[OPENMP] Codegen for 'reduction' clause in 'parallel' directive.

Emit a code for reduction clause. Next code should be emitted for reductions:

static kmp_critical_name lock = { 0 };

void reduce_func(void *lhs[<n>], void *rhs[<n>]) {
  ...
  *(Type<i> *)lhs[i] = RedOp<i>(*(Type<i> *)lhs[i], *(Type<i> *)rhs[i]);
  ...
}

... void *RedList[<n>] = {&<RHSExprs>[0], ..., &<RHSExprs>[<n> - 1]};
switch (__kmpc_reduce{_nowait}(<loc>, <gtid>, <n>, sizeof(RedList), RedList, reduce_func, &<lock>)) {
case 1:
  ...
  <LHSExprs>[i] = RedOp<i>(*<LHSExprs>[i], *<RHSExprs>[i]);
  ...
  __kmpc_end_reduce{_nowait}(<loc>, <gtid>, &<lock>);
  break;
case 2:
  ...
  Atomic(<LHSExprs>[i] = RedOp<i>(*<LHSExprs>[i], *<RHSExprs>[i]));
  ...
  break;
default:
  ;
}
Reduction variables are a kind of a private variables, they have private copies, but initial values are chosen in accordance with the reduction operation.

Differential Revision: http://reviews.llvm.org/D8915

llvm-svn: 234583

clang/include/clang/AST/DataRecursiveASTVisitor.h

@@ -2580,6 +2580,15 @@ RecursiveASTVisitor<Derived>::VisitOMPReductionClause(OMPReductionClause *C) {
   TRY_TO(TraverseNestedNameSpecifierLoc(C->getQualifierLoc()));
   TRY_TO(TraverseDeclarationNameInfo(C->getNameInfo()));
   TRY_TO(VisitOMPClauseList(C));
+  for (auto *E : C->lhs_exprs()) {
+    TRY_TO(TraverseStmt(E));
+  }
+  for (auto *E : C->rhs_exprs()) {
+    TRY_TO(TraverseStmt(E));
+  }
+  for (auto *E : C->reduction_ops()) {
+    TRY_TO(TraverseStmt(E));
+  }
   return true;
 }
 

clang/include/clang/AST/OpenMPClause.h

@@ -1301,6 +1301,48 @@ class OMPReductionClause : public OMPVarListClause<OMPReductionClause> {
   /// \brief Sets the nested name specifier.
   void setQualifierLoc(NestedNameSpecifierLoc NSL) { QualifierLoc = NSL; }
 
+  /// \brief Set list of helper expressions, required for proper codegen of the
+  /// clause. These expressions represent LHS expression in the final
+  /// reduction expression performed by the reduction clause.
+  void setLHSExprs(ArrayRef<Expr *> LHSExprs);
+
+  /// \brief Get the list of helper LHS expressions.
+  MutableArrayRef<Expr *> getLHSExprs() {
+    return MutableArrayRef<Expr *>(varlist_end(), varlist_size());
+  }
+  ArrayRef<const Expr *> getLHSExprs() const {
+    return llvm::makeArrayRef(varlist_end(), varlist_size());
+  }
+
+  /// \brief Set list of helper expressions, required for proper codegen of the
+  /// clause. These expressions represent RHS expression in the final
+  /// reduction expression performed by the reduction clause.
+  /// Also, variables in these expressions are used for proper initialization of
+  /// reduction copies.
+  void setRHSExprs(ArrayRef<Expr *> RHSExprs);
+
+  /// \brief Get the list of helper destination expressions.
+  MutableArrayRef<Expr *> getRHSExprs() {
+    return MutableArrayRef<Expr *>(getLHSExprs().end(), varlist_size());
+  }
+  ArrayRef<const Expr *> getRHSExprs() const {
+    return llvm::makeArrayRef(getLHSExprs().end(), varlist_size());
+  }
+
+  /// \brief Set list of helper reduction expressions, required for proper
+  /// codegen of the clause. These expressions are binary expressions or
+  /// operator/custom reduction call that calculates new value from source
+  /// helper expressions to destination helper expressions.
+  void setReductionOps(ArrayRef<Expr *> ReductionOps);
+
+  /// \brief Get the list of helper reduction expressions.
+  MutableArrayRef<Expr *> getReductionOps() {
+    return MutableArrayRef<Expr *>(getRHSExprs().end(), varlist_size());
+  }
+  ArrayRef<const Expr *> getReductionOps() const {
+    return llvm::makeArrayRef(getRHSExprs().end(), varlist_size());
+  }
+
 public:
   /// \brief Creates clause with a list of variables \a VL.
   ///
@@ -1311,12 +1353,30 @@ public:
   /// \param VL The variables in the clause.
   /// \param QualifierLoc The nested-name qualifier with location information
   /// \param NameInfo The full name info for reduction identifier.
+  /// \param LHSExprs List of helper expressions for proper generation of
+  /// assignment operation required for copyprivate clause. This list represents
+  /// LHSs of the reduction expressions.
+  /// \param RHSExprs List of helper expressions for proper generation of
+  /// assignment operation required for copyprivate clause. This list represents
+  /// RHSs of the reduction expressions.
+  /// Also, variables in these expressions are used for proper initialization of
+  /// reduction copies.
+  /// \param ReductionOps List of helper expressions that represents reduction
+  /// expressions:
+  /// \code
+  /// LHSExprs binop RHSExprs;
+  /// operator binop(LHSExpr, RHSExpr);
+  /// <CutomReduction>(LHSExpr, RHSExpr);
+  /// \endcode
+  /// Required for proper codegen of final reduction operation performed by the
+  /// reduction clause.
   ///
   static OMPReductionClause *
   Create(const ASTContext &C, SourceLocation StartLoc, SourceLocation LParenLoc,
          SourceLocation ColonLoc, SourceLocation EndLoc, ArrayRef<Expr *> VL,
          NestedNameSpecifierLoc QualifierLoc,
-         const DeclarationNameInfo &NameInfo);
+         const DeclarationNameInfo &NameInfo, ArrayRef<Expr *> LHSExprs,
+         ArrayRef<Expr *> RHSExprs, ArrayRef<Expr *> ReductionOps);
   /// \brief Creates an empty clause with the place for \a N variables.
   ///
   /// \param C AST context.
@@ -1331,6 +1391,33 @@ public:
   /// \brief Gets the nested name specifier.
   NestedNameSpecifierLoc getQualifierLoc() const { return QualifierLoc; }
 
+  typedef MutableArrayRef<Expr *>::iterator helper_expr_iterator;
+  typedef ArrayRef<const Expr *>::iterator helper_expr_const_iterator;
+  typedef llvm::iterator_range<helper_expr_iterator> helper_expr_range;
+  typedef llvm::iterator_range<helper_expr_const_iterator>
+      helper_expr_const_range;
+
+  helper_expr_const_range lhs_exprs() const {
+    return helper_expr_const_range(getLHSExprs().begin(), getLHSExprs().end());
+  }
+  helper_expr_range lhs_exprs() {
+    return helper_expr_range(getLHSExprs().begin(), getLHSExprs().end());
+  }
+  helper_expr_const_range rhs_exprs() const {
+    return helper_expr_const_range(getRHSExprs().begin(), getRHSExprs().end());
+  }
+  helper_expr_range rhs_exprs() {
+    return helper_expr_range(getRHSExprs().begin(), getRHSExprs().end());
+  }
+  helper_expr_const_range reduction_ops() const {
+    return helper_expr_const_range(getReductionOps().begin(),
+                                   getReductionOps().end());
+  }
+  helper_expr_range reduction_ops() {
+    return helper_expr_range(getReductionOps().begin(),
+                             getReductionOps().end());
+  }
+
   StmtRange children() {
     return StmtRange(reinterpret_cast<Stmt **>(varlist_begin()),
                      reinterpret_cast<Stmt **>(varlist_end()));

clang/include/clang/AST/RecursiveASTVisitor.h

@@ -2610,6 +2610,15 @@ RecursiveASTVisitor<Derived>::VisitOMPReductionClause(OMPReductionClause *C) {
   TRY_TO(TraverseNestedNameSpecifierLoc(C->getQualifierLoc()));
   TRY_TO(TraverseDeclarationNameInfo(C->getNameInfo()));
   TRY_TO(VisitOMPClauseList(C));
+  for (auto *E : C->lhs_exprs()) {
+    TRY_TO(TraverseStmt(E));
+  }
+  for (auto *E : C->rhs_exprs()) {
+    TRY_TO(TraverseStmt(E));
+  }
+  for (auto *E : C->reduction_ops()) {
+    TRY_TO(TraverseStmt(E));
+  }
   return true;
 }
 

clang/include/clang/Basic/DiagnosticSemaKinds.td

@@ -7451,7 +7451,7 @@ def note_omp_referenced : Note<
 def err_omp_reduction_in_task : Error<
   "reduction variables may not be accessed in an explicit task">;
 def err_omp_reduction_id_not_compatible : Error<
-  "variable of type %0 is not valid for specified reduction operation">;
+  "variable of type %0 is not valid for specified reduction operation: unable to provide default initialization value">;
 def err_omp_prohibited_region : Error<
   "region cannot be%select{| closely}0 nested inside '%1' region"
   "%select{|; perhaps you forget to enclose 'omp %3' directive into a parallel region?|"

clang/lib/AST/Stmt.cpp

@@ -1457,16 +1457,42 @@ void OMPLoopDirective::setFinals(ArrayRef<Expr *> A) {
   std::copy(A.begin(), A.end(), getFinals().begin());
 }
 
+void OMPReductionClause::setLHSExprs(ArrayRef<Expr *> LHSExprs) {
+  assert(
+      LHSExprs.size() == varlist_size() &&
+      "Number of LHS expressions is not the same as the preallocated buffer");
+  std::copy(LHSExprs.begin(), LHSExprs.end(), varlist_end());
+}
+
+void OMPReductionClause::setRHSExprs(ArrayRef<Expr *> RHSExprs) {
+  assert(
+      RHSExprs.size() == varlist_size() &&
+      "Number of RHS expressions is not the same as the preallocated buffer");
+  std::copy(RHSExprs.begin(), RHSExprs.end(), getLHSExprs().end());
+}
+
+void OMPReductionClause::setReductionOps(ArrayRef<Expr *> ReductionOps) {
+  assert(ReductionOps.size() == varlist_size() && "Number of reduction "
+                                                  "expressions is not the same "
+                                                  "as the preallocated buffer");
+  std::copy(ReductionOps.begin(), ReductionOps.end(), getRHSExprs().end());
+}
+
 OMPReductionClause *OMPReductionClause::Create(
     const ASTContext &C, SourceLocation StartLoc, SourceLocation LParenLoc,
     SourceLocation EndLoc, SourceLocation ColonLoc, ArrayRef<Expr *> VL,
-    NestedNameSpecifierLoc QualifierLoc, const DeclarationNameInfo &NameInfo) {
+    NestedNameSpecifierLoc QualifierLoc, const DeclarationNameInfo &NameInfo,
+    ArrayRef<Expr *> LHSExprs, ArrayRef<Expr *> RHSExprs,
+    ArrayRef<Expr *> ReductionOps) {
   void *Mem = C.Allocate(llvm::RoundUpToAlignment(sizeof(OMPReductionClause),
                                                   llvm::alignOf<Expr *>()) +
-                         sizeof(Expr *) * VL.size());
+                         4 * sizeof(Expr *) * VL.size());
   OMPReductionClause *Clause = new (Mem) OMPReductionClause(
       StartLoc, LParenLoc, EndLoc, ColonLoc, VL.size(), QualifierLoc, NameInfo);
   Clause->setVarRefs(VL);
+  Clause->setLHSExprs(LHSExprs);
+  Clause->setRHSExprs(RHSExprs);
+  Clause->setReductionOps(ReductionOps);
   return Clause;
 }
 
@@ -1474,7 +1500,7 @@ OMPReductionClause *OMPReductionClause::CreateEmpty(const ASTContext &C,
                                                     unsigned N) {
   void *Mem = C.Allocate(llvm::RoundUpToAlignment(sizeof(OMPReductionClause),
                                                   llvm::alignOf<Expr *>()) +
-                         sizeof(Expr *) * N);
+                         4 * sizeof(Expr *) * N);
   return new (Mem) OMPReductionClause(N);
 }
 

clang/lib/AST/StmtProfile.cpp

@@ -356,6 +356,15 @@ void OMPClauseProfiler::VisitOMPReductionClause(
       C->getQualifierLoc().getNestedNameSpecifier());
   Profiler->VisitName(C->getNameInfo().getName());
   VisitOMPClauseList(C);
+  for (auto *E : C->lhs_exprs()) {
+    Profiler->VisitStmt(E);
+  }
+  for (auto *E : C->rhs_exprs()) {
+    Profiler->VisitStmt(E);
+  }
+  for (auto *E : C->reduction_ops()) {
+    Profiler->VisitStmt(E);
+  }
 }
 void OMPClauseProfiler::VisitOMPLinearClause(const OMPLinearClause *C) {
   VisitOMPClauseList(C);

clang/lib/CodeGen/CGOpenMPRuntime.cpp

@@ -657,6 +657,62 @@ CGOpenMPRuntime::createRuntimeFunction(OpenMPRTLFunction Function) {
     RTLFn = CGM.CreateRuntimeFunction(FnTy, /*Name=*/"__kmpc_copyprivate");
     break;
   }
+  case OMPRTL__kmpc_reduce: {
+    // Build kmp_int32 __kmpc_reduce(ident_t *loc, kmp_int32 global_tid,
+    // kmp_int32 num_vars, size_t reduce_size, void *reduce_data, void
+    // (*reduce_func)(void *lhs_data, void *rhs_data), kmp_critical_name *lck);
+    llvm::Type *ReduceTypeParams[] = {CGM.VoidPtrTy, CGM.VoidPtrTy};
+    auto *ReduceFnTy = llvm::FunctionType::get(CGM.VoidTy, ReduceTypeParams,
+                                               /*isVarArg=*/false);
+    llvm::Type *TypeParams[] = {
+        getIdentTyPointerTy(), CGM.Int32Ty, CGM.Int32Ty, CGM.SizeTy,
+        CGM.VoidPtrTy, ReduceFnTy->getPointerTo(),
+        llvm::PointerType::getUnqual(KmpCriticalNameTy)};
+    llvm::FunctionType *FnTy =
+        llvm::FunctionType::get(CGM.Int32Ty, TypeParams, /*isVarArg=*/false);
+    RTLFn = CGM.CreateRuntimeFunction(FnTy, /*Name=*/"__kmpc_reduce");
+    break;
+  }
+  case OMPRTL__kmpc_reduce_nowait: {
+    // Build kmp_int32 __kmpc_reduce_nowait(ident_t *loc, kmp_int32
+    // global_tid, kmp_int32 num_vars, size_t reduce_size, void *reduce_data,
+    // void (*reduce_func)(void *lhs_data, void *rhs_data), kmp_critical_name
+    // *lck);
+    llvm::Type *ReduceTypeParams[] = {CGM.VoidPtrTy, CGM.VoidPtrTy};
+    auto *ReduceFnTy = llvm::FunctionType::get(CGM.VoidTy, ReduceTypeParams,
+                                               /*isVarArg=*/false);
+    llvm::Type *TypeParams[] = {
+        getIdentTyPointerTy(), CGM.Int32Ty, CGM.Int32Ty, CGM.SizeTy,
+        CGM.VoidPtrTy, ReduceFnTy->getPointerTo(),
+        llvm::PointerType::getUnqual(KmpCriticalNameTy)};
+    llvm::FunctionType *FnTy =
+        llvm::FunctionType::get(CGM.Int32Ty, TypeParams, /*isVarArg=*/false);
+    RTLFn = CGM.CreateRuntimeFunction(FnTy, /*Name=*/"__kmpc_reduce_nowait");
+    break;
+  }
+  case OMPRTL__kmpc_end_reduce: {
+    // Build void __kmpc_end_reduce(ident_t *loc, kmp_int32 global_tid,
+    // kmp_critical_name *lck);
+    llvm::Type *TypeParams[] = {
+        getIdentTyPointerTy(), CGM.Int32Ty,
+        llvm::PointerType::getUnqual(KmpCriticalNameTy)};
+    llvm::FunctionType *FnTy =
+        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg=*/false);
+    RTLFn = CGM.CreateRuntimeFunction(FnTy, /*Name=*/"__kmpc_end_reduce");
+    break;
+  }
+  case OMPRTL__kmpc_end_reduce_nowait: {
+    // Build __kmpc_end_reduce_nowait(ident_t *loc, kmp_int32 global_tid,
+    // kmp_critical_name *lck);
+    llvm::Type *TypeParams[] = {
+        getIdentTyPointerTy(), CGM.Int32Ty,
+        llvm::PointerType::getUnqual(KmpCriticalNameTy)};
+    llvm::FunctionType *FnTy =
+        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg=*/false);
+    RTLFn =
+        CGM.CreateRuntimeFunction(FnTy, /*Name=*/"__kmpc_end_reduce_nowait");
+    break;
+  }
   }
   return RTLFn;
 }
@@ -998,11 +1054,11 @@ void CGOpenMPRuntime::emitCriticalRegion(CodeGenFunction &CGF,
     llvm::Value *Args[] = {emitUpdateLocation(CGF, Loc), getThreadID(CGF, Loc),
                            getCriticalRegionLock(CriticalName)};
     CGF.EmitRuntimeCall(createRuntimeFunction(OMPRTL__kmpc_critical), Args);
-    emitInlinedDirective(CGF, CriticalOpGen);
     // Build a call to __kmpc_end_critical
     CGF.EHStack.pushCleanup<CallEndCleanup>(
         NormalAndEHCleanup, createRuntimeFunction(OMPRTL__kmpc_end_critical),
         llvm::makeArrayRef(Args));
+    emitInlinedDirective(CGF, CriticalOpGen);
   }
 }
 
@@ -1037,26 +1093,10 @@ void CGOpenMPRuntime::emitMasterRegion(CodeGenFunction &CGF,
       CGF.EmitRuntimeCall(createRuntimeFunction(OMPRTL__kmpc_master), Args);
   emitIfStmt(CGF, IsMaster, [&](CodeGenFunction &CGF) -> void {
     CodeGenFunction::RunCleanupsScope Scope(CGF);
-    MasterOpGen(CGF);
-    // Build a call to __kmpc_end_master.
-    // OpenMP [1.2.2 OpenMP Language Terminology]
-    // For C/C++, an executable statement, possibly compound, with a single
-    // entry at the top and a single exit at the bottom, or an OpenMP
-    // construct.
-    // * Access to the structured block must not be the result of a branch.
-    // * The point of exit cannot be a branch out of the structured block.
-    // * The point of entry must not be a call to setjmp().
-    // * longjmp() and throw() must not violate the entry/exit criteria.
-    // * An expression statement, iteration statement, selection statement, or
-    // try block is considered to be a structured block if the corresponding
-    // compound statement obtained by enclosing it in { and } would be a
-    // structured block.
-    // It is analyzed in Sema, so we can just call __kmpc_end_master() on
-    // fallthrough rather than pushing a normal cleanup for it.
-    // Build a call to __kmpc_end_critical
     CGF.EHStack.pushCleanup<CallEndCleanup>(
         NormalAndEHCleanup, createRuntimeFunction(OMPRTL__kmpc_end_master),
         llvm::makeArrayRef(Args));
+    MasterOpGen(CGF);
   });
 }
 
@@ -1167,29 +1207,15 @@ void CGOpenMPRuntime::emitSingleRegion(CodeGenFunction &CGF,
       CGF.EmitRuntimeCall(createRuntimeFunction(OMPRTL__kmpc_single), Args);
   emitIfStmt(CGF, IsSingle, [&](CodeGenFunction &CGF) -> void {
     CodeGenFunction::RunCleanupsScope Scope(CGF);
+    CGF.EHStack.pushCleanup<CallEndCleanup>(
+        NormalAndEHCleanup, createRuntimeFunction(OMPRTL__kmpc_end_single),
+        llvm::makeArrayRef(Args));
     SingleOpGen(CGF);
     if (DidIt) {
       // did_it = 1;
       CGF.Builder.CreateAlignedStore(CGF.Builder.getInt32(1), DidIt,
                                      DidIt->getAlignment());
     }
-    // Build a call to __kmpc_end_single.
-    // OpenMP [1.2.2 OpenMP Language Terminology]
-    // For C/C++, an executable statement, possibly compound, with a single
-    // entry at the top and a single exit at the bottom, or an OpenMP construct.
-    // * Access to the structured block must not be the result of a branch.
-    // * The point of exit cannot be a branch out of the structured block.
-    // * The point of entry must not be a call to setjmp().
-    // * longjmp() and throw() must not violate the entry/exit criteria.
-    // * An expression statement, iteration statement, selection statement, or
-    // try block is considered to be a structured block if the corresponding
-    // compound statement obtained by enclosing it in { and } would be a
-    // structured block.
-    // It is analyzed in Sema, so we can just call __kmpc_end_single() on
-    // fallthrough rather than pushing a normal cleanup for it.
-    CGF.EHStack.pushCleanup<CallEndCleanup>(
-        NormalAndEHCleanup, createRuntimeFunction(OMPRTL__kmpc_end_single),
-        llvm::makeArrayRef(Args));
   });
   // call __kmpc_copyprivate(ident_t *, gtid, <buf_size>, <copyprivate list>,
   // <copy_func>, did_it);
@@ -1615,6 +1641,265 @@ void CGOpenMPRuntime::emitTaskCall(
   CGF.EmitRuntimeCall(createRuntimeFunction(OMPRTL__kmpc_omp_task), TaskArgs);
 }
 
