %libomp-run"))
 config.substitutions.append(("%libomp-cxx-compile-c", \
     "%clangXX %openmp_flags %flags -std=c++17 -x c++ %s -o %t" + libs))
 config.substitutions.append(("%libomp-cxx-compile", \
     "%clangXX %openmp_flags %flags -std=c++17 %s -o %t" + libs))
+config.substitutions.append(("%libomp-cxx20-compile", \
+    "%clangXX %openmp_flags %flags -std=c++20 %s -o %t" + libs))
 config.substitutions.append(("%libomp-compile", \
     "%clang %openmp_flags %flags %s -o %t" + libs))
 config.substitutions.append(("%libomp-irbuilder-compile", \
diff --git a/openmp/runtime/test/transform/tile/foreach.cpp b/openmp/runtime/test/transform/tile/foreach.cpp
new file mode 100644
index 0000000000000..4fb3595760974
--- /dev/null
+++ b/openmp/runtime/test/transform/tile/foreach.cpp
@@ -0,0 +1,228 @@
+// 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 tile sizes(2, 2)
+  for (Reporter c{"C"}; auto &&v : Reporter("A"))
+    for (Reporter d{"D"}; auto &&w : Reporter("B"))
+      printf("v=%d w=%d\n", v, w);
+  printf("done\n");
+  return EXIT_SUCCESS;
+}
+
+#endif /* HEADER */
+
+// CHECK:      do
+// CHECK-NEXT: [C] ctor
+// CHECK-NEXT: [A] ctor
+// CHECK-NEXT: [A] end()
+// CHECK-NEXT: [A] begin()
+// CHECK-NEXT: [A] begin()
+// CHECK-NEXT: [A] iterator distance: 3
+// CHECK-NEXT: [D] ctor
+// CHECK-NEXT: [B] ctor
+// CHECK-NEXT: [B] end()
+// CHECK-NEXT: [B] begin()
+// CHECK-NEXT: [B] begin()
+// CHECK-NEXT: [B] iterator distance: 3
+// CHECK-NEXT: [A] iterator advance: 0 += 0
+// CHECK-NEXT: [A] iterator move assign
+// CHECK-NEXT: [A] iterator deref: 0
+// CHECK-NEXT: [B] iterator advance: 0 += 0
+// CHECK-NEXT: [B] iterator move assign
+// CHECK-NEXT: [B] iterator deref: 0
+// CHECK-NEXT: v=0 w=0
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [B] iterator advance: 0 += 1
+// CHECK-NEXT: [B] iterator move assign
+// CHECK-NEXT: [B] iterator deref: 1
+// CHECK-NEXT: v=0 w=1
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [A] iterator dtor
+// CHECK-NEXT: [A] iterator advance: 0 += 1
+// CHECK-NEXT: [A] iterator move assign
+// CHECK-NEXT: [A] iterator deref: 1
+// CHECK-NEXT: [B] iterator advance: 0 += 0
+// CHECK-NEXT: [B] iterator move assign
+// CHECK-NEXT: [B] iterator deref: 0
+// CHECK-NEXT: v=1 w=0
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [B] iterator advance: 0 += 1
+// CHECK-NEXT: [B] iterator move assign
+// CHECK-NEXT: [B] iterator deref: 1
+// CHECK-NEXT: v=1 w=1
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [A] iterator dtor
+// CHECK-NEXT: [A] iterator advance: 0 += 0
+// CHECK-NEXT: [A] iterator move assign
+// CHECK-NEXT: [A] iterator deref: 0
+// CHECK-NEXT: [B] iterator advance: 0 += 2
+// CHECK-NEXT: [B] iterator move assign
+// CHECK-NEXT: [B] iterator deref: 2
+// CHECK-NEXT: v=0 w=2
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [A] iterator dtor
+// CHECK-NEXT: [A] iterator advance: 0 += 1
+// CHECK-NEXT: [A] iterator move assign
+// CHECK-NEXT: [A] iterator deref: 1
+// CHECK-NEXT: [B] iterator advance: 0 += 2
+// CHECK-NEXT: [B] iterator move assign
+// CHECK-NEXT: [B] iterator deref: 2
+// CHECK-NEXT: v=1 w=2
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [A] iterator dtor
+// CHECK-NEXT: [A] iterator advance: 0 += 2
+// CHECK-NEXT: [A] iterator move assign
+// CHECK-NEXT: [A] iterator deref: 2
+// CHECK-NEXT: [B] iterator advance: 0 += 0
+// CHECK-NEXT: [B] iterator move assign
+// CHECK-NEXT: [B] iterator deref: 0
+// CHECK-NEXT: v=2 w=0
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [B] iterator advance: 0 += 1
+// CHECK-NEXT: [B] iterator move assign
+// CHECK-NEXT: [B] iterator deref: 1
+// CHECK-NEXT: v=2 w=1
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [A] iterator dtor
+// CHECK-NEXT: [A] iterator advance: 0 += 2
+// CHECK-NEXT: [A] iterator move assign
+// CHECK-NEXT: [A] iterator deref: 2
+// CHECK-NEXT: [B] iterator advance: 0 += 2
+// CHECK-NEXT: [B] iterator move assign
+// CHECK-NEXT: [B] iterator deref: 2
+// CHECK-NEXT: v=2 w=2
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [A] iterator dtor
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [B] dtor
+// CHECK-NEXT: [D] dtor
+// CHECK-NEXT: [A] iterator dtor
+// CHECK-NEXT: [A] iterator dtor
+// CHECK-NEXT: [A] iterator dtor
+// CHECK-NEXT: [A] dtor
+// CHECK-NEXT: [C] dtor
+// CHECK-NEXT: done
diff --git a/openmp/runtime/test/transform/tile/iterfor.cpp b/openmp/runtime/test/transform/tile/iterfor.cpp
new file mode 100644
index 0000000000000..12613544f6e5b
--- /dev/null
+++ b/openmp/runtime/test/transform/tile/iterfor.cpp
@@ -0,0 +1,233 @@
+// 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 A("A"), B("B");
+#pragma omp tile sizes(2, 2)
+    for (auto it = A.begin(); it != A.end(); ++it)
+      for (auto jt = B.begin(); jt != B.end(); ++jt)
+        printf("i=%d j=%d\n", *it, *jt);
+  }
+  printf("done\n");
+  return EXIT_SUCCESS;
+}
+
+#endif /* HEADER */
+
+// CHECK:      do
+// CHECK-NEXT: [A] ctor
+// CHECK-NEXT: [B] ctor
+// CHECK-NEXT: [A] begin()
+// CHECK-NEXT: [A] begin()
+// CHECK-NEXT: [A] end()
+// CHECK-NEXT: [A] iterator distance: 3
+// CHECK-NEXT: [B] begin()
+// CHECK-NEXT: [B] begin()
+// CHECK-NEXT: [B] end()
+// CHECK-NEXT: [B] iterator distance: 3
+// CHECK-NEXT: [A] iterator advance: 0 += 0
+// CHECK-NEXT: [A] iterator move assign
+// CHECK-NEXT: [B] iterator advance: 0 += 0
+// CHECK-NEXT: [B] iterator move assign
+// CHECK-NEXT: [A] iterator deref: 0
+// CHECK-NEXT: [B] iterator deref: 0
+// CHECK-NEXT: i=0 j=0
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [B] iterator advance: 0 += 1
+// CHECK-NEXT: [B] iterator move assign
+// CHECK-NEXT: [A] iterator deref: 0
+// CHECK-NEXT: [B] iterator deref: 1
+// CHECK-NEXT: i=0 j=1
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [A] iterator dtor
+// CHECK-NEXT: [A] iterator advance: 0 += 1
+// CHECK-NEXT: [A] iterator move assign
+// CHECK-NEXT: [B] iterator advance: 0 += 0
+// CHECK-NEXT: [B] iterator move assign
+// CHECK-NEXT: [A] iterator deref: 1
+// CHECK-NEXT: [B] iterator deref: 0
+// CHECK-NEXT: i=1 j=0
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [B] iterator advance: 0 += 1