+static llvm::Value *emitReductionFunction(CodeGenModule &CGM,
+                                          llvm::Type *ArgsType,
+                                          ArrayRef<const Expr *> LHSExprs,
+                                          ArrayRef<const Expr *> RHSExprs,
+                                          ArrayRef<const Expr *> ReductionOps) {
+  auto &C = CGM.getContext();
+
+  // void reduction_func(void *LHSArg, void *RHSArg);
+  FunctionArgList Args;
+  ImplicitParamDecl LHSArg(C, /*DC=*/nullptr, SourceLocation(), /*Id=*/nullptr,
+                           C.VoidPtrTy);
+  ImplicitParamDecl RHSArg(C, /*DC=*/nullptr, SourceLocation(), /*Id=*/nullptr,
+                           C.VoidPtrTy);
+  Args.push_back(&LHSArg);
+  Args.push_back(&RHSArg);
+  FunctionType::ExtInfo EI;
+  auto &CGFI = CGM.getTypes().arrangeFreeFunctionDeclaration(
+      C.VoidTy, Args, EI, /*isVariadic=*/false);
+  auto *Fn = llvm::Function::Create(
+      CGM.getTypes().GetFunctionType(CGFI), llvm::GlobalValue::InternalLinkage,
+      ".omp.reduction.reduction_func", &CGM.getModule());
+  CGM.SetLLVMFunctionAttributes(/*D=*/nullptr, CGFI, Fn);
+  CodeGenFunction CGF(CGM);
+  CGF.StartFunction(GlobalDecl(), C.VoidTy, Fn, CGFI, Args);
+
+  // Dst = (void*[n])(LHSArg);
+  // Src = (void*[n])(RHSArg);
+  auto *LHS = CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
+      CGF.Builder.CreateAlignedLoad(CGF.GetAddrOfLocalVar(&LHSArg),
+                                    CGF.PointerAlignInBytes),
+      ArgsType);
+  auto *RHS = CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
+      CGF.Builder.CreateAlignedLoad(CGF.GetAddrOfLocalVar(&RHSArg),
+                                    CGF.PointerAlignInBytes),
+      ArgsType);
+
+  //  ...
+  //  *(Type<i>*)lhs[i] = RedOp<i>(*(Type<i>*)lhs[i], *(Type<i>*)rhs[i]);
+  //  ...
+  CodeGenFunction::OMPPrivateScope Scope(CGF);
+  for (unsigned I = 0, E = ReductionOps.size(); I < E; ++I) {
+    Scope.addPrivate(
+        cast<VarDecl>(cast<DeclRefExpr>(RHSExprs[I])->getDecl()),
+        [&]() -> llvm::Value *{
+          return CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
+              CGF.Builder.CreateAlignedLoad(
+                  CGF.Builder.CreateStructGEP(/*Ty=*/nullptr, RHS, I),
+                  CGM.PointerAlignInBytes),
+              CGF.ConvertTypeForMem(C.getPointerType(RHSExprs[I]->getType())));
+        });
+    Scope.addPrivate(
+        cast<VarDecl>(cast<DeclRefExpr>(LHSExprs[I])->getDecl()),
+        [&]() -> llvm::Value *{
+          return CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
+              CGF.Builder.CreateAlignedLoad(
+                  CGF.Builder.CreateStructGEP(/*Ty=*/nullptr, LHS, I),
+                  CGM.PointerAlignInBytes),
+              CGF.ConvertTypeForMem(C.getPointerType(LHSExprs[I]->getType())));
+        });
+  }
+  Scope.Privatize();
+  for (auto *E : ReductionOps) {
+    CGF.EmitIgnoredExpr(E);
+  }
+  Scope.ForceCleanup();
+  CGF.FinishFunction();
+  return Fn;
+}
+
+void CGOpenMPRuntime::emitReduction(CodeGenFunction &CGF, SourceLocation Loc,
+                                    ArrayRef<const Expr *> LHSExprs,
+                                    ArrayRef<const Expr *> RHSExprs,
+                                    ArrayRef<const Expr *> ReductionOps,
+                                    bool WithNowait) {
+  // Next code should be emitted for reduction:
+  //
+  // static kmp_critical_name lock = { 0 };
+  //
+  // void reduce_func(void *lhs[<n>], void *rhs[<n>]) {
+  //  *(Type0*)lhs[0] = ReductionOperation0(*(Type0*)lhs[0], *(Type0*)rhs[0]);
+  //  ...
+  //  *(Type<n>-1*)lhs[<n>-1] = ReductionOperation<n>-1(*(Type<n>-1*)lhs[<n>-1],
+  //  *(Type<n>-1*)rhs[<n>-1]);
+  // }
+  //
+  // ...
+  // void *RedList[<n>] = {&<RHSExprs>[0], ..., &<RHSExprs>[<n>-1]};
+  // switch (__kmpc_reduce{_nowait}(<loc>, <gtid>, <n>, sizeof(RedList),
+  // RedList, reduce_func, &<lock>)) {
+  // case 1:
+  //  ...
+  //  <LHSExprs>[i] = RedOp<i>(*<LHSExprs>[i], *<RHSExprs>[i]);
+  //  ...
+  // __kmpc_end_reduce{_nowait}(<loc>, <gtid>, &<lock>);
+  // break;
+  // case 2:
+  //  ...
+  //  Atomic(<LHSExprs>[i] = RedOp<i>(*<LHSExprs>[i], *<RHSExprs>[i]));
+  //  ...
+  // break;
+  // default:;
+  // }
+
+  auto &C = CGM.getContext();
+
+  // 1. Build a list of reduction variables.
+  // void *RedList[<n>] = {<ReductionVars>[0], ..., <ReductionVars>[<n>-1]};
+  llvm::APInt ArraySize(/*unsigned int numBits=*/32, RHSExprs.size());
+  QualType ReductionArrayTy =
+      C.getConstantArrayType(C.VoidPtrTy, ArraySize, ArrayType::Normal,
+                             /*IndexTypeQuals=*/0);
+  auto *ReductionList =
+      CGF.CreateMemTemp(ReductionArrayTy, ".omp.reduction.red_list");
+  for (unsigned I = 0, E = RHSExprs.size(); I < E; ++I) {
+    auto *Elem = CGF.Builder.CreateStructGEP(/*Ty=*/nullptr, ReductionList, I);
+    CGF.Builder.CreateAlignedStore(
+        CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
+            CGF.EmitLValue(RHSExprs[I]).getAddress(), CGF.VoidPtrTy),
+        Elem, CGM.PointerAlignInBytes);
+  }
+
+  // 2. Emit reduce_func().
+  auto *ReductionFn = emitReductionFunction(
+      CGM, CGF.ConvertTypeForMem(ReductionArrayTy)->getPointerTo(), LHSExprs,
+      RHSExprs, ReductionOps);
+
+  // 3. Create static kmp_critical_name lock = { 0 };
+  auto *Lock = getCriticalRegionLock(".reduction");
+
+  // 4. Build res = __kmpc_reduce{_nowait}(<loc>, <gtid>, <n>, sizeof(RedList),
+  // RedList, reduce_func, &<lock>);
+  auto *IdentTLoc = emitUpdateLocation(
+      CGF, Loc,
+      static_cast<OpenMPLocationFlags>(OMP_IDENT_KMPC | OMP_ATOMIC_REDUCE));
+  auto *ThreadId = getThreadID(CGF, Loc);
+  auto *ReductionArrayTySize = llvm::ConstantInt::get(
+      CGM.SizeTy, C.getTypeSizeInChars(ReductionArrayTy).getQuantity());
+  auto *RL = CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(ReductionList,
+                                                             CGF.VoidPtrTy);
+  llvm::Value *Args[] = {
+      IdentTLoc,                             // ident_t *<loc>
+      ThreadId,                              // i32 <gtid>
+      CGF.Builder.getInt32(RHSExprs.size()), // i32 <n>
+      ReductionArrayTySize,                  // size_type sizeof(RedList)
+      RL,                                    // void *RedList
+      ReductionFn, // void (*) (void *, void *) <reduce_func>
+      Lock         // kmp_critical_name *&<lock>
+  };
+  auto Res = CGF.EmitRuntimeCall(
+      createRuntimeFunction(WithNowait ? OMPRTL__kmpc_reduce_nowait
+                                       : OMPRTL__kmpc_reduce),
+      Args);
+
+  // 5. Build switch(res)
+  auto *DefaultBB = CGF.createBasicBlock(".omp.reduction.default");
+  auto *SwInst = CGF.Builder.CreateSwitch(Res, DefaultBB, /*NumCases=*/2);
+
+  // 6. Build case 1:
+  //  ...
+  //  <LHSExprs>[i] = RedOp<i>(*<LHSExprs>[i], *<RHSExprs>[i]);
+  //  ...
+  // __kmpc_end_reduce{_nowait}(<loc>, <gtid>, &<lock>);
+  // break;
+  auto *Case1BB = CGF.createBasicBlock(".omp.reduction.case1");
+  SwInst->addCase(CGF.Builder.getInt32(1), Case1BB);
+  CGF.EmitBlock(Case1BB);
+
+  {
+    CodeGenFunction::RunCleanupsScope Scope(CGF);
+    // Add emission of __kmpc_end_reduce{_nowait}(<loc>, <gtid>, &<lock>);
+    llvm::Value *EndArgs[] = {
+        IdentTLoc, // ident_t *<loc>
+        ThreadId,  // i32 <gtid>
+        Lock       // kmp_critical_name *&<lock>
+    };
+    CGF.EHStack.pushCleanup<CallEndCleanup>(
+        NormalAndEHCleanup,
+        createRuntimeFunction(WithNowait ? OMPRTL__kmpc_end_reduce_nowait
+                                         : OMPRTL__kmpc_end_reduce),
+        llvm::makeArrayRef(EndArgs));
+    for (auto *E : ReductionOps) {
+      CGF.EmitIgnoredExpr(E);
+    }
+  }
+
+  CGF.EmitBranch(DefaultBB);
+
+  // 7. Build case 2:
+  //  ...
+  //  Atomic(<LHSExprs>[i] = RedOp<i>(*<LHSExprs>[i], *<RHSExprs>[i]));
+  //  ...
+  // break;
+  auto *Case2BB = CGF.createBasicBlock(".omp.reduction.case2");
+  SwInst->addCase(CGF.Builder.getInt32(2), Case2BB);
+  CGF.EmitBlock(Case2BB);
+
+  {
+    CodeGenFunction::RunCleanupsScope Scope(CGF);
+    auto I = LHSExprs.begin();
+    for (auto *E : ReductionOps) {
+      const Expr *XExpr = nullptr;
+      const Expr *EExpr = nullptr;
+      const Expr *UpExpr = nullptr;
+      BinaryOperatorKind BO = BO_Comma;
+      // Try to emit update expression as a simple atomic.
+      if (auto *ACO = dyn_cast<AbstractConditionalOperator>(E)) {
+        // If this is a conditional operator, analyze it's condition for
+        // min/max reduction operator.
+        E = ACO->getCond();
+      }
+      if (auto *BO = dyn_cast<BinaryOperator>(E)) {
+        if (BO->getOpcode() == BO_Assign) {
+          XExpr = BO->getLHS();
+          UpExpr = BO->getRHS();
+        }
+      }
+      // Analyze RHS part of the whole expression.
+      if (UpExpr) {
+        if (auto *BORHS =
+                dyn_cast<BinaryOperator>(UpExpr->IgnoreParenImpCasts())) {
+          EExpr = BORHS->getRHS();
+          BO = BORHS->getOpcode();
+        }
+      }
+      if (XExpr) {
+        auto *VD = cast<VarDecl>(cast<DeclRefExpr>(*I)->getDecl());
+        LValue X = CGF.EmitLValue(XExpr);
+        RValue E;
+        if (EExpr)
+          E = CGF.EmitAnyExpr(EExpr);
+        CGF.EmitOMPAtomicSimpleUpdateExpr(
+            X, E, BO, /*IsXLHSInRHSPart=*/true, llvm::Monotonic, Loc,
+            [&CGF, UpExpr, VD](RValue XRValue) {
+              CodeGenFunction::OMPPrivateScope PrivateScope(CGF);
+              PrivateScope.addPrivate(
+                  VD, [&CGF, VD, XRValue]() -> llvm::Value *{
+                    auto *LHSTemp = CGF.CreateMemTemp(VD->getType());
+                    CGF.EmitStoreThroughLValue(
+                        XRValue,
+                        CGF.MakeNaturalAlignAddrLValue(LHSTemp, VD->getType()));
+                    return LHSTemp;
+                  });
+              (void)PrivateScope.Privatize();
+              return CGF.EmitAnyExpr(UpExpr);
+            });
+      } else {
+        // Emit as a critical region.
+        emitCriticalRegion(CGF, ".atomic_reduction", [E](CodeGenFunction &CGF) {
+          CGF.EmitIgnoredExpr(E);
+        }, Loc);
+      }
+      ++I;
+    }
+  }
+
+  CGF.EmitBranch(DefaultBB);
+  CGF.EmitBlock(DefaultBB, /*IsFinished=*/true);
+}
+
 void CGOpenMPRuntime::emitInlinedDirective(CodeGenFunction &CGF,
                                            const RegionCodeGenTy &CodeGen) {
   InlinedOpenMPRegionRAII Region(CGF, CodeGen);

clang/lib/CodeGen/CGOpenMPRuntime.h

@@ -103,6 +103,21 @@ private:
     // kmp_int32 cpy_size, void *cpy_data, void(*cpy_func)(void *, void *),
     // kmp_int32 didit);
     OMPRTL__kmpc_copyprivate,
+    // Call to kmp_int32 __kmpc_reduce(ident_t *loc, kmp_int32 global_tid,
+    // kmp_int32 num_vars, size_t reduce_size, void *reduce_data, void
+    // (*reduce_func)(void *lhs_data, void *rhs_data), kmp_critical_name *lck);
+    OMPRTL__kmpc_reduce,
+    // Call to kmp_int32 __kmpc_reduce_nowait(ident_t *loc, kmp_int32
+    // global_tid, kmp_int32 num_vars, size_t reduce_size, void *reduce_data,
+    // void (*reduce_func)(void *lhs_data, void *rhs_data), kmp_critical_name
+    // *lck);
+    OMPRTL__kmpc_reduce_nowait,
+    // Call to void __kmpc_end_reduce(ident_t *loc, kmp_int32 global_tid,
+    // kmp_critical_name *lck);
+    OMPRTL__kmpc_end_reduce,
+    // Call to void __kmpc_end_reduce_nowait(ident_t *loc, kmp_int32 global_tid,
+    // kmp_critical_name *lck);
+    OMPRTL__kmpc_end_reduce_nowait,
   };
 
   /// \brief Values for bit flags used in the ident_t to describe the fields.
@@ -511,11 +526,54 @@ public:
                             llvm::PointerIntPair<llvm::Value *, 1, bool> Final,
                             llvm::Value *TaskFunction, QualType SharedsTy,
                             llvm::Value *Shareds);
+
   /// \brief Emit code for the directive that does not require outlining.
   ///
   /// \param CodeGen Code generation sequence for the \a D directive.
   virtual void emitInlinedDirective(CodeGenFunction &CGF,
                                     const RegionCodeGenTy &CodeGen);
+  /// \brief Emit a code for reduction clause. Next code should be emitted for
+  /// reduction:
+  /// \code
+  ///
+  /// static kmp_critical_name lock = { 0 };
+  ///
+  /// void reduce_func(void *lhs[<n>], void *rhs[<n>]) {
+  ///  ...
+  ///  *(Type<i>*)lhs[i] = RedOp<i>(*(Type<i>*)lhs[i], *(Type<i>*)rhs[i]);
+  ///  ...
+  /// }
+  ///
+  /// ...
+  /// void *RedList[<n>] = {&<RHSExprs>[0], ..., &<RHSExprs>[<n>-1]};
+  /// switch (__kmpc_reduce{_nowait}(<loc>, <gtid>, <n>, sizeof(RedList),
+  /// RedList, reduce_func, &<lock>)) {
+  /// case 1:
+  ///  ...
+  ///  <LHSExprs>[i] = RedOp<i>(*<LHSExprs>[i], *<RHSExprs>[i]);
+  ///  ...
+  /// __kmpc_end_reduce{_nowait}(<loc>, <gtid>, &<lock>);
+  /// break;
+  /// case 2:
+  ///  ...
+  ///  Atomic(<LHSExprs>[i] = RedOp<i>(*<LHSExprs>[i], *<RHSExprs>[i]));
+  ///  ...
+  /// break;
+  /// default:;
+  /// }
+  /// \endcode
+  ///
+  /// \param LHSExprs List of LHS in \a ReductionOps reduction operations.
+  /// \param RHSExprs List of RHS in \a ReductionOps reduction operations.
+  /// \param ReductionOps List of reduction operations in form 'LHS binop RHS'
+  /// or 'operator binop(LHS, RHS)'.
+  /// \param WithNowait true if parent directive has also nowait clause, false
+  /// otherwise.
+  virtual void emitReduction(CodeGenFunction &CGF, SourceLocation Loc,
+                             ArrayRef<const Expr *> LHSExprs,
+                             ArrayRef<const Expr *> RHSExprs,
+                             ArrayRef<const Expr *> ReductionOps,
+                             bool WithNowait);
 };
 