+// CHECK-NEXT: [B] iterator move assign
+// CHECK-NEXT: [A] iterator deref: 1
+// CHECK-NEXT: [B] iterator deref: 1
+// CHECK-NEXT: i=1 j=1
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [A] iterator dtor
+// CHECK-NEXT: [A] iterator advance: 0 += 0
+// CHECK-NEXT: [A] iterator move assign
+// CHECK-NEXT: [B] iterator advance: 0 += 2
+// CHECK-NEXT: [B] iterator move assign
+// CHECK-NEXT: [A] iterator deref: 0
+// CHECK-NEXT: [B] iterator deref: 2
+// CHECK-NEXT: i=0 j=2
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [A] iterator dtor
+// CHECK-NEXT: [A] iterator advance: 0 += 1
+// CHECK-NEXT: [A] iterator move assign
+// CHECK-NEXT: [B] iterator advance: 0 += 2
+// CHECK-NEXT: [B] iterator move assign
+// CHECK-NEXT: [A] iterator deref: 1
+// CHECK-NEXT: [B] iterator deref: 2
+// CHECK-NEXT: i=1 j=2
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [A] iterator dtor
+// CHECK-NEXT: [A] iterator advance: 0 += 2
+// CHECK-NEXT: [A] iterator move assign
+// CHECK-NEXT: [B] iterator advance: 0 += 0
+// CHECK-NEXT: [B] iterator move assign
+// CHECK-NEXT: [A] iterator deref: 2
+// CHECK-NEXT: [B] iterator deref: 0
+// CHECK-NEXT: i=2 j=0
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [B] iterator advance: 0 += 1
+// CHECK-NEXT: [B] iterator move assign
+// CHECK-NEXT: [A] iterator deref: 2
+// CHECK-NEXT: [B] iterator deref: 1
+// CHECK-NEXT: i=2 j=1
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [A] iterator dtor
+// CHECK-NEXT: [A] iterator advance: 0 += 2
+// CHECK-NEXT: [A] iterator move assign
+// CHECK-NEXT: [B] iterator advance: 0 += 2
+// CHECK-NEXT: [B] iterator move assign
+// CHECK-NEXT: [A] iterator deref: 2
+// CHECK-NEXT: [B] iterator deref: 2
+// CHECK-NEXT: i=2 j=2
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [A] iterator dtor
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [A] iterator dtor
+// CHECK-NEXT: [A] iterator dtor
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [A] iterator dtor
+// CHECK-NEXT: [B] dtor
+// CHECK-NEXT: [A] dtor
+// CHECK-NEXT: done
diff --git a/openmp/runtime/test/transform/tile/parallel-wsloop-collapse-foreach.cpp b/openmp/runtime/test/transform/tile/parallel-wsloop-collapse-foreach.cpp
new file mode 100644
index 0000000000000..b1f4d98a52ddc
--- /dev/null
+++ b/openmp/runtime/test/transform/tile/parallel-wsloop-collapse-foreach.cpp
@@ -0,0 +1,366 @@
+// 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);
+    }
+  };
+
+  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 tile sizes(2, 2)
+    for (Reporter c{"C"}; auto &&v : Reporter("A"))
+      for (Reporter d{"D"}; auto &&w : Reporter("B"))
+        printf("i=%d v=%d w=%d\n", i, v, w);
+  printf("done\n");
+  return EXIT_SUCCESS;
+}
+
+#endif /* HEADER */
+
+// CHECK:      do
+// CHECK-NEXT: [C] ctor
+// CHECK-NEXT: [A] ctor
+// CHECK-NEXT: [A] end()
+// CHECK-NEXT: [A] begin()
+// CHECK-NEXT: [A] begin()
+// CHECK-NEXT: [A] iterator distance: 3
+// CHECK-NEXT: [D] ctor
+// CHECK-NEXT: [B] ctor
+// CHECK-NEXT: [B] end()
+// CHECK-NEXT: [B] begin()
+// CHECK-NEXT: [B] begin()
+// CHECK-NEXT: [B] iterator distance: 3
+// CHECK-NEXT: [A] iterator advance: 0 += 0
+// CHECK-NEXT: [A] iterator move assign
+// CHECK-NEXT: [A] iterator deref: 0
+// CHECK-NEXT: [B] iterator advance: 0 += 0
+// CHECK-NEXT: [B] iterator move assign
+// CHECK-NEXT: [B] iterator deref: 0
+// CHECK-NEXT: i=0 v=0 w=0
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [B] iterator advance: 0 += 1
+// CHECK-NEXT: [B] iterator move assign
+// CHECK-NEXT: [B] iterator deref: 1
+// CHECK-NEXT: i=0 v=0 w=1
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [A] iterator dtor
+// CHECK-NEXT: [A] iterator advance: 0 += 1
+// CHECK-NEXT: [A] iterator move assign
+// CHECK-NEXT: [A] iterator deref: 1
+// CHECK-NEXT: [B] iterator advance: 0 += 0
+// CHECK-NEXT: [B] iterator move assign
+// CHECK-NEXT: [B] iterator deref: 0
+// CHECK-NEXT: i=0 v=1 w=0
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [B] iterator advance: 0 += 1
+// CHECK-NEXT: [B] iterator move assign
+// CHECK-NEXT: [B] iterator deref: 1
+// CHECK-NEXT: i=0 v=1 w=1
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [A] iterator dtor
+// CHECK-NEXT: [A] iterator advance: 0 += 0
+// CHECK-NEXT: [A] iterator move assign
+// CHECK-NEXT: [A] iterator deref: 0
+// CHECK-NEXT: [B] iterator advance: 0 += 2
+// CHECK-NEXT: [B] iterator move assign
+// CHECK-NEXT: [B] iterator deref: 2
+// CHECK-NEXT: i=0 v=0 w=2
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [A] iterator dtor
+// CHECK-NEXT: [A] iterator advance: 0 += 1
+// CHECK-NEXT: [A] iterator move assign
+// CHECK-NEXT: [A] iterator deref: 1
+// CHECK-NEXT: [B] iterator advance: 0 += 2
+// CHECK-NEXT: [B] iterator move assign
+// CHECK-NEXT: [B] iterator deref: 2
+// CHECK-NEXT: i=0 v=1 w=2
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [A] iterator dtor
+// CHECK-NEXT: [A] iterator advance: 0 += 2
+// CHECK-NEXT: [A] iterator move assign
+// CHECK-NEXT: [A] iterator deref: 2
+// CHECK-NEXT: [B] iterator advance: 0 += 0
+// CHECK-NEXT: [B] iterator move assign
+// CHECK-NEXT: [B] iterator deref: 0
+// CHECK-NEXT: i=0 v=2 w=0
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [B] iterator advance: 0 += 1
+// CHECK-NEXT: [B] iterator move assign
+// CHECK-NEXT: [B] iterator deref: 1
+// CHECK-NEXT: i=0 v=2 w=1
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [A] iterator dtor
+// CHECK-NEXT: [A] iterator advance: 0 += 2
+// CHECK-NEXT: [A] iterator move assign
+// CHECK-NEXT: [A] iterator deref: 2
+// CHECK-NEXT: [B] iterator advance: 0 += 2
+// CHECK-NEXT: [B] iterator move assign
+// CHECK-NEXT: [B] iterator deref: 2
+// CHECK-NEXT: i=0 v=2 w=2
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [A] iterator dtor
+// CHECK-NEXT: [A] iterator advance: 0 += 0
+// CHECK-NEXT: [A] iterator move assign
+// CHECK-NEXT: [A] iterator deref: 0
+// CHECK-NEXT: [B] iterator advance: 0 += 0
+// CHECK-NEXT: [B] iterator move assign
+// CHECK-NEXT: [B] iterator deref: 0
+// CHECK-NEXT: i=1 v=0 w=0
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [B] iterator advance: 0 += 1
+// CHECK-NEXT: [B] iterator move assign
+// CHECK-NEXT: [B] iterator deref: 1
+// CHECK-NEXT: i=1 v=0 w=1
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [A] iterator dtor
+// CHECK-NEXT: [A] iterator advance: 0 += 1
+// CHECK-NEXT: [A] iterator move assign
+// CHECK-NEXT: [A] iterator deref: 1
+// CHECK-NEXT: [B] iterator advance: 0 += 0
+// CHECK-NEXT: [B] iterator move assign
+// CHECK-NEXT: [B] iterator deref: 0
+// CHECK-NEXT: i=1 v=1 w=0
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [B] iterator advance: 0 += 1
+// CHECK-NEXT: [B] iterator move assign
+// CHECK-NEXT: [B] iterator deref: 1
+// CHECK-NEXT: i=1 v=1 w=1
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [A] iterator dtor
+// CHECK-NEXT: [A] iterator advance: 0 += 0
+// CHECK-NEXT: [A] iterator move assign
+// CHECK-NEXT: [A] iterator deref: 0
+// CHECK-NEXT: [B] iterator advance: 0 += 2
+// CHECK-NEXT: [B] iterator move assign
+// CHECK-NEXT: [B] iterator deref: 2
+// CHECK-NEXT: i=1 v=0 w=2
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [A] iterator dtor
+// CHECK-NEXT: [A] iterator advance: 0 += 1
+// CHECK-NEXT: [A] iterator move assign
+// CHECK-NEXT: [A] iterator deref: 1
+// CHECK-NEXT: [B] iterator advance: 0 += 2
+// CHECK-NEXT: [B] iterator move assign
+// CHECK-NEXT: [B] iterator deref: 2
+// CHECK-NEXT: i=1 v=1 w=2
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [A] iterator dtor
+// CHECK-NEXT: [A] iterator advance: 0 += 2
+// CHECK-NEXT: [A] iterator move assign
+// CHECK-NEXT: [A] iterator deref: 2
+// CHECK-NEXT: [B] iterator advance: 0 += 0
+// CHECK-NEXT: [B] iterator move assign
+// CHECK-NEXT: [B] iterator deref: 0
+// CHECK-NEXT: i=1 v=2 w=0
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [B] iterator advance: 0 += 1
+// CHECK-NEXT: [B] iterator move assign
+// CHECK-NEXT: [B] iterator deref: 1
+// CHECK-NEXT: i=1 v=2 w=1
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [A] iterator dtor
+// CHECK-NEXT: [A] iterator advance: 0 += 2
+// CHECK-NEXT: [A] iterator move assign
+// CHECK-NEXT: [A] iterator deref: 2
+// CHECK-NEXT: [B] iterator advance: 0 += 2
+// CHECK-NEXT: [B] iterator move assign
+// CHECK-NEXT: [B] iterator deref: 2
+// CHECK-NEXT: i=1 v=2 w=2
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [A] iterator dtor
+// CHECK-NEXT: [A] iterator advance: 0 += 0
+// CHECK-NEXT: [A] iterator move assign
+// CHECK-NEXT: [A] iterator deref: 0
+// CHECK-NEXT: [B] iterator advance: 0 += 0
+// CHECK-NEXT: [B] iterator move assign
+// CHECK-NEXT: [B] iterator deref: 0
+// CHECK-NEXT: i=2 v=0 w=0
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [B] iterator advance: 0 += 1
+// CHECK-NEXT: [B] iterator move assign
+// CHECK-NEXT: [B] iterator deref: 1
+// CHECK-NEXT: i=2 v=0 w=1
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [A] iterator dtor
+// CHECK-NEXT: [A] iterator advance: 0 += 1
+// CHECK-NEXT: [A] iterator move assign
+// CHECK-NEXT: [A] iterator deref: 1
+// CHECK-NEXT: [B] iterator advance: 0 += 0
+// CHECK-NEXT: [B] iterator move assign
+// CHECK-NEXT: [B] iterator deref: 0
+// CHECK-NEXT: i=2 v=1 w=0
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [B] iterator advance: 0 += 1
+// CHECK-NEXT: [B] iterator move assign
+// CHECK-NEXT: [B] iterator deref: 1
+// CHECK-NEXT: i=2 v=1 w=1
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [A] iterator dtor
+// CHECK-NEXT: [A] iterator advance: 0 += 0
+// CHECK-NEXT: [A] iterator move assign
+// CHECK-NEXT: [A] iterator deref: 0
+// CHECK-NEXT: [B] iterator advance: 0 += 2
+// CHECK-NEXT: [B] iterator move assign
+// CHECK-NEXT: [B] iterator deref: 2
+// CHECK-NEXT: i=2 v=0 w=2
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [A] iterator dtor
+// CHECK-NEXT: [A] iterator advance: 0 += 1
+// CHECK-NEXT: [A] iterator move assign
+// CHECK-NEXT: [A] iterator deref: 1
+// CHECK-NEXT: [B] iterator advance: 0 += 2
+// CHECK-NEXT: [B] iterator move assign
+// CHECK-NEXT: [B] iterator deref: 2
+// CHECK-NEXT: i=2 v=1 w=2
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [A] iterator dtor
+// CHECK-NEXT: [A] iterator advance: 0 += 2
+// CHECK-NEXT: [A] iterator move assign
+// CHECK-NEXT: [A] iterator deref: 2
+// CHECK-NEXT: [B] iterator advance: 0 += 0
+// CHECK-NEXT: [B] iterator move assign
+// CHECK-NEXT: [B] iterator deref: 0
+// CHECK-NEXT: i=2 v=2 w=0
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [B] iterator advance: 0 += 1
+// CHECK-NEXT: [B] iterator move assign
+// CHECK-NEXT: [B] iterator deref: 1
+// CHECK-NEXT: i=2 v=2 w=1
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [A] iterator dtor
+// CHECK-NEXT: [A] iterator advance: 0 += 2
+// CHECK-NEXT: [A] iterator move assign
+// CHECK-NEXT: [A] iterator deref: 2
+// CHECK-NEXT: [B] iterator advance: 0 += 2
+// CHECK-NEXT: [B] iterator move assign
+// CHECK-NEXT: [B] iterator deref: 2
+// CHECK-NEXT: i=2 v=2 w=2
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [A] iterator dtor
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [B] iterator dtor
+// CHECK-NEXT: [B] dtor
+// CHECK-NEXT: [D] dtor
+// CHECK-NEXT: [A] iterator dtor
+// CHECK-NEXT: [A] iterator dtor
+// CHECK-NEXT: [A] iterator dtor
+// CHECK-NEXT: [A] dtor
+// CHECK-NEXT: [C] dtor
+// CHECK-NEXT: done
diff --git a/openmp/runtime/test/transform/unroll/factor_foreach.cpp b/openmp/runtime/test/transform/unroll/factor_foreach.cpp
new file mode 100644
index 0000000000000..29fef7c187362
--- /dev/null
+++ b/openmp/runtime/test/transform/unroll/factor_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;
+    }
+
+    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 unroll partial(2)
+  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] begin()
+// CHECK-NEXT: [range] end()
+// CHECK-NEXT: [range] iterator 0 != 3
+// CHECK-NEXT: [range] iterator deref: 0
+// CHECK-NEXT: v=0
+// CHECK-NEXT: [range] iterator prefix ++
+// CHECK-NEXT: [range] iterator 1 != 3
+// CHECK-NEXT: [range] iterator deref: 1
+// CHECK-NEXT: v=1
+// CHECK-NEXT: [range] iterator prefix ++
+// CHECK-NEXT: [range] iterator 2 != 3