 } // namespace CodeGen

clang/lib/CodeGen/CGStmtOpenMP.cpp

@@ -209,6 +209,73 @@ void CodeGenFunction::EmitOMPPrivateClause(
   }
 }
 
+void CodeGenFunction::EmitOMPReductionClauseInit(
+    const OMPExecutableDirective &D,
+    CodeGenFunction::OMPPrivateScope &PrivateScope) {
+  auto ReductionFilter = [](const OMPClause *C) -> bool {
+    return C->getClauseKind() == OMPC_reduction;
+  };
+  for (OMPExecutableDirective::filtered_clause_iterator<decltype(
+           ReductionFilter)> I(D.clauses(), ReductionFilter);
+       I; ++I) {
+    auto *C = cast<OMPReductionClause>(*I);
+    auto ILHS = C->lhs_exprs().begin();
+    auto IRHS = C->rhs_exprs().begin();
+    for (auto IRef : C->varlists()) {
+      auto *OrigVD = cast<VarDecl>(cast<DeclRefExpr>(IRef)->getDecl());
+      auto *LHSVD = cast<VarDecl>(cast<DeclRefExpr>(*ILHS)->getDecl());
+      auto *PrivateVD = cast<VarDecl>(cast<DeclRefExpr>(*IRHS)->getDecl());
+      // Store the address of the original variable associated with the LHS
+      // implicit variable.
+      PrivateScope.addPrivate(LHSVD, [this, OrigVD, IRef]() -> llvm::Value *{
+        DeclRefExpr DRE(const_cast<VarDecl *>(OrigVD),
+                        CapturedStmtInfo->lookup(OrigVD) != nullptr,
+                        IRef->getType(), VK_LValue, IRef->getExprLoc());
+        return EmitLValue(&DRE).getAddress();
+      });
+      // Emit reduction copy.
+      bool IsRegistered =
+          PrivateScope.addPrivate(OrigVD, [this, PrivateVD]() -> llvm::Value *{
+            // Emit private VarDecl with reduction init.
+            EmitDecl(*PrivateVD);
+            return GetAddrOfLocalVar(PrivateVD);
+          });
+      assert(IsRegistered && "private var already registered as private");
+      // Silence the warning about unused variable.
+      (void)IsRegistered;
+      ++ILHS, ++IRHS;
+    }
+  }
+}
+
+void CodeGenFunction::EmitOMPReductionClauseFinal(
+    const OMPExecutableDirective &D) {
+  llvm::SmallVector<const Expr *, 8> LHSExprs;
+  llvm::SmallVector<const Expr *, 8> RHSExprs;
+  llvm::SmallVector<const Expr *, 8> ReductionOps;
+  auto ReductionFilter = [](const OMPClause *C) -> bool {
+    return C->getClauseKind() == OMPC_reduction;
+  };
+  bool HasAtLeastOneReduction = false;
+  for (OMPExecutableDirective::filtered_clause_iterator<decltype(
+           ReductionFilter)> I(D.clauses(), ReductionFilter);
+       I; ++I) {
+    HasAtLeastOneReduction = true;
+    auto *C = cast<OMPReductionClause>(*I);
+    LHSExprs.append(C->lhs_exprs().begin(), C->lhs_exprs().end());
+    RHSExprs.append(C->rhs_exprs().begin(), C->rhs_exprs().end());
+    ReductionOps.append(C->reduction_ops().begin(), C->reduction_ops().end());
+  }
+  if (HasAtLeastOneReduction) {
+    // Emit nowait reduction if nowait clause is present or directive is a
+    // parallel directive (it always has implicit barrier).
+    CGM.getOpenMPRuntime().emitReduction(
+        *this, D.getLocEnd(), LHSExprs, RHSExprs, ReductionOps,
+        D.getSingleClause(OMPC_nowait) ||
+            isOpenMPParallelDirective(D.getDirectiveKind()));
+  }
+}
+
 /// \brief Emits code for OpenMP parallel directive in the parallel region.
 static void emitOMPParallelCall(CodeGenFunction &CGF,
                                 const OMPExecutableDirective &S,
@@ -253,11 +320,13 @@ void CodeGenFunction::EmitOMPParallelDirective(const OMPParallelDirective &S) {
     OMPPrivateScope PrivateScope(CGF);
     CGF.EmitOMPPrivateClause(S, PrivateScope);
     CGF.EmitOMPFirstprivateClause(S, PrivateScope);
+    CGF.EmitOMPReductionClauseInit(S, PrivateScope);
     if (PrivateScope.Privatize())
       // Emit implicit barrier to synchronize threads and avoid data races.
       CGF.CGM.getOpenMPRuntime().emitBarrierCall(CGF, S.getLocStart(),
                                                  OMPD_unknown);
     CGF.EmitStmt(cast<CapturedStmt>(S.getAssociatedStmt())->getCapturedStmt());
+    CGF.EmitOMPReductionClauseFinal(S);
     // Emit implicit barrier at the end of the 'parallel' directive.
     CGF.CGM.getOpenMPRuntime().emitBarrierCall(CGF, S.getLocStart(),
                                                OMPD_unknown);
@@ -1155,73 +1224,113 @@ static void EmitOMPAtomicWriteExpr(CodeGenFunction &CGF, bool IsSeqCst,
     CGF.CGM.getOpenMPRuntime().emitFlush(CGF, llvm::None, Loc);
 }
 
-static Optional<llvm::AtomicRMWInst::BinOp>
-getCompatibleAtomicRMWBinOp(ASTContext &Context, BinaryOperatorKind Op,
-                            bool IsXLHSInRHSPart, LValue XLValue,
-                            RValue ExprRValue) {
-  Optional<llvm::AtomicRMWInst::BinOp> RMWOp;
-  // Allow atomicrmw only if 'x' and 'expr' are integer values, lvalue for 'x'
+bool emitOMPAtomicRMW(CodeGenFunction &CGF, LValue X, RValue Update,
+                      BinaryOperatorKind BO, llvm::AtomicOrdering AO,
+                      bool IsXLHSInRHSPart) {
+  auto &Context = CGF.CGM.getContext();
+  // Allow atomicrmw only if 'x' and 'update' are integer values, lvalue for 'x'
   // expression is simple and atomic is allowed for the given type for the
   // target platform.
-  if (ExprRValue.isScalar() &&
-      ExprRValue.getScalarVal()->getType()->isIntegerTy() &&
-      XLValue.isSimple() &&
-      (isa<llvm::ConstantInt>(ExprRValue.getScalarVal()) ||
-       (ExprRValue.getScalarVal()->getType() ==
-        XLValue.getAddress()->getType()->getPointerElementType())) &&
-      Context.getTargetInfo().hasBuiltinAtomic(
-          Context.getTypeSize(XLValue.getType()),
-          Context.toBits(XLValue.getAlignment()))) {
-    switch (Op) {
-    case BO_Add:
-      RMWOp = llvm::AtomicRMWInst::Add;
-      break;
-    case BO_Sub:
-      if (IsXLHSInRHSPart) {
-        RMWOp = llvm::AtomicRMWInst::Sub;
-      }
-      break;
-    case BO_And:
-      RMWOp = llvm::AtomicRMWInst::And;
-      break;
-    case BO_Or:
-      RMWOp = llvm::AtomicRMWInst::Or;
-      break;
-    case BO_Xor:
-      RMWOp = llvm::AtomicRMWInst::Xor;
-      break;
-    case BO_Mul:
-    case BO_Div:
-    case BO_Rem:
-    case BO_Shl:
-    case BO_Shr:
-      break;
-    case BO_PtrMemD:
-    case BO_PtrMemI:
-    case BO_LT:
-    case BO_GT:
-    case BO_LE:
-    case BO_GE:
-    case BO_EQ:
-    case BO_NE:
-    case BO_LAnd:
-    case BO_LOr:
-    case BO_Assign:
-    case BO_MulAssign:
-    case BO_DivAssign:
-    case BO_RemAssign:
-    case BO_AddAssign:
-    case BO_SubAssign:
-    case BO_ShlAssign:
-    case BO_ShrAssign:
-    case BO_AndAssign:
-    case BO_XorAssign:
-    case BO_OrAssign:
-    case BO_Comma:
-      llvm_unreachable("Unexpected binary operation in 'atomic update'.");
+  if (BO == BO_Comma || !Update.isScalar() ||
+      !Update.getScalarVal()->getType()->isIntegerTy() || !X.isSimple() ||
+      (!isa<llvm::ConstantInt>(Update.getScalarVal()) &&
+       (Update.getScalarVal()->getType() !=
+        X.getAddress()->getType()->getPointerElementType())) ||
+      !Context.getTargetInfo().hasBuiltinAtomic(
+          Context.getTypeSize(X.getType()), Context.toBits(X.getAlignment())))
+    return false;
+
+  llvm::AtomicRMWInst::BinOp RMWOp;
+  switch (BO) {
+  case BO_Add:
+    RMWOp = llvm::AtomicRMWInst::Add;
+    break;
+  case BO_Sub:
+    if (!IsXLHSInRHSPart)
+      return false;
+    RMWOp = llvm::AtomicRMWInst::Sub;
+    break;
+  case BO_And:
+    RMWOp = llvm::AtomicRMWInst::And;
+    break;
+  case BO_Or:
+    RMWOp = llvm::AtomicRMWInst::Or;
+    break;
+  case BO_Xor:
+    RMWOp = llvm::AtomicRMWInst::Xor;
+    break;
+  case BO_LT:
+    RMWOp = X.getType()->hasSignedIntegerRepresentation()
+                ? (IsXLHSInRHSPart ? llvm::AtomicRMWInst::Min
+                                   : llvm::AtomicRMWInst::Max)
+                : (IsXLHSInRHSPart ? llvm::AtomicRMWInst::UMin
+                                   : llvm::AtomicRMWInst::UMax);
+    break;
+  case BO_GT:
+    RMWOp = X.getType()->hasSignedIntegerRepresentation()
+                ? (IsXLHSInRHSPart ? llvm::AtomicRMWInst::Max
+                                   : llvm::AtomicRMWInst::Min)
+                : (IsXLHSInRHSPart ? llvm::AtomicRMWInst::UMax
+                                   : llvm::AtomicRMWInst::UMin);
+    break;
+  case BO_Mul:
+  case BO_Div:
+  case BO_Rem:
+  case BO_Shl:
+  case BO_Shr:
+  case BO_LAnd:
+  case BO_LOr:
+    return false;
+  case BO_PtrMemD:
+  case BO_PtrMemI:
+  case BO_LE:
+  case BO_GE:
+  case BO_EQ:
+  case BO_NE:
+  case BO_Assign:
+  case BO_AddAssign:
+  case BO_SubAssign:
+  case BO_AndAssign:
+  case BO_OrAssign:
+  case BO_XorAssign:
+  case BO_MulAssign:
+  case BO_DivAssign:
+  case BO_RemAssign:
+  case BO_ShlAssign:
+  case BO_ShrAssign:
+  case BO_Comma:
+    llvm_unreachable("Unsupported atomic update operation");
+  }
+  auto *UpdateVal = Update.getScalarVal();
+  if (auto *IC = dyn_cast<llvm::ConstantInt>(UpdateVal)) {
+    UpdateVal = CGF.Builder.CreateIntCast(
+        IC, X.getAddress()->getType()->getPointerElementType(),
+        X.getType()->hasSignedIntegerRepresentation());
+  }
+  CGF.Builder.CreateAtomicRMW(RMWOp, X.getAddress(), UpdateVal, AO);
+  return true;
+}
+
+void CodeGenFunction::EmitOMPAtomicSimpleUpdateExpr(
+    LValue X, RValue E, BinaryOperatorKind BO, bool IsXLHSInRHSPart,
+    llvm::AtomicOrdering AO, SourceLocation Loc,
+    const llvm::function_ref<RValue(RValue)> &CommonGen) {
+  // Update expressions are allowed to have the following forms:
+  // x binop= expr; -> xrval + expr;
+  // x++, ++x -> xrval + 1;
+  // x--, --x -> xrval - 1;
+  // x = x binop expr; -> xrval binop expr
+  // x = expr Op x; - > expr binop xrval;
+  if (!emitOMPAtomicRMW(*this, X, E, BO, AO, IsXLHSInRHSPart)) {
+    if (X.isGlobalReg()) {
+      // Emit an update expression: 'xrval' binop 'expr' or 'expr' binop
+      // 'xrval'.
+      EmitStoreThroughLValue(CommonGen(EmitLoadOfLValue(X, Loc)), X);
+    } else {
+      // Perform compare-and-swap procedure.
+      EmitAtomicUpdate(X, AO, CommonGen, X.getType().isVolatileQualified());
     }
   }
-  return std::move(RMWOp);
 }
 
 static void EmitOMPAtomicUpdateExpr(CodeGenFunction &CGF, bool IsSeqCst,
@@ -1237,42 +1346,22 @@ static void EmitOMPAtomicUpdateExpr(CodeGenFunction &CGF, bool IsSeqCst,
   // x--, --x -> xrval - 1;
   // x = x binop expr; -> xrval binop expr
   // x = expr Op x; - > expr binop xrval;
-  assert(X->isLValue() && "X of 'omp atomic write' is not lvalue");
+  assert(X->isLValue() && "X of 'omp atomic update' is not lvalue");
   LValue XLValue = CGF.EmitLValue(X);
   RValue ExprRValue = CGF.EmitAnyExpr(E);
-  const auto &Op =
-      getCompatibleAtomicRMWBinOp(CGF.CGM.getContext(), BOUE->getOpcode(),
-                                  IsXLHSInRHSPart, XLValue, ExprRValue);
   auto AO = IsSeqCst ? llvm::SequentiallyConsistent : llvm::Monotonic;
-  if (Op) {
-    auto *ExprVal = ExprRValue.getScalarVal();
-    if (auto *IC = dyn_cast<llvm::ConstantInt>(ExprVal)) {
-      ExprVal = CGF.Builder.CreateIntCast(
-          IC, XLValue.getAddress()->getType()->getPointerElementType(),
-          XLValue.getType()->hasSignedIntegerRepresentation());
-    }
-    CGF.Builder.CreateAtomicRMW(*Op, XLValue.getAddress(), ExprVal, AO);
-  } else {
-    auto *LHS = cast<OpaqueValueExpr>(BOUE->getLHS()->IgnoreImpCasts());
-    auto *RHS = cast<OpaqueValueExpr>(BOUE->getRHS()->IgnoreImpCasts());
-    CodeGenFunction::OpaqueValueMapping MapExpr(
-        CGF, IsXLHSInRHSPart ? RHS : LHS, ExprRValue);
-    auto *XRValExpr = IsXLHSInRHSPart ? LHS : RHS;
-    if (XLValue.isGlobalReg()) {
-      // Emit an update expression: 'xrval' binop 'expr' or 'expr' binop
-      // 'xrval'.
-      CodeGenFunction::OpaqueValueMapping MapX(
-          CGF, XRValExpr, CGF.EmitLoadOfLValue(XLValue, Loc));
-      CGF.EmitStoreThroughLValue(CGF.EmitAnyExpr(UE), XLValue);
-    } else {
-      // Perform compare-and-swap procedure.
-      CGF.EmitAtomicUpdate(
-          XLValue, AO, [&CGF, &UE, &XRValExpr](RValue XRVal) -> RValue {
-            CodeGenFunction::OpaqueValueMapping MapX(CGF, XRValExpr, XRVal);
-            return CGF.EmitAnyExpr(UE);
-          }, /*IsVolatile=*/false);
-    }
-  }
+  auto *LHS = cast<OpaqueValueExpr>(BOUE->getLHS()->IgnoreImpCasts());
+  auto *RHS = cast<OpaqueValueExpr>(BOUE->getRHS()->IgnoreImpCasts());
+  auto *XRValExpr = IsXLHSInRHSPart ? LHS : RHS;
+  auto *ERValExpr = IsXLHSInRHSPart ? RHS : LHS;
+  auto Gen =
+      [&CGF, UE, ExprRValue, XRValExpr, ERValExpr](RValue XRValue) -> RValue {
+        CodeGenFunction::OpaqueValueMapping MapExpr(CGF, ERValExpr, ExprRValue);
+        CodeGenFunction::OpaqueValueMapping MapX(CGF, XRValExpr, XRValue);
+        return CGF.EmitAnyExpr(UE);
+      };
+  CGF.EmitOMPAtomicSimpleUpdateExpr(XLValue, ExprRValue, BOUE->getOpcode(),
+                                    IsXLHSInRHSPart, AO, Loc, Gen);
   // OpenMP, 2.12.6, atomic Construct
   // Any atomic construct with a seq_cst clause forces the atomically
   // performed operation to include an implicit flush operation without a

clang/lib/CodeGen/CodeGenFunction.h

@@ -2032,10 +2032,39 @@ public:
   void EmitOMPAggregateAssign(LValue OriginalAddr, llvm::Value *PrivateAddr,
                               const Expr *AssignExpr, QualType Type,
                               const VarDecl *VDInit);
+  /// \brief Emit atomic update code for constructs: \a X = \a X \a BO \a E or
+  /// \a X = \a E \a BO \a E.
+  ///
+  /// \param X Value to be updated.
+  /// \param E Update value.
+  /// \param BO Binary operation for update operation.
+  /// \param IsXLHSInRHSPart true if \a X is LHS in RHS part of the update
+  /// expression, false otherwise.
+  /// \param AO Atomic ordering of the generated atomic instructions.
+  /// \param CommonGen Code generator for complex expressions that cannot be
+  /// expressed through atomicrmw instruction.
+  void EmitOMPAtomicSimpleUpdateExpr(
+      LValue X, RValue E, BinaryOperatorKind BO, bool IsXLHSInRHSPart,
+      llvm::AtomicOrdering AO, SourceLocation Loc,
+      const llvm::function_ref<RValue(RValue)> &CommonGen);
   void EmitOMPFirstprivateClause(const OMPExecutableDirective &D,
                                  OMPPrivateScope &PrivateScope);
   void EmitOMPPrivateClause(const OMPExecutableDirective &D,
                             OMPPrivateScope &PrivateScope);
+  /// \brief Emit initial code for reduction variables. Creates reduction copies
+  /// and initializes them with the values according to OpenMP standard.
+  ///
+  /// \param D Directive (possibly) with the 'reduction' clause.
+  /// \param PrivateScope Private scope for capturing reduction variables for
+  /// proper codegen in internal captured statement.
+  ///
+  void EmitOMPReductionClauseInit(const OMPExecutableDirective &D,
+                                  OMPPrivateScope &PrivateScope);
+  /// \brief Emit final update of reduction values to original variables at
+  /// the end of the directive.
+  ///
+  /// \param D Directive that has at least one 'reduction' directives.
+  void EmitOMPReductionClauseFinal(const OMPExecutableDirective &D);
 
   void EmitOMPParallelDirective(const OMPParallelDirective &S);
   void EmitOMPSimdDirective(const OMPSimdDirective &S);

clang/lib/Sema/SemaOpenMP.cpp

@@ -5234,19 +5234,19 @@ OMPClause *Sema::ActOnOpenMPReductionClause(
   switch (OOK) {
   case OO_Plus:
   case OO_Minus:
-    BOK = BO_AddAssign;
+    BOK = BO_Add;
     break;
   case OO_Star:
-    BOK = BO_MulAssign;
+    BOK = BO_Mul;
     break;
   case OO_Amp:
-    BOK = BO_AndAssign;
+    BOK = BO_And;
     break;
   case OO_Pipe:
-    BOK = BO_OrAssign;
+    BOK = BO_Or;
     break;
   case OO_Caret:
-    BOK = BO_XorAssign;
+    BOK = BO_Xor;
     break;
   case OO_AmpAmp:
     BOK = BO_LAnd;
@@ -5254,7 +5254,44 @@ OMPClause *Sema::ActOnOpenMPReductionClause(
   case OO_PipePipe:
     BOK = BO_LOr;
     break;
-  default:
+  case OO_New:
+  case OO_Delete:
+  case OO_Array_New:
+  case OO_Array_Delete:
+  case OO_Slash:
+  case OO_Percent:
+  case OO_Tilde:
+  case OO_Exclaim:
+  case OO_Equal:
+  case OO_Less:
+  case OO_Greater:
+  case OO_LessEqual:
+  case OO_GreaterEqual:
+  case OO_PlusEqual:
+  case OO_MinusEqual:
+  case OO_StarEqual:
+  case OO_SlashEqual:
+  case OO_PercentEqual:
+  case OO_CaretEqual:
+  case OO_AmpEqual:
+  case OO_PipeEqual:
+  case OO_LessLess:
+  case OO_GreaterGreater:
+  case OO_LessLessEqual:
+  case OO_GreaterGreaterEqual:
+  case OO_EqualEqual:
+  case OO_ExclaimEqual:
+  case OO_PlusPlus:
+  case OO_MinusMinus:
+  case OO_Comma:
+  case OO_ArrowStar:
+  case OO_Arrow:
+  case OO_Call:
+  case OO_Subscript:
+  case OO_Conditional:
+  case NUM_OVERLOADED_OPERATORS:
+    llvm_unreachable("Unexpected reduction identifier");
+  case OO_None:
     if (auto II = DN.getAsIdentifierInfo()) {
       if (II->isStr("max"))
         BOK = BO_GT;
@@ -5276,11 +5313,17 @@ OMPClause *Sema::ActOnOpenMPReductionClause(
   }
 
   SmallVector<Expr *, 8> Vars;
+  SmallVector<Expr *, 8> LHSs;
+  SmallVector<Expr *, 8> RHSs;
+  SmallVector<Expr *, 8> ReductionOps;
   for (auto RefExpr : VarList) {
     assert(RefExpr && "nullptr expr in OpenMP reduction clause.");
     if (isa<DependentScopeDeclRefExpr>(RefExpr)) {
       // It will be analyzed later.
       Vars.push_back(RefExpr);
+      LHSs.push_back(nullptr);
+      RHSs.push_back(nullptr);
+      ReductionOps.push_back(nullptr);
       continue;
     }
 
@@ -5289,6 +5332,9 @@ OMPClause *Sema::ActOnOpenMPReductionClause(
         RefExpr->containsUnexpandedParameterPack()) {
       // It will be analyzed later.
       Vars.push_back(RefExpr);
+      LHSs.push_back(nullptr);
+      RHSs.push_back(nullptr);
+      ReductionOps.push_back(nullptr);
       continue;
     }
 