+// CHECK-NEXT: [range] iterator deref: 2
+// CHECK-NEXT: v=2
+// CHECK-NEXT: [range] iterator prefix ++
+// CHECK-NEXT: [range] iterator 3 != 3
+// 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/unroll/factor_intfor.c b/openmp/runtime/test/transform/unroll/factor_intfor.c
new file mode 100644
index 0000000000000..42ebeb48e41c8
--- /dev/null
+++ b/openmp/runtime/test/transform/unroll/factor_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 unroll partial(2)
+  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=7
+// CHECK-NEXT: i=10
+// CHECK-NEXT: i=13
+// CHECK-NEXT: i=16
+// CHECK-NEXT: done
diff --git a/openmp/runtime/test/transform/unroll/factor_iterfor.cpp b/openmp/runtime/test/transform/unroll/factor_iterfor.cpp
new file mode 100644
index 0000000000000..0298477110b25
--- /dev/null
+++ b/openmp/runtime/test/transform/unroll/factor_iterfor.cpp
@@ -0,0 +1,169 @@
+// 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) {
+      print("iterator copy ctor");
+    }
+
+    Iterator(Iterator &&that) : owner(that.owner), pos(that.pos) {
+      print("iterator move ctor");
+    }
+
+    ~Iterator() { print("iterator dtor"); }
+
+    const Iterator &operator=(const Iterator &that) {
+      print("iterator copy assign");
+      this->owner = that.owner;
+      this->pos = that.pos;
+      return *this;
+    }
+
+    const Iterator &operator=(Iterator &&that) {
+      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++() {
+      print("iterator prefix ++");
+      pos -= 1;
+      return *this;
+    }
+
+    Iterator operator++(int) {
+      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");
+  {
+    Reporter range("range");
+#pragma omp unroll partial(2)
+    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] end()
+// CHECK-NEXT: [range] iterator 0 != 3
+// CHECK-NEXT: [range] iterator deref: 0
+// CHECK-NEXT: v=0
+// CHECK-NEXT: [range] iterator prefix ++
+// CHECK-NEXT: [range] iterator dtor
+// CHECK-NEXT: [range] end()
+// CHECK-NEXT: [range] iterator 1 != 3
+// CHECK-NEXT: [range] iterator deref: 1
+// CHECK-NEXT: v=1
+// CHECK-NEXT: [range] iterator prefix ++
+// CHECK-NEXT: [range] iterator dtor
+// CHECK-NEXT: [range] end()
+// CHECK-NEXT: [range] iterator 2 != 3
+// CHECK-NEXT: [range] iterator deref: 2
+// CHECK-NEXT: v=2
+// CHECK-NEXT: [range] iterator prefix ++
+// CHECK-NEXT: [range] iterator dtor
+// CHECK-NEXT: [range] end()
+// CHECK-NEXT: [range] iterator 3 != 3
+// CHECK-NEXT: [range] iterator dtor
+// CHECK-NEXT: [range] iterator dtor
+// CHECK-NEXT: [range] dtor
+// CHECK-NEXT: done
diff --git a/openmp/runtime/test/transform/unroll/factor_parallel-wsloop-collapse-foreach.cpp b/openmp/runtime/test/transform/unroll/factor_parallel-wsloop-collapse-foreach.cpp
new file mode 100644
index 0000000000000..71567faf79646
--- /dev/null
+++ b/openmp/runtime/test/transform/unroll/factor_parallel-wsloop-collapse-foreach.cpp
@@ -0,0 +1,199 @@
+// 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);
+    }
+
+    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(2) num_threads(1)
+  for (int i = 0; i < 3; ++i)
+#pragma omp unroll partial(2)
+    for (Reporter c{"init-stmt"}; auto &&v : Reporter("range"))
+      printf("i=%d v=%d\n", i, 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 += 0
+// CHECK-NEXT: [range] iterator move assign
+// CHECK-NEXT: [range] iterator deref: 0
+// CHECK-NEXT: i=0 v=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 v=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 v=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 v=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 v=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 v=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 v=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 v=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 v=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/unroll/factor_parallel-wsloop-collapse-intfor.cpp b/openmp/runtime/test/transform/unroll/factor_parallel-wsloop-collapse-intfor.cpp
new file mode 100644
index 0000000000000..0a31f8db07016
--- /dev/null
+++ b/openmp/runtime/test/transform/unroll/factor_parallel-wsloop-collapse-intfor.cpp
@@ -0,0 +1,32 @@
+// 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(2) num_threads(1)
+  for (int i = 0; i < 3; ++i)
+#pragma omp unroll partial(2)
+    for (int j = 0; j < 3; ++j)
+      printf("i=%d j=%d\n", i, j);
+  printf("done\n");
+  return EXIT_SUCCESS;
+}
+
+#endif /* HEADER */
+
+// CHECK:      do
+// CHECK-NEXT: i=0 j=0
+// CHECK-NEXT: i=0 j=1
+// CHECK-NEXT: i=0 j=2
+// CHECK-NEXT: i=1 j=0
+// CHECK-NEXT: i=1 j=1
+// CHECK-NEXT: i=1 j=2
+// CHECK-NEXT: i=2 j=0
+// CHECK-NEXT: i=2 j=1
+// CHECK-NEXT: i=2 j=2
+// CHECK-NEXT: done
diff --git a/openmp/runtime/test/transform/unroll/full_intfor.c b/openmp/runtime/test/transform/unroll/full_intfor.c
new file mode 100644
index 0000000000000..0814511091766
--- /dev/null
+++ b/openmp/runtime/test/transform/unroll/full_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 unroll full
+  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=7
+// CHECK-NEXT: i=10
+// CHECK-NEXT: i=13
+// CHECK-NEXT: i=16
+// CHECK-NEXT: done
diff --git a/openmp/runtime/test/transform/unroll/heuristic_intfor.c b/openmp/runtime/test/transform/unroll/heuristic_intfor.c
new file mode 100644
index 0000000000000..b07bec7d82f0b
--- /dev/null
+++ b/openmp/runtime/test/transform/unroll/heuristic_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 unroll
+  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=7
+// CHECK-NEXT: i=10
+// CHECK-NEXT: i=13
+// CHECK-NEXT: i=16
+// CHECK-NEXT: done
diff --git a/openmp/runtime/test/transform/unroll/partial_intfor.c b/openmp/runtime/test/transform/unroll/partial_intfor.c
new file mode 100644
index 0000000000000..2ede94e70e12d
--- /dev/null
+++ b/openmp/runtime/test/transform/unroll/partial_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 unroll partial
+  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=7
+// CHECK-NEXT: i=10
+// CHECK-NEXT: i=13
+// CHECK-NEXT: i=16
+// CHECK-NEXT: done

>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 02/10] [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 c2bd6a50eed4001c013dc98a21567c092fd2eeb8 Mon Sep 17 00:00:00 2001
From: Michael Kruse <llvm-project at meinersbur.de>
Date: Tue, 21 May 2024 17:08:28 +0200
Subject: [PATCH 03/10] Extract out appendFlattendedStmtList

---
 clang/lib/Sema/SemaOpenMP.cpp | 52 +++++++++++++++++++----------------
 1 file changed, 29 insertions(+), 23 deletions(-)

diff --git a/clang/lib/Sema/SemaOpenMP.cpp b/clang/lib/Sema/SemaOpenMP.cpp
index 663dbb927250e..d6a38cb58450b 100644
--- a/clang/lib/Sema/SemaOpenMP.cpp
+++ b/clang/lib/Sema/SemaOpenMP.cpp
@@ -9815,6 +9815,25 @@ static Stmt *buildPreInits(ASTContext &Context,
   return nullptr;
 }
 
+/// Append the \p Item or the content of a CompoundStmt to the list \p
+/// TargetList.
+///
+/// A CompoundStmt is used as container in case multiple statements need to be
+/// stored in lieu of using an explicit list. Flattening is necessary because
+/// contained DeclStmts need to be visible after the execution of the list. Used
+/// for OpenMP pre-init declarations/statements.
+static void appendFlattendedStmtList(SmallVectorImpl<Stmt *> &TargetList,
+                                     Stmt *Item) {
+  // nullptr represents an empty list.
+  if (!Item)
+    return;
+
+  if (auto *CS = dyn_cast<CompoundStmt>(Item))
+    llvm::append_range(TargetList, CS->body());
+  else
+    TargetList.push_back(Item);
+}
+
 /// Build preinits statement for the given declarations.
 static Stmt *
 buildPreInits(ASTContext &Context,
@@ -9830,19 +9849,13 @@ buildPreInits(ASTContext &Context,
 