@@ -5381,23 +5427,6 @@ OMPClause *Sema::ActOnOpenMPReductionClause(
           << VD;
       continue;
     }
-    bool Suppress = getDiagnostics().getSuppressAllDiagnostics();
-    getDiagnostics().setSuppressAllDiagnostics(true);
-    ExprResult ReductionOp =
-        BuildBinOp(DSAStack->getCurScope(), ReductionId.getLocStart(), BOK,
-                   RefExpr, RefExpr);
-    getDiagnostics().setSuppressAllDiagnostics(Suppress);
-    if (ReductionOp.isInvalid()) {
-      Diag(ELoc, diag::err_omp_reduction_id_not_compatible) << Type
-                                                            << ReductionIdRange;
-      bool IsDecl =
-          VD->isThisDeclarationADefinition(Context) == VarDecl::DeclarationOnly;
-      Diag(VD->getLocation(),
-           IsDecl ? diag::note_previous_decl : diag::note_defined_here)
-          << VD;
-      continue;
-    }
-
     // OpenMP [2.14.1.1, Data-sharing Attribute Rules for Variables Referenced
     // in a Construct]
     //  Variables with the predetermined data-sharing attributes may not be
@@ -5440,55 +5469,160 @@ OMPClause *Sema::ActOnOpenMPReductionClause(
         continue;
       }
     }
-
-    CXXRecordDecl *RD = getLangOpts().CPlusPlus
-                            ? Type.getNonReferenceType()->getAsCXXRecordDecl()
-                            : nullptr;
-    // FIXME This code must be replaced by actual constructing/destructing of
-    // the reduction variable.
-    if (RD) {
-      CXXConstructorDecl *CD = LookupDefaultConstructor(RD);
-      PartialDiagnostic PD =
-          PartialDiagnostic(PartialDiagnostic::NullDiagnostic());
-      if (!CD ||
-          CheckConstructorAccess(ELoc, CD,
-                                 InitializedEntity::InitializeTemporary(Type),
-                                 CD->getAccess(), PD) == AR_inaccessible ||
-          CD->isDeleted()) {
-        Diag(ELoc, diag::err_omp_required_method)
-            << getOpenMPClauseName(OMPC_reduction) << 0;
-        bool IsDecl = VD->isThisDeclarationADefinition(Context) ==
-                      VarDecl::DeclarationOnly;
-        Diag(VD->getLocation(),
-             IsDecl ? diag::note_previous_decl : diag::note_defined_here)
-            << VD;
-        Diag(RD->getLocation(), diag::note_previous_decl) << RD;
-        continue;
+    Type = Type.getNonLValueExprType(Context).getUnqualifiedType();
+    auto *LHSVD = BuildVarDecl(*this, ELoc, Type, ".reduction.lhs");
+    auto *RHSVD = BuildVarDecl(*this, ELoc, Type, VD->getName());
+    // Add initializer for private variable.
+    Expr *Init = nullptr;
+    switch (BOK) {
+    case BO_Add:
+    case BO_Xor:
+    case BO_Or:
+    case BO_LOr:
+      // '+', '-', '^', '|', '||' reduction ops - initializer is '0'.
+      if (Type->isScalarType() || Type->isAnyComplexType()) {
+        Init = ActOnIntegerConstant(ELoc, /*Val=*/0).get();
       }
-      MarkFunctionReferenced(ELoc, CD);
-      DiagnoseUseOfDecl(CD, ELoc);
-
-      CXXDestructorDecl *DD = RD->getDestructor();
-      if (DD) {
-        if (CheckDestructorAccess(ELoc, DD, PD) == AR_inaccessible ||
-            DD->isDeleted()) {
-          Diag(ELoc, diag::err_omp_required_method)
-              << getOpenMPClauseName(OMPC_reduction) << 4;
-          bool IsDecl = VD->isThisDeclarationADefinition(Context) ==
-                        VarDecl::DeclarationOnly;
-          Diag(VD->getLocation(),
-               IsDecl ? diag::note_previous_decl : diag::note_defined_here)
-              << VD;
-          Diag(RD->getLocation(), diag::note_previous_decl) << RD;
-          continue;
+      break;
+    case BO_Mul:
+    case BO_LAnd:
+      if (Type->isScalarType() || Type->isAnyComplexType()) {
+        // '*' and '&&' reduction ops - initializer is '1'.
+        Init = ActOnIntegerConstant(ELoc, /*Val=*/1).get();
+      }
+      break;
+    case BO_And: {
+      // '&' reduction op - initializer is '~0'.
+      QualType OrigType = Type;
+      if (auto *ComplexTy = OrigType->getAs<ComplexType>()) {
+        Type = ComplexTy->getElementType();
+      }
+      if (Type->isRealFloatingType()) {
+        llvm::APFloat InitValue =
+            llvm::APFloat::getAllOnesValue(Context.getTypeSize(Type),
+                                           /*isIEEE=*/true);
+        Init = FloatingLiteral::Create(Context, InitValue, /*isexact=*/true,
+                                       Type, ELoc);
+      } else if (Type->isScalarType()) {
+        auto Size = Context.getTypeSize(Type);
+        QualType IntTy = Context.getIntTypeForBitwidth(Size, /*Signed=*/0);
+        llvm::APInt InitValue = llvm::APInt::getAllOnesValue(Size);
+        Init = IntegerLiteral::Create(Context, InitValue, IntTy, ELoc);
+      }
+      if (Init && OrigType->isAnyComplexType()) {
+        // Init = 0xFFFF + 0xFFFFi;
+        auto *Im = new (Context) ImaginaryLiteral(Init, OrigType);
+        Init = CreateBuiltinBinOp(ELoc, BO_Add, Init, Im).get();
+      }
+      Type = OrigType;
+      break;
+    }
+    case BO_LT:
+    case BO_GT: {
+      // 'min' reduction op - initializer is 'Largest representable number in
+      // the reduction list item type'.
+      // 'max' reduction op - initializer is 'Least representable number in
+      // the reduction list item type'.
+      if (Type->isIntegerType() || Type->isPointerType()) {
+        bool IsSigned = Type->hasSignedIntegerRepresentation();
+        auto Size = Context.getTypeSize(Type);
+        QualType IntTy =
+            Context.getIntTypeForBitwidth(Size, /*Signed=*/IsSigned);
+        llvm::APInt InitValue =
+            (BOK != BO_LT)
+                ? IsSigned ? llvm::APInt::getSignedMinValue(Size)
+                           : llvm::APInt::getMinValue(Size)
+                : IsSigned ? llvm::APInt::getSignedMaxValue(Size)
+                           : llvm::APInt::getMaxValue(Size);
+        Init = IntegerLiteral::Create(Context, InitValue, IntTy, ELoc);
+        if (Type->isPointerType()) {
+          // Cast to pointer type.
+          auto CastExpr = BuildCStyleCastExpr(
+              SourceLocation(), Context.getTrivialTypeSourceInfo(Type, ELoc),
+              SourceLocation(), Init);
+          if (CastExpr.isInvalid())
+            continue;
+          Init = CastExpr.get();
         }
-        MarkFunctionReferenced(ELoc, DD);
-        DiagnoseUseOfDecl(DD, ELoc);
+      } else if (Type->isRealFloatingType()) {
+        llvm::APFloat InitValue = llvm::APFloat::getLargest(
+            Context.getFloatTypeSemantics(Type), BOK != BO_LT);
+        Init = FloatingLiteral::Create(Context, InitValue, /*isexact=*/true,
+                                       Type, ELoc);
+      }
+      break;
+    }
+    case BO_PtrMemD:
+    case BO_PtrMemI:
+    case BO_MulAssign:
+    case BO_Div:
+    case BO_Rem:
+    case BO_Sub:
+    case BO_Shl:
+    case BO_Shr:
+    case BO_LE:
+    case BO_GE:
+    case BO_EQ:
+    case BO_NE:
+    case BO_AndAssign:
+    case BO_XorAssign:
+    case BO_OrAssign:
+    case BO_Assign:
+    case BO_AddAssign:
+    case BO_SubAssign:
+    case BO_DivAssign:
+    case BO_RemAssign:
+    case BO_ShlAssign:
+    case BO_ShrAssign:
+    case BO_Comma:
+      llvm_unreachable("Unexpected reduction operation");
+    }
+    if (Init) {
+      AddInitializerToDecl(RHSVD, Init, /*DirectInit=*/false,
+                           /*TypeMayContainAuto=*/false);
+    } else {
+      ActOnUninitializedDecl(RHSVD, /*TypeMayContainAuto=*/false);
+    }
+    if (!RHSVD->hasInit()) {
+      Diag(ELoc, diag::err_omp_reduction_id_not_compatible) << Type
+                                                            << ReductionIdRange;
+      bool IsDecl =
+          VD->isThisDeclarationADefinition(Context) == VarDecl::DeclarationOnly;
+      Diag(VD->getLocation(),
+           IsDecl ? diag::note_previous_decl : diag::note_defined_here)
+          << VD;
+      continue;
+    }
+    auto *LHSDRE = BuildDeclRefExpr(LHSVD, Type, VK_LValue, ELoc).get();
+    auto *RHSDRE = BuildDeclRefExpr(RHSVD, Type, VK_LValue, ELoc).get();
+    ExprResult ReductionOp =
+        BuildBinOp(DSAStack->getCurScope(), ReductionId.getLocStart(), BOK,
+                   LHSDRE, RHSDRE);
+    if (ReductionOp.isUsable()) {
+      if (BOK != BO_LOr && BOK != BO_LAnd) {
+        ReductionOp =
+            BuildBinOp(DSAStack->getCurScope(), ReductionId.getLocStart(),
+                       BO_Assign, LHSDRE, ReductionOp.get());
+      } else {
+        auto *ConditionalOp = new (Context) ConditionalOperator(
+            ReductionOp.get(), SourceLocation(), LHSDRE, SourceLocation(),
+            RHSDRE, Type, VK_LValue, OK_Ordinary);
+        ReductionOp =
+            BuildBinOp(DSAStack->getCurScope(), ReductionId.getLocStart(),
+                       BO_Assign, LHSDRE, ConditionalOp);
+      }
+      if (ReductionOp.isUsable()) {
+        ReductionOp = ActOnFinishFullExpr(ReductionOp.get());
       }
     }
+    if (ReductionOp.isInvalid())
+      continue;
 
     DSAStack->addDSA(VD, DE, OMPC_reduction);
     Vars.push_back(DE);
+    LHSs.push_back(LHSDRE);
+    RHSs.push_back(RHSDRE);
+    ReductionOps.push_back(ReductionOp.get());
   }
 
   if (Vars.empty())
@@ -5496,7 +5630,8 @@ OMPClause *Sema::ActOnOpenMPReductionClause(
 
   return OMPReductionClause::Create(
       Context, StartLoc, LParenLoc, ColonLoc, EndLoc, Vars,
-      ReductionIdScopeSpec.getWithLocInContext(Context), ReductionId);
+      ReductionIdScopeSpec.getWithLocInContext(Context), ReductionId, LHSs,
+      RHSs, ReductionOps);
 }
 
 OMPClause *Sema::ActOnOpenMPLinearClause(ArrayRef<Expr *> VarList, Expr *Step,

clang/lib/Serialization/ASTReaderStmt.cpp

@@ -1917,6 +1917,18 @@ void OMPClauseReader::VisitOMPReductionClause(OMPReductionClause *C) {
   for (unsigned i = 0; i != NumVars; ++i)
     Vars.push_back(Reader->Reader.ReadSubExpr());
   C->setVarRefs(Vars);
+  Vars.clear();
+  for (unsigned i = 0; i != NumVars; ++i)
+    Vars.push_back(Reader->Reader.ReadSubExpr());
+  C->setLHSExprs(Vars);
+  Vars.clear();
+  for (unsigned i = 0; i != NumVars; ++i)
+    Vars.push_back(Reader->Reader.ReadSubExpr());
+  C->setRHSExprs(Vars);
+  Vars.clear();
+  for (unsigned i = 0; i != NumVars; ++i)
+    Vars.push_back(Reader->Reader.ReadSubExpr());
+  C->setReductionOps(Vars);
 }
 
 void OMPClauseReader::VisitOMPLinearClause(OMPLinearClause *C) {

clang/lib/Serialization/ASTWriterStmt.cpp

@@ -1815,6 +1815,12 @@ void OMPClauseWriter::VisitOMPReductionClause(OMPReductionClause *C) {
   Writer->Writer.AddDeclarationNameInfo(C->getNameInfo(), Record);
   for (auto *VE : C->varlists())
     Writer->Writer.AddStmt(VE);
+  for (auto *E : C->lhs_exprs())
+    Writer->Writer.AddStmt(E);
+  for (auto *E : C->rhs_exprs())
+    Writer->Writer.AddStmt(E);
+  for (auto *E : C->reduction_ops())
+    Writer->Writer.AddStmt(E);
 }
 
 void OMPClauseWriter::VisitOMPLinearClause(OMPLinearClause *C) {

clang/test/OpenMP/for_reduction_messages.cpp

@@ -1,4 +1,4 @@
-// RUN: %clang_cc1 -verify -fopenmp=libiomp5 -ferror-limit 100 -o - %s
+// RUN: %clang_cc1 -verify -fopenmp=libiomp5 -ferror-limit 150 -o - %s
 
 void foo() {
 }
@@ -11,7 +11,7 @@ struct S1; // expected-note {{declared here}} expected-note 4 {{forward declarat
 extern S1 a;
 class S2 {
   mutable int a;
-  S2 &operator+=(const S2 &arg) { return (*this); }
+  S2 &operator+(const S2 &arg) { return (*this); } // expected-note 4 {{implicitly declared private here}}
 
 public:
   S2() : a(0) {}
@@ -28,17 +28,17 @@ class S3 {
 public:
   S3() : a(0) {}
   S3(const S3 &s3) : a(s3.a) {}
-  S3 operator+=(const S3 &arg1) { return arg1; }
+  S3 operator+(const S3 &arg1) { return arg1; }
 };
-int operator+=(const S3 &arg1, const S3 &arg2) { return 5; }
+int operator+(const S3 &arg1, const S3 &arg2) { return 5; }
 S3 c;               // expected-note 2 {{'c' defined here}}
 const S3 ca[5];     // expected-note 2 {{'ca' defined here}}
 extern const int f; // expected-note 4 {{'f' declared here}}
-class S4 {          // expected-note {{'S4' declared here}}
+class S4 {
   int a;
-  S4();
+  S4(); // expected-note {{implicitly declared private here}}
   S4(const S4 &s4);
-  S4 &operator+=(const S4 &arg) { return (*this); }
+  S4 &operator+(const S4 &arg) { return (*this); }
 
 public:
   S4(int v) : a(v) {}
@@ -46,26 +46,26 @@ public:
 S4 &operator&=(S4 &arg1, S4 &arg2) { return arg1; }
 class S5 {
   int a;
-  S5() : a(0) {}
+  S5() : a(0) {} // expected-note {{implicitly declared private here}}
   S5(const S5 &s5) : a(s5.a) {}
-  S5 &operator+=(const S5 &arg);
+  S5 &operator+(const S5 &arg);
 
 public:
   S5(int v) : a(v) {}
 };
-class S6 {
+class S6 { // expected-note 2 {{candidate function (the implicit copy assignment operator) not viable: no known conversion from 'int' to 'const S6' for 1st argument}}
   int a;
 
 public:
   S6() : a(6) {}
   operator int() { return 6; }
-} o; // expected-note 2 {{'o' defined here}}
+} o;
 
 S3 h, k;
 #pragma omp threadprivate(h) // expected-note 2 {{defined as threadprivate or thread local}}
 
 template <class T>       // expected-note {{declared here}}
-T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
+T tmain(T argc) {
   const T d = T();       // expected-note 4 {{'d' defined here}}
   const T da[5] = {T()}; // expected-note 2 {{'da' defined here}}
   T qa[5] = {T()};
@@ -74,7 +74,7 @@ T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
   S3 &p = k;               // expected-note 2 {{'p' defined here}}
   const T &r = da[(int)i]; // expected-note 2 {{'r' defined here}}
   T &q = qa[(int)i];       // expected-note 2 {{'q' defined here}}
-  T fl;                    // expected-note {{'fl' defined here}}
+  T fl;
 #pragma omp parallel
 #pragma omp for reduction // expected-error {{expected '(' after 'reduction'}}
   for (int i = 0; i < 10; ++i)
@@ -104,11 +104,11 @@ T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp parallel
-#pragma omp for reduction(& : argc // expected-error {{expected ')'}} expected-note {{to match this '('}} expected-error {{variable of type 'float' is not valid for specified reduction operation}}
+#pragma omp for reduction(& : argc // expected-error {{expected ')'}} expected-note {{to match this '('}} expected-error {{invalid operands to binary expression ('float' and 'float')}}
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp parallel
-#pragma omp for reduction(| : argc, // expected-error {{expected expression}} expected-error {{expected ')'}} expected-note {{to match this '('}} expected-error {{variable of type 'float' is not valid for specified reduction operation}}
+#pragma omp for reduction(| : argc, // expected-error {{expected expression}} expected-error {{expected ')'}} expected-note {{to match this '('}} expected-error {{invalid operands to binary expression ('float' and 'float')}}
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp parallel
@@ -128,7 +128,7 @@ T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp parallel
-#pragma omp for reduction(+ : a, b, c, d, f) // expected-error {{reduction variable with incomplete type 'S1'}} expected-error 3 {{const-qualified variable cannot be reduction}}
+#pragma omp for reduction(+ : a, b, c, d, f) // expected-error {{reduction variable with incomplete type 'S1'}} expected-error 3 {{const-qualified variable cannot be reduction}} expected-error 3 {{'operator+' is a private member of 'S2'}}
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp parallel
@@ -152,7 +152,7 @@ T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp parallel
-#pragma omp for reduction(^ : fl) // expected-error {{variable of type 'float' is not valid for specified reduction operation}}
+#pragma omp for reduction(^ : fl) // expected-error {{invalid operands to binary expression ('float' and 'float')}}
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp parallel
@@ -168,7 +168,7 @@ T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp parallel
-#pragma omp for reduction(+ : o) // expected-error {{variable of type 'class S6' is not valid for specified reduction operation}}
+#pragma omp for reduction(+ : o) // expected-error {{no viable overloaded '='}}
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp parallel
@@ -208,14 +208,14 @@ int main(int argc, char **argv) {
   const int d = 5;       // expected-note 2 {{'d' defined here}}
   const int da[5] = {0}; // expected-note {{'da' defined here}}
   int qa[5] = {0};
-  S4 e(4); // expected-note {{'e' defined here}}
-  S5 g(5); // expected-note {{'g' defined here}}
+  S4 e(4);
+  S5 g(5);
   int i;
   int &j = i;           // expected-note 2 {{'j' defined here}}
   S3 &p = k;            // expected-note 2 {{'p' defined here}}
   const int &r = da[i]; // expected-note {{'r' defined here}}
   int &q = qa[i];       // expected-note {{'q' defined here}}
-  float fl;             // expected-note {{'fl' defined here}}
+  float fl;
 #pragma omp parallel
 #pragma omp for reduction // expected-error {{expected '(' after 'reduction'}}
   for (int i = 0; i < 10; ++i)
@@ -269,7 +269,7 @@ int main(int argc, char **argv) {
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp parallel
-#pragma omp for reduction(+ : a, b, c, d, f) // expected-error {{reduction variable with incomplete type 'S1'}} expected-error 2 {{const-qualified variable cannot be reduction}}
+#pragma omp for reduction(+ : a, b, c, d, f) // expected-error {{reduction variable with incomplete type 'S1'}} expected-error 2 {{const-qualified variable cannot be reduction}} expected-error {{'operator+' is a private member of 'S2'}}
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp parallel
@@ -293,7 +293,7 @@ int main(int argc, char **argv) {
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp parallel
-#pragma omp for reduction(^ : fl) // expected-error {{variable of type 'float' is not valid for specified reduction operation}}
+#pragma omp for reduction(^ : fl) // expected-error {{invalid operands to binary expression ('float' and 'float')}}
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp parallel
@@ -305,7 +305,7 @@ int main(int argc, char **argv) {
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp parallel
-#pragma omp for reduction(& : e, g) // expected-error {{reduction variable must have an accessible, unambiguous default constructor}} expected-error {{variable of type 'S5' is not valid for specified reduction operation}}
+#pragma omp for reduction(& : e, g) // expected-error {{calling a private constructor of class 'S4'}} expected-error {{calling a private constructor of class 'S5'}} expected-error {{invalid operands to binary expression ('S4' and 'S4')}} expected-error {{invalid operands to binary expression ('S5' and 'S5')}}
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp parallel
@@ -313,7 +313,7 @@ int main(int argc, char **argv) {
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp parallel
-#pragma omp for reduction(+ : o) // expected-error {{variable of type 'class S6' is not valid for specified reduction operation}}
+#pragma omp for reduction(+ : o) // expected-error {{no viable overloaded '='}}
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp parallel

clang/test/OpenMP/for_simd_reduction_messages.cpp

@@ -11,7 +11,7 @@ struct S1; // expected-note {{declared here}} expected-note 4 {{forward declarat
 extern S1 a;
 class S2 {
   mutable int a;
-  S2 &operator+=(const S2 &arg) { return (*this); }
+  S2 &operator+(const S2 &arg) { return (*this); } // expected-note 4 {{implicitly declared private here}}
 
 public:
   S2() : a(0) {}
@@ -28,17 +28,17 @@ class S3 {
 public:
   S3() : a(0) {}
   S3(const S3 &s3) : a(s3.a) {}
-  S3 operator+=(const S3 &arg1) { return arg1; }
+  S3 operator+(const S3 &arg1) { return arg1; }
 };
-int operator+=(const S3 &arg1, const S3 &arg2) { return 5; }
+int operator+(const S3 &arg1, const S3 &arg2) { return 5; }
 S3 c;               // expected-note 2 {{'c' defined here}}
 const S3 ca[5];     // expected-note 2 {{'ca' defined here}}
 extern const int f; // expected-note 4 {{'f' declared here}}
-class S4 {          // expected-note {{'S4' declared here}}
+class S4 {
   int a;
-  S4();
+  S4(); // expected-note {{implicitly declared private here}}
   S4(const S4 &s4);
-  S4 &operator+=(const S4 &arg) { return (*this); }
+  S4 &operator+(const S4 &arg) { return (*this); }
 
 public:
   S4(int v) : a(v) {}
@@ -46,26 +46,26 @@ public:
 S4 &operator&=(S4 &arg1, S4 &arg2) { return arg1; }
 class S5 {
   int a;
-  S5() : a(0) {}
+  S5() : a(0) {} // expected-note {{implicitly declared private here}}
   S5(const S5 &s5) : a(s5.a) {}
-  S5 &operator+=(const S5 &arg);
+  S5 &operator+(const S5 &arg);
 
 public:
   S5(int v) : a(v) {}
 };
-class S6 {
+class S6 { // expected-note 2 {{candidate function (the implicit copy assignment operator) not viable: no known conversion from 'int' to 'const S6' for 1st argument}}
   int a;
 
 public:
   S6() : a(6) {}
   operator int() { return 6; }
-} o; // expected-note 2 {{'o' defined here}}
+} o;
 