 /// Build pre-init statement for the given statements.
 static Stmt *buildPreInits(ASTContext &Context, ArrayRef<Stmt *> PreInits) {
-  if (!PreInits.empty()) {
-    SmallVector<Stmt *> Stmts;
-    for (Stmt *S : PreInits) {
-      // Do not nest CompoundStmts.
-      if (auto *CS = dyn_cast<CompoundStmt>(S)) {
-        llvm::append_range(Stmts, CS->body());
-        continue;
-      }
-      Stmts.push_back(S);
-    }
-    return CompoundStmt::Create(Context, PreInits, FPOptionsOverride(), {}, {});
-  }
-  return nullptr;
+  if (PreInits.empty())
+    return nullptr;
+
+  SmallVector<Stmt *> Stmts;
+  for (Stmt *S : PreInits)
+    appendFlattendedStmtList(Stmts, S);
+  return CompoundStmt::Create(Context, PreInits, FPOptionsOverride(), {}, {});
 }
 
 /// Build postupdate expression for the given list of postupdates expressions.
@@ -9950,7 +9963,7 @@ checkOpenMPLoop(OpenMPDirectiveKind DKind, Expr *CollapseLoopCountExpr,
             else
               Constituents.push_back(DependentPreInits);
             for (Stmt *S : Constituents) {
-              if (DeclStmt *DC = dyn_cast<DeclStmt>(S)) {
+              if (auto *DC = dyn_cast<DeclStmt>(S)) {
                 for (Decl *C : DC->decls()) {
                   auto *D = cast<VarDecl>(C);
                   DeclRefExpr *Ref = buildDeclRefExpr(
@@ -15123,15 +15136,8 @@ bool SemaOpenMP::checkTransformableLoopNest(
           DependentPreInits = Dir->getPreInits();
         else
           llvm_unreachable("Unhandled loop transformation");
-        if (!DependentPreInits)
-          return;
-        // CompoundStmts are used as lists of other statements, add their
-        // contents, not the lists themselves to avoid nesting. This is
-        // necessary because DeclStmts need to be visible after the pre-init.
-        else if (auto *CS = dyn_cast<CompoundStmt>(DependentPreInits))
-          llvm::append_range(OriginalInits.back(), CS->body());
-        else
-          OriginalInits.back().push_back(DependentPreInits);
+
+        appendFlattendedStmtList(OriginalInits.back(), DependentPreInits);
       });
   assert(OriginalInits.back().empty() && "No preinit after innermost loop");
   OriginalInits.pop_back();

>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 04/10] 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));
 }

>From 8eb4b90d0fabed1819d88d5092d5ab2dc5fd3c6f Mon Sep 17 00:00:00 2001
From: Michael Kruse <llvm-project at meinersbur.de>
Date: Wed, 22 May 2024 10:51:42 +0200
Subject: [PATCH 05/10] Address review comments

---
 clang/lib/Sema/SemaOpenMP.cpp | 7 ++-----
 1 file changed, 2 insertions(+), 5 deletions(-)

diff --git a/clang/lib/Sema/SemaOpenMP.cpp b/clang/lib/Sema/SemaOpenMP.cpp
index d6a38cb58450b..bab61e8fd54e8 100644
--- a/clang/lib/Sema/SemaOpenMP.cpp
+++ b/clang/lib/Sema/SemaOpenMP.cpp
@@ -9958,10 +9958,7 @@ checkOpenMPLoop(OpenMPDirectiveKind DKind, Expr *CollapseLoopCountExpr,
             // Search for pre-init declared variables that need to be captured
             // to be referenceable inside the directive.
             SmallVector<Stmt *> Constituents;
-            if (auto *CS = dyn_cast<CompoundStmt>(DependentPreInits))
-              llvm::append_range(Constituents, CS->body());
-            else
-              Constituents.push_back(DependentPreInits);
+            appendFlattendedStmtList(Constituents, DependentPreInits);
             for (Stmt *S : Constituents) {
               if (auto *DC = dyn_cast<DeclStmt>(S)) {
                 for (Decl *C : DC->decls()) {
@@ -15196,7 +15193,7 @@ static void collectLoopStmts(Stmt *AStmt, MutableArrayRef<Stmt *> LoopStmts) {
         LoopStmts[Cnt] = CurStmt;
         return false;
       });
-  assert(llvm::all_of(LoopStmts, [](Stmt *LoopStmt) { return LoopStmt; }) &&
+  assert(!is_contained(LoopStmts, nullptr) &&
          "Expecting a loop statement for each affected loop");
 }
 

>From 6c88e4396f51c609b65a6e682232a28439999218 Mon Sep 17 00:00:00 2001
From: Michael Kruse <llvm-project at meinersbur.de>
Date: Fri, 24 May 2024 23:12:55 +0200
Subject: [PATCH 06/10] Address review comment

---
 clang/include/clang/AST/StmtOpenMP.h      | 3 +--
 clang/lib/AST/StmtOpenMP.cpp              | 7 +++----
 clang/lib/Serialization/ASTReaderStmt.cpp | 5 +++--
 3 files changed, 7 insertions(+), 8 deletions(-)

diff --git a/clang/include/clang/AST/StmtOpenMP.h b/clang/include/clang/AST/StmtOpenMP.h
index fb7f413162fad..e41a9e52b7674 100644
--- a/clang/include/clang/AST/StmtOpenMP.h
+++ b/clang/include/clang/AST/StmtOpenMP.h
@@ -5760,8 +5760,7 @@ class OMPReverseDirective final : public OMPLoopTransformationDirective {
   ///
   /// \param C          Context of the AST.
   /// \param NumClauses Number of clauses to allocate.
-  static OMPReverseDirective *CreateEmpty(const ASTContext &C,
-                                          unsigned NumClauses);
+  static OMPReverseDirective *CreateEmpty(const ASTContext &C);
 
   /// Gets/sets the associated loops after the transformation, i.e. after
   /// de-sugaring.
diff --git a/clang/lib/AST/StmtOpenMP.cpp b/clang/lib/AST/StmtOpenMP.cpp
index 83b8a08e9af73..608ca14e446e5 100644
--- a/clang/lib/AST/StmtOpenMP.cpp
+++ b/clang/lib/AST/StmtOpenMP.cpp
@@ -460,11 +460,10 @@ OMPReverseDirective::Create(const ASTContext &C, SourceLocation StartLoc,
   return Dir;
 }
 
-OMPReverseDirective *OMPReverseDirective::CreateEmpty(const ASTContext &C,
-                                                      unsigned NumClauses) {
+OMPReverseDirective *OMPReverseDirective::CreateEmpty(const ASTContext &C) {
   return createEmptyDirective<OMPReverseDirective>(
-      C, NumClauses, /*HasAssociatedStmt=*/true, TransformedStmtOffset + 1,
-      SourceLocation(), SourceLocation());
+      C, /*NumClauses=*/0, /*HasAssociatedStmt=*/true,
+      TransformedStmtOffset + 1, SourceLocation(), SourceLocation());
 }
 
 OMPForSimdDirective *
diff --git a/clang/lib/Serialization/ASTReaderStmt.cpp b/clang/lib/Serialization/ASTReaderStmt.cpp
index ff72679f993e2..0834f79f940d1 100644
--- a/clang/lib/Serialization/ASTReaderStmt.cpp
+++ b/clang/lib/Serialization/ASTReaderStmt.cpp
@@ -3443,8 +3443,9 @@ Stmt *ASTReader::ReadStmtFromStream(ModuleFile &F) {
     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);
+      assert(Record[ASTStmtReader::NumStmtFields + 1] == 0 &&
+             "Reverse directive has no clauses");
+      S = OMPReverseDirective::CreateEmpty(Context);
       break;
     }
 

>From a05b584d3687d8449a53eb990f835e69ede2db6d Mon Sep 17 00:00:00 2001
From: Michael Kruse <llvm-project at meinersbur.de>
Date: Fri, 21 Jun 2024 12:49:19 +0200
Subject: [PATCH 07/10] Insert reverse in alphabetical order

---
 llvm/include/llvm/Frontend/OpenMP/OMP.td | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)

diff --git a/llvm/include/llvm/Frontend/OpenMP/OMP.td b/llvm/include/llvm/Frontend/OpenMP/OMP.td
index 72e5bd0de998b..51b2cbb0ee37b 100644
--- a/llvm/include/llvm/Frontend/OpenMP/OMP.td
+++ b/llvm/include/llvm/Frontend/OpenMP/OMP.td
@@ -626,9 +626,6 @@ def OMP_DeclareVariant : Directive<"declare variant"> {
   let association = AS_Declaration;
   let category = CA_Declarative;
 }
-def OMP_Reverse : Directive<"reverse"> {
-  let association = AS_Loop;
-}
 def OMP_EndDeclareVariant : Directive<"end declare variant"> {
   let association = AS_Delimited;
   let category = OMP_DeclareVariant.category;
@@ -839,6 +836,10 @@ def OMP_Requires : Directive<"requires"> {
   let association = AS_None;
   let category = CA_Informational;
 }
+def OMP_Reverse : Directive<"reverse"> {
+  let association = AS_Loop;
+  let category = CA_Executable;
+}
 def OMP_Scan : Directive<"scan"> {
   let allowedClauses = [
     VersionedClause<OMPC_Exclusive, 50>,

>From bcd0ae42ea073825949da8c4b988bd36532be5e2 Mon Sep 17 00:00:00 2001
From: Michael Kruse <llvm-project at meinersbur.de>
Date: Wed, 17 Jul 2024 13:03:02 +0200
Subject: [PATCH 08/10] Address @alexey-bataev's review

---
 clang/lib/AST/StmtOpenMP.cpp  |  3 ++-
 clang/lib/Sema/SemaOpenMP.cpp | 41 +++++++++++++++++++----------------
 2 files changed, 24 insertions(+), 20 deletions(-)

diff --git a/clang/lib/AST/StmtOpenMP.cpp b/clang/lib/AST/StmtOpenMP.cpp
index 94f0ee6b0470f..8b19b4209757f 100644
--- a/clang/lib/AST/StmtOpenMP.cpp
+++ b/clang/lib/AST/StmtOpenMP.cpp
@@ -454,7 +454,8 @@ OMPReverseDirective::Create(const ASTContext &C, SourceLocation StartLoc,
                             SourceLocation EndLoc, Stmt *AssociatedStmt,
                             Stmt *TransformedStmt, Stmt *PreInits) {
   OMPReverseDirective *Dir = createDirective<OMPReverseDirective>(
-      C, {}, AssociatedStmt, TransformedStmtOffset + 1, StartLoc, EndLoc);
+      C, std::nullopt, 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 2a508b33bd5e4..76c04424f06de 100644
--- a/clang/lib/Sema/SemaOpenMP.cpp
+++ b/clang/lib/Sema/SemaOpenMP.cpp
@@ -14528,10 +14528,9 @@ StmtResult SemaOpenMP::ActOnOpenMPUnrollDirective(ArrayRef<OMPClause *> Clauses,
 