 S3 h, k;
 #pragma omp threadprivate(h) // expected-note 2 {{defined as threadprivate or thread local}}
 
 template <class T>       // expected-note {{declared here}}
-T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
+T tmain(T argc) {
   const T d = T();       // expected-note 4 {{'d' defined here}}
   const T da[5] = {T()}; // expected-note 2 {{'da' defined here}}
   T qa[5] = {T()};
@@ -74,7 +74,7 @@ T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
   S3 &p = k;               // expected-note 2 {{'p' defined here}}
   const T &r = da[(int)i]; // expected-note 2 {{'r' defined here}}
   T &q = qa[(int)i];       // expected-note 2 {{'q' defined here}}
-  T fl;                    // expected-note {{'fl' defined here}}
+  T fl;
 #pragma omp parallel
 #pragma omp for simd reduction // expected-error {{expected '(' after 'reduction'}}
   for (int i = 0; i < 10; ++i)
@@ -104,11 +104,11 @@ T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp parallel
-#pragma omp for simd reduction(& : argc // expected-error {{expected ')'}} expected-note {{to match this '('}} expected-error {{variable of type 'float' is not valid for specified reduction operation}}
+#pragma omp for simd reduction(& : argc // expected-error {{expected ')'}} expected-note {{to match this '('}} expected-error {{invalid operands to binary expression ('float' and 'float')}}
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp parallel
-#pragma omp for simd reduction(| : argc, // expected-error {{expected expression}} expected-error {{expected ')'}} expected-note {{to match this '('}} expected-error {{variable of type 'float' is not valid for specified reduction operation}}
+#pragma omp for simd reduction(| : argc, // expected-error {{expected expression}} expected-error {{expected ')'}} expected-note {{to match this '('}} expected-error {{invalid operands to binary expression ('float' and 'float')}}
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp parallel
@@ -128,7 +128,7 @@ T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp parallel
-#pragma omp for simd reduction(+ : a, b, c, d, f) // expected-error {{reduction variable with incomplete type 'S1'}} expected-error 3 {{const-qualified variable cannot be reduction}}
+#pragma omp for simd reduction(+ : a, b, c, d, f) // expected-error {{reduction variable with incomplete type 'S1'}} expected-error 3 {{const-qualified variable cannot be reduction}} expected-error 3 {{'operator+' is a private member of 'S2'}}
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp parallel
@@ -152,7 +152,7 @@ T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp parallel
-#pragma omp for simd reduction(^ : fl) // expected-error {{variable of type 'float' is not valid for specified reduction operation}}
+#pragma omp for simd reduction(^ : fl) // expected-error {{invalid operands to binary expression ('float' and 'float')}}
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp parallel
@@ -168,7 +168,7 @@ T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp parallel
-#pragma omp for simd reduction(+ : o) // expected-error {{variable of type 'class S6' is not valid for specified reduction operation}}
+#pragma omp for simd reduction(+ : o) // expected-error {{no viable overloaded '='}}
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp parallel
@@ -208,14 +208,14 @@ int main(int argc, char **argv) {
   const int d = 5;       // expected-note 2 {{'d' defined here}}
   const int da[5] = {0}; // expected-note {{'da' defined here}}
   int qa[5] = {0};
-  S4 e(4); // expected-note {{'e' defined here}}
-  S5 g(5); // expected-note {{'g' defined here}}
+  S4 e(4);
+  S5 g(5);
   int i;
   int &j = i;           // expected-note 2 {{'j' defined here}}
   S3 &p = k;            // expected-note 2 {{'p' defined here}}
   const int &r = da[i]; // expected-note {{'r' defined here}}
   int &q = qa[i];       // expected-note {{'q' defined here}}
-  float fl;             // expected-note {{'fl' defined here}}
+  float fl;
 #pragma omp parallel
 #pragma omp for simd reduction // expected-error {{expected '(' after 'reduction'}}
   for (int i = 0; i < 10; ++i)
@@ -269,7 +269,7 @@ int main(int argc, char **argv) {
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp parallel
-#pragma omp for simd reduction(+ : a, b, c, d, f) // expected-error {{reduction variable with incomplete type 'S1'}} expected-error 2 {{const-qualified variable cannot be reduction}}
+#pragma omp for simd reduction(+ : a, b, c, d, f) // expected-error {{reduction variable with incomplete type 'S1'}} expected-error 2 {{const-qualified variable cannot be reduction}} expected-error {{'operator+' is a private member of 'S2'}}
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp parallel
@@ -293,7 +293,7 @@ int main(int argc, char **argv) {
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp parallel
-#pragma omp for simd reduction(^ : fl) // expected-error {{variable of type 'float' is not valid for specified reduction operation}}
+#pragma omp for simd reduction(^ : fl) // expected-error {{invalid operands to binary expression ('float' and 'float')}}
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp parallel
@@ -305,7 +305,7 @@ int main(int argc, char **argv) {
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp parallel
-#pragma omp for simd reduction(& : e, g) // expected-error {{reduction variable must have an accessible, unambiguous default constructor}} expected-error {{variable of type 'S5' is not valid for specified reduction operation}}
+#pragma omp for simd reduction(& : e, g) // expected-error {{calling a private constructor of class 'S4'}} expected-error {{invalid operands to binary expression ('S4' and 'S4')}} expected-error {{calling a private constructor of class 'S5'}} expected-error {{invalid operands to binary expression ('S5' and 'S5')}}
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp parallel
@@ -313,7 +313,7 @@ int main(int argc, char **argv) {
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp parallel
-#pragma omp for simd reduction(+ : o) // expected-error {{variable of type 'class S6' is not valid for specified reduction operation}}
+#pragma omp for simd reduction(+ : o) // expected-error {{no viable overloaded '='}}
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp parallel

clang/test/OpenMP/parallel_for_reduction_messages.cpp

@@ -11,7 +11,7 @@ struct S1; // expected-note {{declared here}} expected-note 4 {{forward declarat
 extern S1 a;
 class S2 {
   mutable int a;
-  S2 &operator+=(const S2 &arg) { return (*this); }
+  S2 &operator+(const S2 &arg) { return (*this); } // expected-note 4 {{implicitly declared private here}}
 
 public:
   S2() : a(0) {}
@@ -28,17 +28,17 @@ class S3 {
 public:
   S3() : a(0) {}
   S3(const S3 &s3) : a(s3.a) {}
-  S3 operator+=(const S3 &arg1) { return arg1; }
+  S3 operator+(const S3 &arg1) { return arg1; }
 };
-int operator+=(const S3 &arg1, const S3 &arg2) { return 5; }
+int operator+(const S3 &arg1, const S3 &arg2) { return 5; }
 S3 c;               // expected-note 2 {{'c' defined here}}
 const S3 ca[5];     // expected-note 2 {{'ca' defined here}}
 extern const int f; // expected-note 4 {{'f' declared here}}
-class S4 {          // expected-note {{'S4' declared here}}
+class S4 {
   int a;
-  S4();
+  S4(); // expected-note {{implicitly declared private here}}
   S4(const S4 &s4);
-  S4 &operator+=(const S4 &arg) { return (*this); }
+  S4 &operator+(const S4 &arg) { return (*this); }
 
 public:
   S4(int v) : a(v) {}
@@ -46,26 +46,26 @@ public:
 S4 &operator&=(S4 &arg1, S4 &arg2) { return arg1; }
 class S5 {
   int a;
-  S5() : a(0) {}
+  S5() : a(0) {} // expected-note {{implicitly declared private here}}
   S5(const S5 &s5) : a(s5.a) {}
-  S5 &operator+=(const S5 &arg);
+  S5 &operator+(const S5 &arg);
 
 public:
   S5(int v) : a(v) {}
 };
-class S6 {
+class S6 { // expected-note 2 {{candidate function (the implicit copy assignment operator) not viable: no known conversion from 'int' to 'const S6' for 1st argument}}
   int a;
 
 public:
   S6() : a(6) {}
   operator int() { return 6; }
-} o; // expected-note 2 {{'o' defined here}}
+} o;
 
 S3 h, k;
 #pragma omp threadprivate(h) // expected-note 2 {{defined as threadprivate or thread local}}
 
 template <class T>       // expected-note {{declared here}}
-T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
+T tmain(T argc) {
   const T d = T();       // expected-note 4 {{'d' defined here}}
   const T da[5] = {T()}; // expected-note 2 {{'da' defined here}}
   T qa[5] = {T()};
@@ -74,7 +74,7 @@ T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
   S3 &p = k;                       // expected-note 2 {{'p' defined here}}
   const T &r = da[(int)i];         // expected-note 2 {{'r' defined here}}
   T &q = qa[(int)i];               // expected-note 2 {{'q' defined here}}
-  T fl;                            // expected-note {{'fl' defined here}}
+  T fl;
 #pragma omp parallel for reduction // expected-error {{expected '(' after 'reduction'}}
   for (int i = 0; i < 10; ++i)
     foo();
@@ -96,10 +96,10 @@ T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
 #pragma omp parallel for reduction(\) // expected-error {{expected unqualified-id}} expected-warning {{missing ':' after reduction identifier - ignoring}}
   for (int i = 0; i < 10; ++i)
     foo();
-#pragma omp parallel for reduction(& : argc // expected-error {{expected ')'}} expected-note {{to match this '('}} expected-error {{variable of type 'float' is not valid for specified reduction operation}}
+#pragma omp parallel for reduction(& : argc // expected-error {{expected ')'}} expected-note {{to match this '('}} expected-error {{invalid operands to binary expression ('float' and 'float')}}
   for (int i = 0; i < 10; ++i)
     foo();
-#pragma omp parallel for reduction(| : argc, // expected-error {{expected expression}} expected-error {{expected ')'}} expected-note {{to match this '('}} expected-error {{variable of type 'float' is not valid for specified reduction operation}}
+#pragma omp parallel for reduction(| : argc, // expected-error {{expected expression}} expected-error {{expected ')'}} expected-note {{to match this '('}} expected-error {{invalid operands to binary expression ('float' and 'float')}}
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp parallel for reduction(|| : argc ? i : argc) // expected-error 2 {{expected variable name}}
@@ -114,7 +114,7 @@ T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
 #pragma omp parallel for reduction(^ : T) // expected-error {{'T' does not refer to a value}}
   for (int i = 0; i < 10; ++i)
     foo();
-#pragma omp parallel for reduction(+ : a, b, c, d, f) // expected-error {{reduction variable with incomplete type 'S1'}} expected-error 3 {{const-qualified variable cannot be reduction}}
+#pragma omp parallel for reduction(+ : a, b, c, d, f) // expected-error {{reduction variable with incomplete type 'S1'}} expected-error 3 {{const-qualified variable cannot be reduction}} expected-error 3 {{'operator+' is a private member of 'S2'}}
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp parallel for reduction(min : a, b, c, d, f) // expected-error {{reduction variable with incomplete type 'S1'}} expected-error 2 {{arguments of OpenMP clause 'reduction' for 'min' or 'max' must be of arithmetic type}} expected-error 3 {{const-qualified variable cannot be reduction}}
@@ -132,7 +132,7 @@ T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
 #pragma omp parallel for reduction(- : da) // expected-error {{a reduction variable with array type 'const int [5]'}} expected-error {{a reduction variable with array type 'const float [5]'}}
   for (int i = 0; i < 10; ++i)
     foo();
-#pragma omp parallel for reduction(^ : fl) // expected-error {{variable of type 'float' is not valid for specified reduction operation}}
+#pragma omp parallel for reduction(^ : fl) // expected-error {{invalid operands to binary expression ('float' and 'float')}}
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp parallel for reduction(&& : S2::S2s)
@@ -144,7 +144,7 @@ T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
 #pragma omp parallel for reduction(+ : h, k) // expected-error {{threadprivate or thread local variable cannot be reduction}}
   for (int i = 0; i < 10; ++i)
     foo();
-#pragma omp parallel for reduction(+ : o) // expected-error {{variable of type 'class S6' is not valid for specified reduction operation}}
+#pragma omp parallel for reduction(+ : o) // expected-error {{no viable overloaded '='}}
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp parallel for private(i), reduction(+ : j), reduction(+ : q) // expected-error 4 {{argument of OpenMP clause 'reduction' must reference the same object in all threads}}
@@ -181,14 +181,14 @@ int main(int argc, char **argv) {
   const int d = 5;       // expected-note 2 {{'d' defined here}}
   const int da[5] = {0}; // expected-note {{'da' defined here}}
   int qa[5] = {0};
-  S4 e(4); // expected-note {{'e' defined here}}
-  S5 g(5); // expected-note {{'g' defined here}}
+  S4 e(4);
+  S5 g(5);
   int i;
   int &j = i;                      // expected-note 2 {{'j' defined here}}
   S3 &p = k;                       // expected-note 2 {{'p' defined here}}
   const int &r = da[i];            // expected-note {{'r' defined here}}
   int &q = qa[i];                  // expected-note {{'q' defined here}}
-  float fl;                        // expected-note {{'fl' defined here}}
+  float fl;
 #pragma omp parallel for reduction // expected-error {{expected '(' after 'reduction'}}
   for (int i = 0; i < 10; ++i)
     foo();
@@ -228,7 +228,7 @@ int main(int argc, char **argv) {
 #pragma omp parallel for reduction(^ : S1) // expected-error {{'S1' does not refer to a value}}
   for (int i = 0; i < 10; ++i)
     foo();
-#pragma omp parallel for reduction(+ : a, b, c, d, f) // expected-error {{reduction variable with incomplete type 'S1'}} expected-error 2 {{const-qualified variable cannot be reduction}}
+#pragma omp parallel for reduction(+ : a, b, c, d, f) // expected-error {{reduction variable with incomplete type 'S1'}} expected-error 2 {{const-qualified variable cannot be reduction}} expected-error {{'operator+' is a private member of 'S2'}}
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp parallel for reduction(min : a, b, c, d, f) // expected-error {{reduction variable with incomplete type 'S1'}} expected-error 2 {{arguments of OpenMP clause 'reduction' for 'min' or 'max' must be of arithmetic type}} expected-error 2 {{const-qualified variable cannot be reduction}}
@@ -246,7 +246,7 @@ int main(int argc, char **argv) {
 #pragma omp parallel for reduction(- : da) // expected-error {{a reduction variable with array type 'const int [5]'}}
   for (int i = 0; i < 10; ++i)
     foo();
-#pragma omp parallel for reduction(^ : fl) // expected-error {{variable of type 'float' is not valid for specified reduction operation}}
+#pragma omp parallel for reduction(^ : fl) // expected-error {{invalid operands to binary expression ('float' and 'float')}}
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp parallel for reduction(&& : S2::S2s)
@@ -255,13 +255,13 @@ int main(int argc, char **argv) {
 #pragma omp parallel for reduction(&& : S2::S2sc) // expected-error {{const-qualified variable cannot be reduction}}
   for (int i = 0; i < 10; ++i)
     foo();
-#pragma omp parallel for reduction(& : e, g) // expected-error {{reduction variable must have an accessible, unambiguous default constructor}} expected-error {{variable of type 'S5' is not valid for specified reduction operation}}
+#pragma omp parallel for reduction(& : e, g) // expected-error {{calling a private constructor of class 'S4'}} expected-error {{invalid operands to binary expression ('S4' and 'S4')}} expected-error {{calling a private constructor of class 'S5'}} expected-error {{invalid operands to binary expression ('S5' and 'S5')}}
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp parallel for reduction(+ : h, k) // expected-error {{threadprivate or thread local variable cannot be reduction}}
   for (int i = 0; i < 10; ++i)
     foo();
-#pragma omp parallel for reduction(+ : o) // expected-error {{variable of type 'class S6' is not valid for specified reduction operation}}
+#pragma omp parallel for reduction(+ : o) // expected-error {{no viable overloaded '='}}
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp parallel for private(i), reduction(+ : j), reduction(+ : q) // expected-error 2 {{argument of OpenMP clause 'reduction' must reference the same object in all threads}}

clang/test/OpenMP/parallel_for_simd_reduction_messages.cpp

@@ -11,7 +11,7 @@ struct S1; // expected-note {{declared here}} expected-note 4 {{forward declarat
 extern S1 a;
 class S2 {
   mutable int a;
-  S2 &operator+=(const S2 &arg) { return (*this); }
+  S2 &operator+(const S2 &arg) { return (*this); } // expected-note 4 {{implicitly declared private here}}
 
 public:
   S2() : a(0) {}
@@ -28,17 +28,17 @@ class S3 {
 public:
   S3() : a(0) {}
   S3(const S3 &s3) : a(s3.a) {}
-  S3 operator+=(const S3 &arg1) { return arg1; }
+  S3 operator+(const S3 &arg1) { return arg1; }
 };
-int operator+=(const S3 &arg1, const S3 &arg2) { return 5; }
+int operator+(const S3 &arg1, const S3 &arg2) { return 5; }
 S3 c;               // expected-note 2 {{'c' defined here}}
 const S3 ca[5];     // expected-note 2 {{'ca' defined here}}
 extern const int f; // expected-note 4 {{'f' declared here}}
-class S4 {          // expected-note {{'S4' declared here}}
+class S4 {
   int a;
-  S4();
+  S4(); // expected-note {{implicitly declared private here}}
   S4(const S4 &s4);
-  S4 &operator+=(const S4 &arg) { return (*this); }
+  S4 &operator+(const S4 &arg) { return (*this); }
 
 public:
   S4(int v) : a(v) {}
@@ -46,26 +46,26 @@ public:
 S4 &operator&=(S4 &arg1, S4 &arg2) { return arg1; }
 class S5 {
   int a;
-  S5() : a(0) {}
+  S5() : a(0) {} // expected-note {{implicitly declared private here}}
   S5(const S5 &s5) : a(s5.a) {}
-  S5 &operator+=(const S5 &arg);
+  S5 &operator+(const S5 &arg);
 
 public:
   S5(int v) : a(v) {}
 };
-class S6 {
+class S6 { // expected-note 2 {{candidate function (the implicit copy assignment operator) not viable: no known conversion from 'int' to 'const S6' for 1st argument}}
   int a;
 
 public:
   S6() : a(6) {}
   operator int() { return 6; }
-} o; // expected-note 2 {{'o' defined here}}
+} o;
 
 S3 h, k;
 #pragma omp threadprivate(h) // expected-note 2 {{defined as threadprivate or thread local}}
 
 template <class T>       // expected-note {{declared here}}
-T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
+T tmain(T argc) {
   const T d = T();       // expected-note 4 {{'d' defined here}}
   const T da[5] = {T()}; // expected-note 2 {{'da' defined here}}
   T qa[5] = {T()};
@@ -74,7 +74,7 @@ T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
   S3 &p = k;                       // expected-note 2 {{'p' defined here}}
   const T &r = da[(int)i];         // expected-note 2 {{'r' defined here}}
   T &q = qa[(int)i];               // expected-note 2 {{'q' defined here}}
-  T fl;                            // expected-note {{'fl' defined here}}
+  T fl;
 #pragma omp parallel for simd reduction // expected-error {{expected '(' after 'reduction'}}
   for (int i = 0; i < 10; ++i)
     foo();
@@ -96,10 +96,10 @@ T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
 #pragma omp parallel for simd reduction(\) // expected-error {{expected unqualified-id}} expected-warning {{missing ':' after reduction identifier - ignoring}}
   for (int i = 0; i < 10; ++i)
     foo();
-#pragma omp parallel for simd reduction(& : argc // expected-error {{expected ')'}} expected-note {{to match this '('}} expected-error {{variable of type 'float' is not valid for specified reduction operation}}
+#pragma omp parallel for simd reduction(& : argc // expected-error {{expected ')'}} expected-note {{to match this '('}} expected-error {{invalid operands to binary expression ('float' and 'float')}}
   for (int i = 0; i < 10; ++i)
     foo();
-#pragma omp parallel for simd reduction(| : argc, // expected-error {{expected expression}} expected-error {{expected ')'}} expected-note {{to match this '('}} expected-error {{variable of type 'float' is not valid for specified reduction operation}}
+#pragma omp parallel for simd reduction(| : argc, // expected-error {{expected expression}} expected-error {{expected ')'}} expected-note {{to match this '('}} expected-error {{invalid operands to binary expression ('float' and 'float')}}
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp parallel for simd reduction(|| : argc ? i : argc) // expected-error 2 {{expected variable name}}
@@ -114,7 +114,7 @@ T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
 #pragma omp parallel for simd reduction(^ : T) // expected-error {{'T' does not refer to a value}}
   for (int i = 0; i < 10; ++i)
     foo();
-#pragma omp parallel for simd reduction(+ : a, b, c, d, f) // expected-error {{reduction variable with incomplete type 'S1'}} expected-error 3 {{const-qualified variable cannot be reduction}}
+#pragma omp parallel for simd reduction(+ : a, b, c, d, f) // expected-error {{reduction variable with incomplete type 'S1'}} expected-error 3 {{const-qualified variable cannot be reduction}} expected-error 3 {{'operator+' is a private member of 'S2'}}
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp parallel for simd reduction(min : a, b, c, d, f) // expected-error {{reduction variable with incomplete type 'S1'}} expected-error 2 {{arguments of OpenMP clause 'reduction' for 'min' or 'max' must be of arithmetic type}} expected-error 3 {{const-qualified variable cannot be reduction}}
@@ -132,7 +132,7 @@ T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
 #pragma omp parallel for simd reduction(- : da) // expected-error {{a reduction variable with array type 'const int [5]'}} expected-error {{a reduction variable with array type 'const float [5]'}}
   for (int i = 0; i < 10; ++i)
     foo();
-#pragma omp parallel for simd reduction(^ : fl) // expected-error {{variable of type 'float' is not valid for specified reduction operation}}
+#pragma omp parallel for simd reduction(^ : fl) // expected-error {{invalid operands to binary expression ('float' and 'float')}}
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp parallel for simd reduction(&& : S2::S2s)
@@ -144,7 +144,7 @@ T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
 #pragma omp parallel for simd reduction(+ : h, k) // expected-error {{threadprivate or thread local variable cannot be reduction}}
   for (int i = 0; i < 10; ++i)
     foo();
-#pragma omp parallel for simd reduction(+ : o) // expected-error {{variable of type 'class S6' is not valid for specified reduction operation}}
+#pragma omp parallel for simd reduction(+ : o) // expected-error {{no viable overloaded '='}}
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp parallel for simd private(i), reduction(+ : j), reduction(+ : q) // expected-error 4 {{argument of OpenMP clause 'reduction' must reference the same object in all threads}}
@@ -181,14 +181,14 @@ int main(int argc, char **argv) {
   const int d = 5;       // expected-note 2 {{'d' defined here}}
   const int da[5] = {0}; // expected-note {{'da' defined here}}
   int qa[5] = {0};
-  S4 e(4); // expected-note {{'e' defined here}}
-  S5 g(5); // expected-note {{'g' defined here}}
+  S4 e(4);
+  S5 g(5);
   int i;
   int &j = i;                      // expected-note 2 {{'j' defined here}}
   S3 &p = k;                       // expected-note 2 {{'p' defined here}}
   const int &r = da[i];            // expected-note {{'r' defined here}}
   int &q = qa[i];                  // expected-note {{'q' defined here}}
-  float fl;                        // expected-note {{'fl' defined here}}
+  float fl;
 #pragma omp parallel for simd reduction // expected-error {{expected '(' after 'reduction'}}
   for (int i = 0; i < 10; ++i)
     foo();
@@ -228,7 +228,7 @@ int main(int argc, char **argv) {
 #pragma omp parallel for simd reduction(^ : S1) // expected-error {{'S1' does not refer to a value}}
   for (int i = 0; i < 10; ++i)
     foo();
-#pragma omp parallel for simd reduction(+ : a, b, c, d, f) // expected-error {{reduction variable with incomplete type 'S1'}} expected-error 2 {{const-qualified variable cannot be reduction}}
+#pragma omp parallel for simd reduction(+ : a, b, c, d, f) // expected-error {{reduction variable with incomplete type 'S1'}} expected-error 2 {{const-qualified variable cannot be reduction}} expected-error {{'operator+' is a private member of 'S2'}}
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp parallel for simd reduction(min : a, b, c, d, f) // expected-error {{reduction variable with incomplete type 'S1'}} expected-error 2 {{arguments of OpenMP clause 'reduction' for 'min' or 'max' must be of arithmetic type}} expected-error 2 {{const-qualified variable cannot be reduction}}
@@ -246,7 +246,7 @@ int main(int argc, char **argv) {
 #pragma omp parallel for simd reduction(- : da) // expected-error {{a reduction variable with array type 'const int [5]'}}
   for (int i = 0; i < 10; ++i)
     foo();
-#pragma omp parallel for simd reduction(^ : fl) // expected-error {{variable of type 'float' is not valid for specified reduction operation}}
+#pragma omp parallel for simd reduction(^ : fl) // expected-error {{invalid operands to binary expression ('float' and 'float')}}
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp parallel for simd reduction(&& : S2::S2s)
@@ -255,13 +255,13 @@ int main(int argc, char **argv) {
 #pragma omp parallel for simd reduction(&& : S2::S2sc) // expected-error {{const-qualified variable cannot be reduction}}
   for (int i = 0; i < 10; ++i)
     foo();
-#pragma omp parallel for simd reduction(& : e, g) // expected-error {{reduction variable must have an accessible, unambiguous default constructor}} expected-error {{variable of type 'S5' is not valid for specified reduction operation}}
+#pragma omp parallel for simd reduction(& : e, g) // expected-error {{calling a private constructor of class 'S4'}} expected-error {{invalid operands to binary expression ('S4' and 'S4')}} expected-error {{calling a private constructor of class 'S5'}} expected-error {{invalid operands to binary expression ('S5' and 'S5')}}
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp parallel for simd reduction(+ : h, k) // expected-error {{threadprivate or thread local variable cannot be reduction}}
   for (int i = 0; i < 10; ++i)
     foo();
-#pragma omp parallel for simd reduction(+ : o) // expected-error {{variable of type 'class S6' is not valid for specified reduction operation}}
+#pragma omp parallel for simd reduction(+ : o) // expected-error {{no viable overloaded '='}}
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp parallel for simd private(i), reduction(+ : j), reduction(+ : q) // expected-error 2 {{argument of OpenMP clause 'reduction' must reference the same object in all threads}}