   // Internal variable names.
   std::string OrigVarName = OrigVar->getNameInfo().getAsString();
-  std::string OuterIVName = (Twine(".unrolled.iv.") + OrigVarName).str();
-  std::string InnerIVName = (Twine(".unroll_inner.iv.") + OrigVarName).str();
-  std::string InnerTripCountName =
-      (Twine(".unroll_inner.tripcount.") + OrigVarName).str();
+  std::string OuterIVName = ".unrolled.iv." + OrigVarName;
+  std::string InnerIVName = ".unroll_inner.iv." + OrigVarName;
+  std::string InnerTripCountName = ".unroll_inner.tripcount." + OrigVarName;
 
   // Create the iteration variable for the unrolled loop.
   VarDecl *OuterIVDecl =
@@ -14731,13 +14730,13 @@ StmtResult SemaOpenMP::ActOnOpenMPReverseDirective(Stmt *AStmt,
   // 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
+  // \code{.c}
   //   for (unsigned i = 0; i < n; ++i)
   //     body(i);
   // \endcode
   //
   // Naive reversal:
-  // \code
+  // \code{.c}
   //   for (unsigned i = n-1; i >= 0; --i)
   //     body(i);
   // \endcode
@@ -14746,7 +14745,7 @@ StmtResult SemaOpenMP::ActOnOpenMPReverseDirective(Stmt *AStmt,
   // 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
+  // \code{.cpp}
   //   for (auto .forward.iv = 0; .forward.iv < n; ++.forward.iv) {
   //     auto .reversed.iv = n - .forward.iv - 1;
   //     i = (.reversed.iv + 0) * 1;                // LoopHelper.Updates
@@ -14778,12 +14777,11 @@ StmtResult SemaOpenMP::ActOnOpenMPReverseDirective(Stmt *AStmt,
       &SemaRef.PP.getIdentifierTable().get(ReversedIVName));
 
   // For init-statement:
-  // \code
-  //   auto .forward.iv = 0
+  // \code{.cpp}
+  //   auto .forward.iv = 0;
   // \endcode
-  IntegerLiteral *Zero =
-      IntegerLiteral::Create(Context, llvm::APInt::getZero(IVWidth),
-                             ForwardIVDecl->getType(), OrigVarLoc);
+  auto *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);
@@ -14791,8 +14789,8 @@ StmtResult SemaOpenMP::ActOnOpenMPReverseDirective(Stmt *AStmt,
     return StmtError();
 
   // Forward iv cond-expression:
-  // \code
-  //   .forward.iv < NumIterations
+  // \code{.cpp}
+  //   .forward.iv < MakeNumIterations()
   // \endcode
   ExprResult Cond =
       SemaRef.BuildBinOp(CurScope, LoopHelper.Cond->getExprLoc(), BO_LT,
@@ -14800,16 +14798,21 @@ StmtResult SemaOpenMP::ActOnOpenMPReverseDirective(Stmt *AStmt,
   if (!Cond.isUsable())
     return StmtError();
 
-  // Forward incr-statement: ++.forward.iv
+  // Forward incr-statement:
+  // \code{.c}
+  //   ++.forward.iv
+  // \endcode
   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);
+  // Reverse the forward-iv:
+  // \code{.cpp}
+  //   auto .reversed.iv = MakeNumIterations() - 1 - .forward.iv
+  // \endcode
+  auto *One = IntegerLiteral::Create(Context, llvm::APInt(IVWidth, 1), IVTy,
+                                     TransformLoc);
   ExprResult Minus = SemaRef.BuildBinOp(CurScope, TransformLoc, BO_Sub,
                                         MakeNumIterations(), One);
   if (!Minus.isUsable())

>From bb801f10d91abb9f9fefd63063eaeff33b519724 Mon Sep 17 00:00:00 2001
From: Michael Kruse <llvm-project at meinersbur.de>
Date: Wed, 17 Jul 2024 13:23:18 +0200
Subject: [PATCH 09/10] Motivated small size

---
 clang/lib/Sema/SemaOpenMP.cpp | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/clang/lib/Sema/SemaOpenMP.cpp b/clang/lib/Sema/SemaOpenMP.cpp
index 76c04424f06de..c8dd73cbce62e 100644
--- a/clang/lib/Sema/SemaOpenMP.cpp
+++ b/clang/lib/Sema/SemaOpenMP.cpp
@@ -14829,7 +14829,7 @@ StmtResult SemaOpenMP::ActOnOpenMPReverseDirective(Stmt *AStmt,
                                /*DirectInit=*/false);
 
   // The new loop body.
-  SmallVector<Stmt *> BodyStmts;
+  SmallVector<Stmt *, 4> BodyStmts;
   BodyStmts.push_back(InitReversed.get());
   llvm::append_range(BodyStmts, LoopHelper.Updates);
   if (auto *CXXRangeFor = dyn_cast<CXXForRangeStmt>(LoopStmt))

>From e6384c72f291f54260874fdbf3350ba63218fc24 Mon Sep 17 00:00:00 2001
From: Michael Kruse <llvm-project at meinersbur.de>
Date: Wed, 17 Jul 2024 13:39:09 +0200
Subject: [PATCH 10/10] Preallocate vector size

---
 clang/lib/Sema/SemaOpenMP.cpp | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/clang/lib/Sema/SemaOpenMP.cpp b/clang/lib/Sema/SemaOpenMP.cpp
index c8dd73cbce62e..b93e6cd649deb 100644
--- a/clang/lib/Sema/SemaOpenMP.cpp
+++ b/clang/lib/Sema/SemaOpenMP.cpp
@@ -14830,6 +14830,8 @@ StmtResult SemaOpenMP::ActOnOpenMPReverseDirective(Stmt *AStmt,
 
   // The new loop body.
   SmallVector<Stmt *, 4> BodyStmts;
+  BodyStmts.reserve(LoopHelper.Updates.size() + 2 +
+                    isa<CXXForRangeStmt>(LoopStmt));
   BodyStmts.push_back(InitReversed.get());
   llvm::append_range(BodyStmts, LoopHelper.Updates);
   if (auto *CXXRangeFor = dyn_cast<CXXForRangeStmt>(LoopStmt))



More information about the Openmp-commits mailing list