clang/test/OpenMP/parallel_reduction_codegen.cpp

@@ -0,0 +1,703 @@
+// RUN: %clang_cc1 -verify -fopenmp=libiomp5 -x c++ -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck %s
+// RUN: %clang_cc1 -fopenmp=libiomp5 -x c++ -std=c++11 -triple x86_64-apple-darwin10 -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp=libiomp5 -x c++ -triple x86_64-apple-darwin10 -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s
+// RUN: %clang_cc1 -verify -fopenmp=libiomp5 -x c++ -std=c++11 -DLAMBDA -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=LAMBDA %s
+// RUN: %clang_cc1 -verify -fopenmp=libiomp5 -x c++ -fblocks -DBLOCKS -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=BLOCKS %s
+// expected-no-diagnostics
+#ifndef HEADER
+#define HEADER
+
+volatile int g = 1212;
+
+template <class T>
+struct S {
+  T f;
+  S(T a) : f(a + g) {}
+  S() : f(g) {}
+  operator T() { return T(); }
+  S &operator&(const S &) { return *this; }
+  ~S() {}
+};
+
+// CHECK-DAG: [[S_FLOAT_TY:%.+]] = type { float }
+// CHECK-DAG: [[S_INT_TY:%.+]] = type { i{{[0-9]+}} }
+// CHECK-DAG: [[CAP_MAIN_TY:%.+]] = type { [2 x i{{[0-9]+}}]*, float*, [2 x [[S_FLOAT_TY]]]*, [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]*, float* }
+// CHECK-DAG: [[CAP_TMAIN_TY:%.+]] = type { [2 x i{{[0-9]+}}]*, i{{[0-9]+}}*, [2 x [[S_INT_TY]]]*, [[S_INT_TY]]*, [[S_INT_TY]]*, i{{[0-9]+}}* }
+// CHECK-DAG: [[IMPLICIT_BARRIER_LOC:@.+]] = private unnamed_addr constant %{{.+}} { i32 0, i32 66, i32 0, i32 0, i8*
+// CHECK-DAG: [[REDUCTION_LOC:@.+]] = private unnamed_addr constant %{{.+}} { i32 0, i32 18, i32 0, i32 0, i8*
+// CHECK-DAG: [[REDUCTION_LOCK:@.+]] = common global [8 x i32] zeroinitializer
+
+template <typename T>
+T tmain() {
+  T t;
+  S<T> test;
+  T t_var = T(), t_var1;
+  T vec[] = {1, 2};
+  S<T> s_arr[] = {1, 2};
+  S<T> var(3), var1;
+#pragma omp parallel reduction(+:t_var) reduction(&:var) reduction(&& : var1) reduction(min: t_var1)
+  {
+    vec[0] = t_var;
+    s_arr[0] = var;
+  }
+  return T();
+}
+
+int main() {
+#ifdef LAMBDA
+  // LAMBDA: [[G:@.+]] = global i{{[0-9]+}} 1212,
+  // LAMBDA-LABEL: @main
+  // LAMBDA: call void [[OUTER_LAMBDA:@.+]](
+  [&]() {
+  // LAMBDA: define{{.*}} internal{{.*}} void [[OUTER_LAMBDA]](
+  // LAMBDA: [[G_LOCAL_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[AGG_CAPTURED:%.+]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
+  // LAMBDA: store i{{[0-9]+}}* [[G]], i{{[0-9]+}}** [[G_LOCAL_REF]]
+  // LAMBDA: [[ARG:%.+]] = bitcast %{{.+}}* [[AGG_CAPTURED]] to i8*
+  // LAMBDA: call void {{.+}}* @__kmpc_fork_call({{.+}}, i32 1, {{.+}}* [[OMP_REGION:@.+]] to {{.+}}, i8* [[ARG]])
+#pragma omp parallel reduction(+:g)
+  {
+    // LAMBDA: define{{.*}} internal{{.*}} void [[OMP_REGION]](i32* %{{.+}}, i32* %{{.+}}, %{{.+}}* [[ARG:%.+]])
+    // LAMBDA: [[G_PRIVATE_ADDR:%.+]] = alloca i{{[0-9]+}},
+
+    // Reduction list for runtime.
+    // LAMBDA: [[RED_LIST:%.+]] = alloca [1 x i8*],
+
+    // LAMBDA: store %{{.+}}* [[ARG]], %{{.+}}** [[ARG_REF:%.+]],
+    // LAMBDA: [[ARG:%.+]] = load %{{.+}}*, %{{.+}}** [[ARG_REF]]
+    // LAMBDA: [[G_REF_ADDR:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[ARG]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
+    // LAMBDA: [[G_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[G_REF_ADDR]]
+    // LAMBDA: store i{{[0-9]+}} 0, i{{[0-9]+}}* [[G_PRIVATE_ADDR]]
+    // LAMBDA: call i32 @__kmpc_cancel_barrier(
+    g = 1;
+    // LAMBDA: store volatile i{{[0-9]+}} 1, i{{[0-9]+}}* [[G_PRIVATE_ADDR]],
+    // LAMBDA: [[G_PRIVATE_ADDR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[ARG:%.+]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
+    // LAMBDA: store i{{[0-9]+}}* [[G_PRIVATE_ADDR]], i{{[0-9]+}}** [[G_PRIVATE_ADDR_REF]]
+    // LAMBDA: call void [[INNER_LAMBDA:@.+]](%{{.+}}* [[ARG]])
+
+    // LAMBDA: [[G_PRIV_REF:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[RED_LIST]], i32 0, i32 0
+    // LAMBDA: [[BITCAST:%.+]] = bitcast i32* [[G_PRIVATE_ADDR]] to i8*
+    // LAMBDA: store i8* [[BITCAST]], i8** [[G_PRIV_REF]],
+    // LAMBDA: call i32 @__kmpc_reduce_nowait(
+    // LAMBDA: switch i32 %{{.+}}, label %[[REDUCTION_DONE:.+]] [
+    // LAMBDA: i32 1, label %[[CASE1:.+]]
+    // LAMBDA: i32 2, label %[[CASE2:.+]]
+    // LAMBDA: [[CASE1]]
+    // LAMBDA: [[G_VAL:%.+]] = load i32, i32* [[G_REF]]
+    // LAMBDA: [[G_PRIV_VAL:%.+]] = load i32, i32* [[G_PRIVATE_ADDR]]
+    // LAMBDA: [[ADD:%.+]] = add nsw i32 [[G_VAL]], [[G_PRIV_VAL]]
+    // LAMBDA: store i32 [[ADD]], i32* [[G_REF]]
+    // LAMBDA: call void @__kmpc_end_reduce_nowait(
+    // LAMBDA: br label %[[REDUCTION_DONE]]
+    // LAMBDA: [[CASE2]]
+    // LAMBDA: [[G_PRIV_VAL:%.+]] = load i32, i32* [[G_PRIVATE_ADDR]]
+    // LAMBDA: atomicrmw add i32* [[G_REF]], i32 [[G_PRIV_VAL]] monotonic
+    // LAMBDA: br label %[[REDUCTION_DONE]]
+    // LAMBDA: [[REDUCTION_DONE]]
+    // LAMBDA: ret void
+    [&]() {
+      // LAMBDA: define {{.+}} void [[INNER_LAMBDA]](%{{.+}}* [[ARG_PTR:%.+]])
+      // LAMBDA: store %{{.+}}* [[ARG_PTR]], %{{.+}}** [[ARG_PTR_REF:%.+]],
+      g = 2;
+      // LAMBDA: [[ARG_PTR:%.+]] = load %{{.+}}*, %{{.+}}** [[ARG_PTR_REF]]
+      // LAMBDA: [[G_PTR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[ARG_PTR]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
+      // LAMBDA: [[G_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[G_PTR_REF]]
+      // LAMBDA: store volatile i{{[0-9]+}} 2, i{{[0-9]+}}* [[G_REF]]
+    }();
+  }
+  }();
+  return 0;
+#elif defined(BLOCKS)
+  // BLOCKS: [[G:@.+]] = global i{{[0-9]+}} 1212,
+  // BLOCKS-LABEL: @main
+  // BLOCKS: call void {{%.+}}(i8*
+  ^{
+  // BLOCKS: define{{.*}} internal{{.*}} void {{.+}}(i8*
+  // BLOCKS: [[G_LOCAL_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[AGG_CAPTURED:%.+]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
+  // BLOCKS: store i{{[0-9]+}}* [[G]], i{{[0-9]+}}** [[G_LOCAL_REF]]
+  // BLOCKS: [[ARG:%.+]] = bitcast %{{.+}}* [[AGG_CAPTURED]] to i8*
+  // BLOCKS: call void {{.+}}* @__kmpc_fork_call({{.+}}, i32 1, {{.+}}* [[OMP_REGION:@.+]] to {{.+}}, i8* [[ARG]])
+#pragma omp parallel reduction(-:g)
+  {
+    // BLOCKS: define{{.*}} internal{{.*}} void [[OMP_REGION]](i32* %{{.+}}, i32* %{{.+}}, %{{.+}}* [[ARG:%.+]])
+    // BLOCKS: [[G_PRIVATE_ADDR:%.+]] = alloca i{{[0-9]+}},
+
+    // Reduction list for runtime.
+    // BLOCKS: [[RED_LIST:%.+]] = alloca [1 x i8*],
+
+    // BLOCKS: store %{{.+}}* [[ARG]], %{{.+}}** [[ARG_REF:%.+]],
+    // BLOCKS: [[ARG:%.+]] = load %{{.+}}*, %{{.+}}** [[ARG_REF]]
+    // BLOCKS: [[G_REF_ADDR:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[ARG]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
+    // BLOCKS: [[G_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[G_REF_ADDR]]
+    // BLOCKS: store i{{[0-9]+}} 0, i{{[0-9]+}}* [[G_PRIVATE_ADDR]]
+    // BLOCKS: call i32 @__kmpc_cancel_barrier(
+    g = 1;
+    // BLOCKS: store volatile i{{[0-9]+}} 1, i{{[0-9]+}}* [[G_PRIVATE_ADDR]],
+    // BLOCKS-NOT: [[G]]{{[[^:word:]]}}
+    // BLOCKS: i{{[0-9]+}}* [[G_PRIVATE_ADDR]]
+    // BLOCKS-NOT: [[G]]{{[[^:word:]]}}
+    // BLOCKS: call void {{%.+}}(i8*
+
+    // BLOCKS: [[G_PRIV_REF:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[RED_LIST]], i32 0, i32 0
+    // BLOCKS: [[BITCAST:%.+]] = bitcast i32* [[G_PRIVATE_ADDR]] to i8*
+    // BLOCKS: store i8* [[BITCAST]], i8** [[G_PRIV_REF]],
+    // BLOCKS: call i32 @__kmpc_reduce_nowait(
+    // BLOCKS: switch i32 %{{.+}}, label %[[REDUCTION_DONE:.+]] [
+    // BLOCKS: i32 1, label %[[CASE1:.+]]
+    // BLOCKS: i32 2, label %[[CASE2:.+]]
+    // BLOCKS: [[CASE1]]
+    // BLOCKS: [[G_VAL:%.+]] = load i32, i32* [[G_REF]]
+    // BLOCKS: [[G_PRIV_VAL:%.+]] = load i32, i32* [[G_PRIVATE_ADDR]]
+    // BLOCKS: [[ADD:%.+]] = add nsw i32 [[G_VAL]], [[G_PRIV_VAL]]
+    // BLOCKS: store i32 [[ADD]], i32* [[G_REF]]
+    // BLOCKS: call void @__kmpc_end_reduce_nowait(
+    // BLOCKS: br label %[[REDUCTION_DONE]]
+    // BLOCKS: [[CASE2]]
+    // BLOCKS: [[G_PRIV_VAL:%.+]] = load i32, i32* [[G_PRIVATE_ADDR]]
+    // BLOCKS: atomicrmw add i32* [[G_REF]], i32 [[G_PRIV_VAL]] monotonic
+    // BLOCKS: br label %[[REDUCTION_DONE]]
+    // BLOCKS: [[REDUCTION_DONE]]
+    // BLOCKS: ret void
+    ^{
+      // BLOCKS: define {{.+}} void {{@.+}}(i8*
+      g = 2;
+      // BLOCKS-NOT: [[G]]{{[[^:word:]]}}
+      // BLOCKS: store volatile i{{[0-9]+}} 2, i{{[0-9]+}}*
+      // BLOCKS-NOT: [[G]]{{[[^:word:]]}}
+      // BLOCKS: ret
+    }();
+  }
+  }();
+  return 0;
+#else
+  S<float> test;
+  float t_var = 0, t_var1;
+  int vec[] = {1, 2};
+  S<float> s_arr[] = {1, 2};
+  S<float> var(3), var1;
+#pragma omp parallel reduction(+:t_var) reduction(&:var) reduction(&& : var1) reduction(min: t_var1)
+  {
+    vec[0] = t_var;
+    s_arr[0] = var;
+  }
+  return tmain<int>();
+#endif
+}
+
+// CHECK: define {{.*}}i{{[0-9]+}} @main()
+// CHECK: [[TEST:%.+]] = alloca [[S_FLOAT_TY]],
+// CHECK: call {{.*}} [[S_FLOAT_TY_CONSTR:@.+]]([[S_FLOAT_TY]]* [[TEST]])
+// CHECK: %{{.+}} = bitcast [[CAP_MAIN_TY]]*
+// CHECK: call void (%{{.+}}*, i{{[0-9]+}}, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)*, ...)* @__kmpc_fork_call(%{{.+}}* @{{.+}}, i{{[0-9]+}} 1, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)* bitcast (void (i{{[0-9]+}}*, i{{[0-9]+}}*, [[CAP_MAIN_TY]]*)* [[MAIN_MICROTASK:@.+]] to void
+// CHECK: = call {{.*}}i{{.+}} [[TMAIN_INT:@.+]]()
+// CHECK: call {{.*}} [[S_FLOAT_TY_DESTR:@.+]]([[S_FLOAT_TY]]*
+// CHECK: ret
+//
+// CHECK: define internal void [[MAIN_MICROTASK]](i{{[0-9]+}}* [[GTID_ADDR:%.+]], i{{[0-9]+}}* %{{.+}}, [[CAP_MAIN_TY]]* %{{.+}})
+// CHECK: [[T_VAR_PRIV:%.+]] = alloca float,
+// CHECK: [[VAR_PRIV:%.+]] = alloca [[S_FLOAT_TY]],
+// CHECK: [[VAR1_PRIV:%.+]] = alloca [[S_FLOAT_TY]],
+// CHECK: [[T_VAR1_PRIV:%.+]] = alloca float,
+
+// Reduction list for runtime.
+// CHECK: [[RED_LIST:%.+]] = alloca [4 x i8*],
+
+// CHECK: store i{{[0-9]+}}* [[GTID_ADDR]], i{{[0-9]+}}** [[GTID_ADDR_ADDR:%.+]],
+
+// CHECK: [[T_VAR_PTR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} {{[0-9]+}}
+// CHECK: [[T_VAR_REF:%.+]] = load float*, float** [[T_VAR_PTR_REF]],
+// For + reduction operation initial value of private variable is 0.
+// CHECK: store float 0.0{{.+}}, float* [[T_VAR_PRIV]],
+
+// CHECK: [[VAR_PTR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} {{[0-9]+}}
+// CHECK: [[VAR_REF:%.+]] = load [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** [[VAR_PTR_REF:%.+]],
+// For & reduction operation initial value of private variable is ones in all bits.
+// CHECK: call {{.*}} [[S_FLOAT_TY_CONSTR:@.+]]([[S_FLOAT_TY]]* [[VAR_PRIV]])
+
+// CHECK: [[VAR1_PTR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} {{[0-9]+}}
+// CHECK: [[VAR1_REF:%.+]] = load [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** [[VAR_PTR_REF:%.+]],
+// For && reduction operation initial value of private variable is 1.0.
+// CHECK: call {{.*}} [[S_FLOAT_TY_CONSTR:@.+]]([[S_FLOAT_TY]]* [[VAR1_PRIV]])
+
+// CHECK: [[T_VAR1_PTR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} {{[0-9]+}}
+// CHECK: [[T_VAR1_REF:%.+]] = load float*, float** [[T_VAR1_PTR_REF]],
+// For min reduction operation initial value of private variable is largest repesentable value.
+// CHECK: store float 0x47EFFFFFE0000000, float* [[T_VAR1_PRIV]],
+
+// CHECK: [[GTID_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[GTID_ADDR_ADDR]]
+// CHECK: [[GTID:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[GTID_REF]]
+// CHECK: call i32 @__kmpc_cancel_barrier(%{{.+}}* [[IMPLICIT_BARRIER_LOC]], i{{[0-9]+}} [[GTID]])
+
+// Skip checks for internal operations.
+
+// void *RedList[<n>] = {<ReductionVars>[0], ..., <ReductionVars>[<n>-1]};
+
+// CHECK: [[T_VAR_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i32 0, i32 0
+// CHECK: [[BITCAST:%.+]] = bitcast float* [[T_VAR_PRIV]] to i8*
+// CHECK: store i8* [[BITCAST]], i8** [[T_VAR_PRIV_REF]],
+// CHECK: [[VAR_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i32 0, i32 1
+// CHECK: [[BITCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR_PRIV]] to i8*
+// CHECK: store i8* [[BITCAST]], i8** [[VAR_PRIV_REF]],
+// CHECK: [[VAR1_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i32 0, i32 2
+// CHECK: [[BITCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR1_PRIV]] to i8*
+// CHECK: store i8* [[BITCAST]], i8** [[VAR1_PRIV_REF]],
+// CHECK: [[T_VAR1_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i32 0, i32 3
+// CHECK: [[BITCAST:%.+]] = bitcast float* [[T_VAR1_PRIV]] to i8*
+// CHECK: store i8* [[BITCAST]], i8** [[T_VAR1_PRIV_REF]],
+
+// res = __kmpc_reduce_nowait(<loc>, <gtid>, <n>, sizeof(RedList), RedList, reduce_func, &<lock>);
+
+// CHECK: [[BITCAST:%.+]] = bitcast [4 x i8*]* [[RED_LIST]] to i8*
+// CHECK: [[RES:%.+]] = call i32 @__kmpc_reduce_nowait(%{{.+}}* [[REDUCTION_LOC]], i32 [[GTID]], i32 4, i64 32, i8* [[BITCAST]], void (i8*, i8*)* [[REDUCTION_FUNC:@.+]], [8 x i32]* [[REDUCTION_LOCK]])
+
+// switch(res)
+// CHECK: switch i32 [[RES]], label %[[RED_DONE:.+]] [
+// CHECK: i32 1, label %[[CASE1:.+]]
+// CHECK: i32 2, label %[[CASE2:.+]]
+// CHECK: ]
+
+// case 1:
+// t_var += t_var_reduction;
+// CHECK: [[T_VAR_VAL:%.+]] = load float, float* [[T_VAR_REF]],
+// CHECK: [[T_VAR_PRIV_VAL:%.+]] = load float, float* [[T_VAR_PRIV]],
+// CHECK: [[UP:%.+]] = fadd float [[T_VAR_VAL]], [[T_VAR_PRIV_VAL]]
+// CHECK: store float [[UP]], float* [[T_VAR_REF]],
+
+// var = var.operator &(var_reduction);
+// CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_FLOAT_TY]]* @{{.+}}([[S_FLOAT_TY]]* [[VAR_REF]], [[S_FLOAT_TY]]* dereferenceable(4) [[VAR_PRIV]])
+// CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR_REF]] to i8*
+// CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[UP]] to i8*
+// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false)
+
+// var1 = var1.operator &&(var1_reduction);
+// CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_REF]])
+// CHECK: [[VAR1_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0
+// CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[FALSE:.+]]
+// CHECK: [[TRUE]]
+// CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_PRIV]])
+// CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0
+// CHECK: br i1 [[VAR1_REDUCTION_BOOL]], label %[[TRUE2:.+]], label %[[FALSE2:.+]]
+// CHECK: [[TRUE2]]
+// CHECK: br label %[[END2:.+]]
+// CHECK: [[FALSE2]]
+// CHECK: br label %[[END2]]
+// CHECK: [[END2]]
+// CHECK: [[COND_LVALUE:%.+]] = phi [[S_FLOAT_TY]]* [ [[VAR1_REF]], %[[TRUE2]] ], [ [[VAR1_PRIV]], %[[FALSE2]] ]
+// CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR1_REF]] to i8*
+// CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[COND_LVALUE]] to i8*
+// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false)
+
+// t_var1 = min(t_var1, t_var1_reduction);
+// CHECK: [[T_VAR1_VAL:%.+]] = load float, float* [[T_VAR1_REF]],
+// CHECK: [[T_VAR1_PRIV_VAL:%.+]] = load float, float* [[T_VAR1_PRIV]],
+// CHECK: [[CMP:%.+]] = fcmp olt float [[T_VAR1_VAL]], [[T_VAR1_PRIV_VAL]]
+// CHECK: [[UP:%.+]] = uitofp i1 [[CMP]] to float
+// CHECK: store float [[UP]], float* [[T_VAR1_REF]],
+
+// __kmpc_end_reduce_nowait(<loc>, <gtid>, &<lock>);
+// CHECK: call void @__kmpc_end_reduce_nowait(%{{.+}}* [[REDUCTION_LOC]], i32 [[GTID]], [8 x i32]* [[REDUCTION_LOCK]])
+
+// break;
+// CHECK: br label %[[RED_DONE]]
+
+// case 2:
+// t_var += t_var_reduction;
+// CHECK: load float, float* [[T_VAR_PRIV]]
+// CHECK: [[T_VAR_REF_INT:%.+]] = bitcast float* [[T_VAR_REF]] to i32*
+// CHECK: load atomic i32, i32* [[T_VAR_REF_INT]] monotonic,
+// CHECK: [[OLD1:%.+]] = bitcast i32 %{{.+}} to float
+// CHECK: br label %[[CONT:.+]]
+// CHECK: [[CONT]]
+// CHECK: [[ORIG_OLD:%.+]] = phi float [ [[OLD1]], %{{.+}} ], [ [[OLD2:%.+]], %[[CONT]] ]
+// CHECK: [[UP:%.+]] = fadd float
+// CHECK: [[ORIG_OLD_INT:%.+]] = bitcast float [[ORIG_OLD]] to i32
+// CHECK: [[UP_INT:%.+]] = bitcast float [[UP]] to i32
+// CHECK: [[T_VAR_REF_INT:%.+]] = bitcast float* [[T_VAR_REF]] to i32*
+// CHECK: [[RES:%.+]] = cmpxchg i32* [[T_VAR_REF_INT]], i32 [[ORIG_OLD_INT]], i32 [[UP_INT]] monotonic monotonic
+// CHECK: [[OLD_VAL_INT:%.+]] = extractvalue { i32, i1 } [[RES]], 0
+// CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i32, i1 } [[RES]], 1
+// CHECK: [[OLD2]] = bitcast i32 [[OLD_VAL_INT]] to float
+// CHECK: br i1 [[SUCCESS_FAIL]], label %[[ATOMIC_DONE:.+]], label %[[CONT]]
+// CHECK: [[ATOMIC_DONE]]
+
+// var = var.operator &(var_reduction);
+// CHECK: call void @__kmpc_critical(
+// CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_FLOAT_TY]]* @{{.+}}([[S_FLOAT_TY]]* [[VAR_REF]], [[S_FLOAT_TY]]* dereferenceable(4) [[VAR_PRIV]])
+// CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR_REF]] to i8*
+// CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[UP]] to i8*
+// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false)
+// CHECK: call void @__kmpc_end_critical(
+
+// var1 = var1.operator &&(var1_reduction);
+// CHECK: call void @__kmpc_critical(
+// CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_REF]])
+// CHECK: [[VAR1_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0
+// CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[FALSE:.+]]
+// CHECK: [[TRUE]]
+// CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_PRIV]])
+// CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0
+// CHECK: br i1 [[VAR1_REDUCTION_BOOL]], label %[[TRUE2:.+]], label %[[FALSE2:.+]]
+// CHECK: [[TRUE2]]
+// CHECK: br label %[[END2:.+]]
+// CHECK: [[FALSE2]]
+// CHECK: br label %[[END2]]
+// CHECK: [[END2]]
+// CHECK: [[COND_LVALUE:%.+]] = phi [[S_FLOAT_TY]]* [ [[VAR1_REF]], %[[TRUE2]] ], [ [[VAR1_PRIV]], %[[FALSE2]] ]
+// CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR1_REF]] to i8*
+// CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[COND_LVALUE]] to i8*
+// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false)
+// CHECK: call void @__kmpc_end_critical(
+
+// t_var1 = min(t_var1, t_var1_reduction);
+// CHECK: load float, float* [[T_VAR1_PRIV]]
+// CHECK: [[T_VAR1_REF_INT:%.+]] = bitcast float* [[T_VAR1_REF]] to i32*
+// CHECK: load atomic i32, i32* [[T_VAR1_REF_INT]] monotonic,
+// CHECK: [[OLD1:%.+]] = bitcast i32 %{{.+}} to float
+// CHECK: br label %[[CONT:.+]]
+// CHECK: [[CONT]]
+// CHECK: [[ORIG_OLD:%.+]] = phi float [ [[OLD1]], %{{.+}} ], [ [[OLD2:%.+]], %[[CONT]] ]
+// CHECK: [[CMP:%.+]] = fcmp olt float
+// CHECK: [[UP:%.+]] = uitofp i1 [[CMP]] to float
+// CHECK: [[ORIG_OLD_INT:%.+]] = bitcast float [[ORIG_OLD]] to i32
+// CHECK: [[UP_INT:%.+]] = bitcast float [[UP]] to i32
+// CHECK: [[T_VAR1_REF_INT:%.+]] = bitcast float* [[T_VAR1_REF]] to i32*
+// CHECK: [[RES:%.+]] = cmpxchg i32* [[T_VAR1_REF_INT]], i32 [[ORIG_OLD_INT]], i32 [[UP_INT]] monotonic monotonic
+// CHECK: [[OLD_VAL_INT:%.+]] = extractvalue { i32, i1 } [[RES]], 0
+// CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i32, i1 } [[RES]], 1
+// CHECK: [[OLD2]] = bitcast i32 [[OLD_VAL_INT]] to float
+// CHECK: br i1 [[SUCCESS_FAIL]], label %[[ATOMIC_DONE:.+]], label %[[CONT]]
+// CHECK: [[ATOMIC_DONE]]
+
+// break;
+// CHECK: br label %[[RED_DONE]]
+// CHECK: [[RED_DONE]]
+// CHECK: call i32 @__kmpc_cancel_barrier(%{{.+}}* [[IMPLICIT_BARRIER_LOC]], i{{[0-9]+}} [[GTID]])
+
+// CHECK-DAG: call {{.*}} [[S_FLOAT_TY_DESTR]]([[S_FLOAT_TY]]* [[VAR_PRIV]])
+// CHECK-DAG: call {{.*}} [[S_FLOAT_TY_DESTR]]([[S_FLOAT_TY]]*
+// CHECK: ret void
+
+// void reduce_func(void *lhs[<n>], void *rhs[<n>]) {
+//  *(Type0*)lhs[0] = ReductionOperation0(*(Type0*)lhs[0], *(Type0*)rhs[0]);
+//  ...
+//  *(Type<n>-1*)lhs[<n>-1] = ReductionOperation<n>-1(*(Type<n>-1*)lhs[<n>-1],
+//  *(Type<n>-1*)rhs[<n>-1]);
+// }
+// CHECK: define internal void [[REDUCTION_FUNC]](i8*, i8*)
+// t_var_lhs = (float*)lhs[0];
+// CHECK: [[T_VAR_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS:%.+]], i32 0, i32 0
+// CHECK: [[T_VAR_RHS_VOID:%.+]] = load i8*, i8** [[T_VAR_RHS_REF]],
+// CHECK: [[T_VAR_RHS:%.+]] = bitcast i8* [[T_VAR_RHS_VOID]] to float*
+// t_var_rhs = (float*)rhs[0];
+// CHECK: [[T_VAR_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS:%.+]], i32 0, i32 0
+// CHECK: [[T_VAR_LHS_VOID:%.+]] = load i8*, i8** [[T_VAR_LHS_REF]],
+// CHECK: [[T_VAR_LHS:%.+]] = bitcast i8* [[T_VAR_LHS_VOID]] to float*
+
+// var_lhs = (S<float>*)lhs[1];
+// CHECK: [[VAR_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS]], i32 0, i32 1
+// CHECK: [[VAR_RHS_VOID:%.+]] = load i8*, i8** [[VAR_RHS_REF]],
+// CHECK: [[VAR_RHS:%.+]] = bitcast i8* [[VAR_RHS_VOID]] to [[S_FLOAT_TY]]*
+// var_rhs = (S<float>*)rhs[1];
+// CHECK: [[VAR_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i32 0, i32 1
+// CHECK: [[VAR_LHS_VOID:%.+]] = load i8*, i8** [[VAR_LHS_REF]],
+// CHECK: [[VAR_LHS:%.+]] = bitcast i8* [[VAR_LHS_VOID]] to [[S_FLOAT_TY]]*
+
+// var1_lhs = (S<float>*)lhs[2];
+// CHECK: [[VAR1_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS]], i32 0, i32 2
+// CHECK: [[VAR1_RHS_VOID:%.+]] = load i8*, i8** [[VAR1_RHS_REF]],
+// CHECK: [[VAR1_RHS:%.+]] = bitcast i8* [[VAR1_RHS_VOID]] to [[S_FLOAT_TY]]*
+// var1_rhs = (S<float>*)rhs[2];
+// CHECK: [[VAR1_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i32 0, i32 2
+// CHECK: [[VAR1_LHS_VOID:%.+]] = load i8*, i8** [[VAR1_LHS_REF]],
+// CHECK: [[VAR1_LHS:%.+]] = bitcast i8* [[VAR1_LHS_VOID]] to [[S_FLOAT_TY]]*
+
+// t_var1_lhs = (float*)lhs[3];
+// CHECK: [[T_VAR1_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS]], i32 0, i32 3
+// CHECK: [[T_VAR1_RHS_VOID:%.+]] = load i8*, i8** [[T_VAR1_RHS_REF]],
+// CHECK: [[T_VAR1_RHS:%.+]] = bitcast i8* [[T_VAR1_RHS_VOID]] to float*
+// t_var1_rhs = (float*)rhs[3];
+// CHECK: [[T_VAR1_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i32 0, i32 3
+// CHECK: [[T_VAR1_LHS_VOID:%.+]] = load i8*, i8** [[T_VAR1_LHS_REF]],
+// CHECK: [[T_VAR1_LHS:%.+]] = bitcast i8* [[T_VAR1_LHS_VOID]] to float*
+
+// t_var_lhs += t_var_rhs;
+// CHECK: [[T_VAR_LHS_VAL:%.+]] = load float, float* [[T_VAR_LHS]],
+// CHECK: [[T_VAR_RHS_VAL:%.+]] = load float, float* [[T_VAR_RHS]],
+// CHECK: [[UP:%.+]] = fadd float [[T_VAR_LHS_VAL]], [[T_VAR_RHS_VAL]]
+// CHECK: store float [[UP]], float* [[T_VAR_LHS]],
+
+// var_lhs = var_lhs.operator &(var_rhs);
+// CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_FLOAT_TY]]* @{{.+}}([[S_FLOAT_TY]]* [[VAR_LHS]], [[S_FLOAT_TY]]* dereferenceable(4) [[VAR_RHS]])
+// CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR_LHS]] to i8*
+// CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[UP]] to i8*
+// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false)
+
+// var1_lhs = var1_lhs.operator &&(var1_rhs);
+// CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_LHS]])
+// CHECK: [[VAR1_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0
+// CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[FALSE:.+]]
+// CHECK: [[TRUE]]
+// CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_RHS]])
+// CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0
+// CHECK: br i1 [[VAR1_REDUCTION_BOOL]], label %[[TRUE2:.+]], label %[[FALSE2:.+]]
+// CHECK: [[TRUE2]]
+// CHECK: br label %[[END2:.+]]
+// CHECK: [[FALSE2]]
+// CHECK: br label %[[END2]]
+// CHECK: [[END2]]
+// CHECK: [[COND_LVALUE:%.+]] = phi [[S_FLOAT_TY]]* [ [[VAR1_LHS]], %[[TRUE2]] ], [ [[VAR1_RHS]], %[[FALSE2]] ]
+// CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR1_LHS]] to i8*
+// CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[COND_LVALUE]] to i8*
+// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false)
+
+// t_var1_lhs = min(t_var1_lhs, t_var1_rhs);
+// CHECK: [[T_VAR1_LHS_VAL:%.+]] = load float, float* [[T_VAR1_LHS]],
+// CHECK: [[T_VAR1_RHS_VAL:%.+]] = load float, float* [[T_VAR1_RHS]],
+// CHECK: [[CMP:%.+]] = fcmp olt float [[T_VAR1_LHS_VAL]], [[T_VAR1_RHS_VAL]]
+// CHECK: [[UP:%.+]] = uitofp i1 [[CMP]] to float
+// CHECK: store float [[UP]], float* [[T_VAR1_LHS]],
+// CHECK: ret void
+
+// CHECK: define {{.*}} i{{[0-9]+}} [[TMAIN_INT]]()
+// CHECK: [[TEST:%.+]] = alloca [[S_INT_TY]],
+// CHECK: call {{.*}} [[S_INT_TY_CONSTR:@.+]]([[S_INT_TY]]* [[TEST]])
+// CHECK: %{{.+}} = bitcast [[CAP_TMAIN_TY]]*
+// CHECK: call void (%{{.+}}*, i{{[0-9]+}}, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)*, ...)* @__kmpc_fork_call(%{{.+}}* @{{.+}}, i{{[0-9]+}} 1, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)* bitcast (void (i{{[0-9]+}}*, i{{[0-9]+}}*, [[CAP_TMAIN_TY]]*)* [[TMAIN_MICROTASK:@.+]] to void
+// CHECK: call {{.*}} [[S_INT_TY_DESTR:@.+]]([[S_INT_TY]]*
+// CHECK: ret
+//
+// CHECK: define internal void [[TMAIN_MICROTASK]](i{{[0-9]+}}* [[GTID_ADDR:%.+]], i{{[0-9]+}}* %{{.+}}, [[CAP_TMAIN_TY]]* %{{.+}})
+// CHECK: [[T_VAR_PRIV:%.+]] = alloca i{{[0-9]+}},
+// CHECK: [[VAR_PRIV:%.+]] = alloca [[S_INT_TY]],
+// CHECK: [[VAR1_PRIV:%.+]] = alloca [[S_INT_TY]],
+// CHECK: [[T_VAR1_PRIV:%.+]] = alloca i{{[0-9]+}},
+
+// Reduction list for runtime.
+// CHECK: [[RED_LIST:%.+]] = alloca [4 x i8*],
+
+// CHECK: store i{{[0-9]+}}* [[GTID_ADDR]], i{{[0-9]+}}** [[GTID_ADDR_ADDR:%.+]],
+
+// CHECK: [[T_VAR_PTR_REF:%.+]] = getelementptr inbounds [[CAP_TMAIN_TY]], [[CAP_TMAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} {{[0-9]+}}
+// CHECK: [[T_VAR_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[T_VAR_PTR_REF]],
+// For + reduction operation initial value of private variable is 0.
+// CHECK: store i{{[0-9]+}} 0, i{{[0-9]+}}* [[T_VAR_PRIV]],
+
+// CHECK: [[VAR_PTR_REF:%.+]] = getelementptr inbounds [[CAP_TMAIN_TY]], [[CAP_TMAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} {{[0-9]+}}
+// CHECK: [[VAR_REF:%.+]] = load [[S_INT_TY]]*, [[S_INT_TY]]** [[VAR_PTR_REF:%.+]],
+// For & reduction operation initial value of private variable is ones in all bits.
+// CHECK: call {{.*}} [[S_INT_TY_CONSTR:@.+]]([[S_INT_TY]]* [[VAR_PRIV]])
+
+// CHECK: [[VAR1_PTR_REF:%.+]] = getelementptr inbounds [[CAP_TMAIN_TY]], [[CAP_TMAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} {{[0-9]+}}
+// CHECK: [[VAR1_REF:%.+]] = load [[S_INT_TY]]*, [[S_INT_TY]]** [[VAR_PTR_REF:%.+]],
+// For && reduction operation initial value of private variable is 1.0.
+// CHECK: call {{.*}} [[S_INT_TY_CONSTR:@.+]]([[S_INT_TY]]* [[VAR1_PRIV]])
+
+// CHECK: [[T_VAR1_PTR_REF:%.+]] = getelementptr inbounds [[CAP_TMAIN_TY]], [[CAP_TMAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} {{[0-9]+}}
+// CHECK: [[T_VAR1_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[T_VAR1_PTR_REF]],
+// For min reduction operation initial value of private variable is largest repesentable value.
+// CHECK: store i{{[0-9]+}} 2147483647, i{{[0-9]+}}* [[T_VAR1_PRIV]],
+
+// CHECK: [[GTID_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[GTID_ADDR_ADDR]]
+// CHECK: [[GTID:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[GTID_REF]]
+// CHECK: call i32 @__kmpc_cancel_barrier(%{{.+}}* [[IMPLICIT_BARRIER_LOC]], i{{[0-9]+}} [[GTID]])
+
+// Skip checks for internal operations.
+
+// void *RedList[<n>] = {<ReductionVars>[0], ..., <ReductionVars>[<n>-1]};
+
+// CHECK: [[T_VAR_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i32 0, i32 0
+// CHECK: [[BITCAST:%.+]] = bitcast i{{[0-9]+}}* [[T_VAR_PRIV]] to i8*
+// CHECK: store i8* [[BITCAST]], i8** [[T_VAR_PRIV_REF]],
+// CHECK: [[VAR_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i32 0, i32 1
+// CHECK: [[BITCAST:%.+]] = bitcast [[S_INT_TY]]* [[VAR_PRIV]] to i8*
+// CHECK: store i8* [[BITCAST]], i8** [[VAR_PRIV_REF]],
+// CHECK: [[VAR1_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i32 0, i32 2
+// CHECK: [[BITCAST:%.+]] = bitcast [[S_INT_TY]]* [[VAR1_PRIV]] to i8*
+// CHECK: store i8* [[BITCAST]], i8** [[VAR1_PRIV_REF]],
+// CHECK: [[T_VAR1_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i32 0, i32 3
+// CHECK: [[BITCAST:%.+]] = bitcast i{{[0-9]+}}* [[T_VAR1_PRIV]] to i8*
+// CHECK: store i8* [[BITCAST]], i8** [[T_VAR1_PRIV_REF]],
+
+// res = __kmpc_reduce_nowait(<loc>, <gtid>, <n>, sizeof(RedList), RedList, reduce_func, &<lock>);
+
+// CHECK: [[BITCAST:%.+]] = bitcast [4 x i8*]* [[RED_LIST]] to i8*
+// CHECK: [[RES:%.+]] = call i32 @__kmpc_reduce_nowait(%{{.+}}* [[REDUCTION_LOC]], i32 [[GTID]], i32 4, i64 32, i8* [[BITCAST]], void (i8*, i8*)* [[REDUCTION_FUNC:@.+]], [8 x i32]* [[REDUCTION_LOCK]])
+
+// switch(res)
+// CHECK: switch i32 [[RES]], label %[[RED_DONE:.+]] [
+// CHECK: i32 1, label %[[CASE1:.+]]
+// CHECK: i32 2, label %[[CASE2:.+]]
+// CHECK: ]
+
+// case 1:
+// t_var += t_var_reduction;
+// CHECK: [[T_VAR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR_REF]],
+// CHECK: [[T_VAR_PRIV_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR_PRIV]],
+// CHECK: [[UP:%.+]] = add nsw i{{[0-9]+}} [[T_VAR_VAL]], [[T_VAR_PRIV_VAL]]
+// CHECK: store i{{[0-9]+}} [[UP]], i{{[0-9]+}}* [[T_VAR_REF]],
+
+// var = var.operator &(var_reduction);
+// CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_INT_TY]]* @{{.+}}([[S_INT_TY]]* [[VAR_REF]], [[S_INT_TY]]* dereferenceable(4) [[VAR_PRIV]])
+// CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR_REF]] to i8*
+// CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[UP]] to i8*
+// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false)
+
+// var1 = var1.operator &&(var1_reduction);
+// CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_REF]])
+// CHECK: [[VAR1_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0
+// CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[FALSE:.+]]
+// CHECK: [[TRUE]]
+// CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_PRIV]])
+// CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0
+// CHECK: br i1 [[VAR1_REDUCTION_BOOL]], label %[[TRUE2:.+]], label %[[FALSE2:.+]]
+// CHECK: [[TRUE2]]
+// CHECK: br label %[[END2:.+]]
+// CHECK: [[FALSE2]]
+// CHECK: br label %[[END2]]
+// CHECK: [[END2]]
+// CHECK: [[COND_LVALUE:%.+]] = phi [[S_INT_TY]]* [ [[VAR1_REF]], %[[TRUE2]] ], [ [[VAR1_PRIV]], %[[FALSE2]] ]
+// CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR1_REF]] to i8*
+// CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[COND_LVALUE]] to i8*
+// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false)
+
+// t_var1 = min(t_var1, t_var1_reduction);
+// CHECK: [[T_VAR1_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_REF]],
+// CHECK: [[T_VAR1_PRIV_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_PRIV]],
+// CHECK: [[CMP:%.+]] = icmp slt i{{[0-9]+}} [[T_VAR1_VAL]], [[T_VAR1_PRIV_VAL]]
+// CHECK: [[UP:%.+]] = zext i1 [[CMP]] to i{{[0-9]+}}
+// CHECK: store i{{[0-9]+}} [[UP]], i{{[0-9]+}}* [[T_VAR1_REF]],
+
+// __kmpc_end_reduce_nowait(<loc>, <gtid>, &<lock>);
+// CHECK: call void @__kmpc_end_reduce_nowait(%{{.+}}* [[REDUCTION_LOC]], i32 [[GTID]], [8 x i32]* [[REDUCTION_LOCK]])
+
+// break;
+// CHECK: br label %[[RED_DONE]]
+
+// case 2:
+// t_var += t_var_reduction;
+// CHECK: [[T_VAR_PRIV_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR_PRIV]]
+// CHECK: atomicrmw add i32* [[T_VAR_REF]], i32 [[T_VAR_PRIV_VAL]] monotonic
+
+// var = var.operator &(var_reduction);
+// CHECK: call void @__kmpc_critical(
+// CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_INT_TY]]* @{{.+}}([[S_INT_TY]]* [[VAR_REF]], [[S_INT_TY]]* dereferenceable(4) [[VAR_PRIV]])
+// CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR_REF]] to i8*
+// CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[UP]] to i8*
+// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false)
+// CHECK: call void @__kmpc_end_critical(
+
+// var1 = var1.operator &&(var1_reduction);
+// CHECK: call void @__kmpc_critical(
+// CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_REF]])
+// CHECK: [[VAR1_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0
+// CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[FALSE:.+]]
+// CHECK: [[TRUE]]
+// CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_PRIV]])
+// CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0
+// CHECK: br i1 [[VAR1_REDUCTION_BOOL]], label %[[TRUE2:.+]], label %[[FALSE2:.+]]
+// CHECK: [[TRUE2]]
+// CHECK: br label %[[END2:.+]]
+// CHECK: [[FALSE2]]
+// CHECK: br label %[[END2]]
+// CHECK: [[END2]]
+// CHECK: [[COND_LVALUE:%.+]] = phi [[S_INT_TY]]* [ [[VAR1_REF]], %[[TRUE2]] ], [ [[VAR1_PRIV]], %[[FALSE2]] ]
+// CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR1_REF]] to i8*
+// CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[COND_LVALUE]] to i8*
+// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false)
+// CHECK: call void @__kmpc_end_critical(
+
+// t_var1 = min(t_var1, t_var1_reduction);
+// CHECK: [[T_VAR1_PRIV_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_PRIV]]
+// CHECK: atomicrmw min i32* [[T_VAR1_REF]], i32 [[T_VAR1_PRIV_VAL]] monotonic
+
+// break;
+// CHECK: br label %[[RED_DONE]]
+// CHECK: [[RED_DONE]]
+// CHECK: call i32 @__kmpc_cancel_barrier(%{{.+}}* [[IMPLICIT_BARRIER_LOC]], i{{[0-9]+}} [[GTID]])
+
+// CHECK-DAG: call {{.*}} [[S_INT_TY_DESTR]]([[S_INT_TY]]* [[VAR_PRIV]])
+// CHECK-DAG: call {{.*}} [[S_INT_TY_DESTR]]([[S_INT_TY]]*
+// CHECK: ret void
+
+// void reduce_func(void *lhs[<n>], void *rhs[<n>]) {
+//  *(Type0*)lhs[0] = ReductionOperation0(*(Type0*)lhs[0], *(Type0*)rhs[0]);
+//  ...
+//  *(Type<n>-1*)lhs[<n>-1] = ReductionOperation<n>-1(*(Type<n>-1*)lhs[<n>-1],
+//  *(Type<n>-1*)rhs[<n>-1]);
+// }
+// CHECK: define internal void [[REDUCTION_FUNC]](i8*, i8*)
+// t_var_lhs = (i{{[0-9]+}}*)lhs[0];
+// CHECK: [[T_VAR_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS:%.+]], i32 0, i32 0
+// CHECK: [[T_VAR_RHS_VOID:%.+]] = load i8*, i8** [[T_VAR_RHS_REF]],
+// CHECK: [[T_VAR_RHS:%.+]] = bitcast i8* [[T_VAR_RHS_VOID]] to i{{[0-9]+}}*
+// t_var_rhs = (i{{[0-9]+}}*)rhs[0];
+// CHECK: [[T_VAR_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS:%.+]], i32 0, i32 0
+// CHECK: [[T_VAR_LHS_VOID:%.+]] = load i8*, i8** [[T_VAR_LHS_REF]],
+// CHECK: [[T_VAR_LHS:%.+]] = bitcast i8* [[T_VAR_LHS_VOID]] to i{{[0-9]+}}*
+
+// var_lhs = (S<i{{[0-9]+}}>*)lhs[1];
+// CHECK: [[VAR_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS]], i32 0, i32 1
+// CHECK: [[VAR_RHS_VOID:%.+]] = load i8*, i8** [[VAR_RHS_REF]],
+// CHECK: [[VAR_RHS:%.+]] = bitcast i8* [[VAR_RHS_VOID]] to [[S_INT_TY]]*
+// var_rhs = (S<i{{[0-9]+}}>*)rhs[1];
+// CHECK: [[VAR_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i32 0, i32 1
+// CHECK: [[VAR_LHS_VOID:%.+]] = load i8*, i8** [[VAR_LHS_REF]],
+// CHECK: [[VAR_LHS:%.+]] = bitcast i8* [[VAR_LHS_VOID]] to [[S_INT_TY]]*
+
+// var1_lhs = (S<i{{[0-9]+}}>*)lhs[2];
+// CHECK: [[VAR1_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS]], i32 0, i32 2
+// CHECK: [[VAR1_RHS_VOID:%.+]] = load i8*, i8** [[VAR1_RHS_REF]],
+// CHECK: [[VAR1_RHS:%.+]] = bitcast i8* [[VAR1_RHS_VOID]] to [[S_INT_TY]]*
+// var1_rhs = (S<i{{[0-9]+}}>*)rhs[2];
+// CHECK: [[VAR1_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i32 0, i32 2
+// CHECK: [[VAR1_LHS_VOID:%.+]] = load i8*, i8** [[VAR1_LHS_REF]],
+// CHECK: [[VAR1_LHS:%.+]] = bitcast i8* [[VAR1_LHS_VOID]] to [[S_INT_TY]]*
+
+// t_var1_lhs = (i{{[0-9]+}}*)lhs[3];
+// CHECK: [[T_VAR1_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS]], i32 0, i32 3
+// CHECK: [[T_VAR1_RHS_VOID:%.+]] = load i8*, i8** [[T_VAR1_RHS_REF]],
+// CHECK: [[T_VAR1_RHS:%.+]] = bitcast i8* [[T_VAR1_RHS_VOID]] to i{{[0-9]+}}*
+// t_var1_rhs = (i{{[0-9]+}}*)rhs[3];
+// CHECK: [[T_VAR1_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i32 0, i32 3
+// CHECK: [[T_VAR1_LHS_VOID:%.+]] = load i8*, i8** [[T_VAR1_LHS_REF]],
+// CHECK: [[T_VAR1_LHS:%.+]] = bitcast i8* [[T_VAR1_LHS_VOID]] to i{{[0-9]+}}*
+
+// t_var_lhs += t_var_rhs;
+// CHECK: [[T_VAR_LHS_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR_LHS]],
+// CHECK: [[T_VAR_RHS_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR_RHS]],
+// CHECK: [[UP:%.+]] = add nsw i{{[0-9]+}} [[T_VAR_LHS_VAL]], [[T_VAR_RHS_VAL]]
+// CHECK: store i{{[0-9]+}} [[UP]], i{{[0-9]+}}* [[T_VAR_LHS]],
+
+// var_lhs = var_lhs.operator &(var_rhs);
+// CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_INT_TY]]* @{{.+}}([[S_INT_TY]]* [[VAR_LHS]], [[S_INT_TY]]* dereferenceable(4) [[VAR_RHS]])
+// CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR_LHS]] to i8*
+// CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[UP]] to i8*
+// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false)
+
+// var1_lhs = var1_lhs.operator &&(var1_rhs);
+// CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_LHS]])
+// CHECK: [[VAR1_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0
+// CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[FALSE:.+]]
+// CHECK: [[TRUE]]
+// CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_RHS]])
+// CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0
+// CHECK: br i1 [[VAR1_REDUCTION_BOOL]], label %[[TRUE2:.+]], label %[[FALSE2:.+]]
+// CHECK: [[TRUE2]]
+// CHECK: br label %[[END2:.+]]
+// CHECK: [[FALSE2]]
+// CHECK: br label %[[END2]]
+// CHECK: [[END2]]
+// CHECK: [[COND_LVALUE:%.+]] = phi [[S_INT_TY]]* [ [[VAR1_LHS]], %[[TRUE2]] ], [ [[VAR1_RHS]], %[[FALSE2]] ]
+// CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR1_LHS]] to i8*
+// CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[COND_LVALUE]] to i8*
+// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false)
+
+// t_var1_lhs = min(t_var1_lhs, t_var1_rhs);
+// CHECK: [[T_VAR1_LHS_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_LHS]],
+// CHECK: [[T_VAR1_RHS_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_RHS]],
+// CHECK: [[CMP:%.+]] = icmp slt i{{[0-9]+}} [[T_VAR1_LHS_VAL]], [[T_VAR1_RHS_VAL]]
+// CHECK: [[UP:%.+]] = zext i1 [[CMP]] to i{{[0-9]+}}
+// CHECK: store i{{[0-9]+}} [[UP]], i{{[0-9]+}}* [[T_VAR1_LHS]],
+// CHECK: ret void
+
+#endif
+

clang/test/OpenMP/parallel_reduction_messages.cpp

@@ -11,7 +11,7 @@ struct S1; // expected-note {{declared here}} expected-note 4 {{forward declarat
 extern S1 a;
 class S2 {
   mutable int a;
-  S2 &operator+=(const S2 &arg) { return (*this); }
+  S2 &operator+(const S2 &arg) { return (*this); } // expected-note 4 {{implicitly declared private here}}
 
 public:
   S2() : a(0) {}
@@ -28,17 +28,17 @@ class S3 {
 public:
   S3() : a(0) {}
   S3(const S3 &s3) : a(s3.a) {}
-  S3 operator+=(const S3 &arg1) { return arg1; }
+  S3 operator+(const S3 &arg1) { return arg1; }
 };
-int operator+=(const S3 &arg1, const S3 &arg2) { return 5; }
+int operator+(const S3 &arg1, const S3 &arg2) { return 5; }
 S3 c;               // expected-note 2 {{'c' defined here}}
 const S3 ca[5];     // expected-note 2 {{'ca' defined here}}
 extern const int f; // expected-note 4 {{'f' declared here}}
-class S4 {          // expected-note {{'S4' declared here}}
+class S4 {
   int a;
-  S4();
+  S4(); // expected-note {{implicitly declared private here}}
   S4(const S4 &s4);
-  S4 &operator+=(const S4 &arg) { return (*this); }
+  S4 &operator+(const S4 &arg) { return (*this); }
 
 public:
   S4(int v) : a(v) {}
@@ -46,26 +46,26 @@ public:
 S4 &operator&=(S4 &arg1, S4 &arg2) { return arg1; }
 class S5 {
   int a;
-  S5() : a(0) {}
+  S5() : a(0) {} // expected-note {{implicitly declared private here}}
   S5(const S5 &s5) : a(s5.a) {}
-  S5 &operator+=(const S5 &arg);
+  S5 &operator+(const S5 &arg);
 
 public:
   S5(int v) : a(v) {}
 };
-class S6 {
+class S6 { // expected-note 2 {{candidate function (the implicit copy assignment operator) not viable: no known conversion from 'int' to 'const S6' for 1st argument}}
   int a;
 
 public:
   S6() : a(6) {}
   operator int() { return 6; }
-} o; // expected-note 2 {{'o' defined here}}
+} o;
 
 S3 h, k;
 #pragma omp threadprivate(h) // expected-note 2 {{defined as threadprivate or thread local}}
 
 template <class T>       // expected-note {{declared here}}
-T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
+T tmain(T argc) {
   const T d = T();       // expected-note 4 {{'d' defined here}}
   const T da[5] = {T()}; // expected-note 2 {{'da' defined here}}
   T qa[5] = {T()};
@@ -74,7 +74,7 @@ T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
   S3 &p = k;                   // expected-note 2 {{'p' defined here}}
   const T &r = da[(int)i];     // expected-note 2 {{'r' defined here}}
   T &q = qa[(int)i];           // expected-note 2 {{'q' defined here}}
-  T fl;                        // expected-note {{'fl' defined here}}
+  T fl;
 #pragma omp parallel reduction // expected-error {{expected '(' after 'reduction'}}
   foo();
 #pragma omp parallel reduction + // expected-error {{expected '(' after 'reduction'}} expected-warning {{extra tokens at the end of '#pragma omp parallel' are ignored}}
@@ -89,9 +89,9 @@ T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
   foo();
 #pragma omp parallel reduction(\) // expected-error {{expected unqualified-id}} expected-warning {{missing ':' after reduction identifier - ignoring}}
   foo();
-#pragma omp parallel reduction(& : argc // expected-error {{expected ')'}} expected-note {{to match this '('}} expected-error {{variable of type 'float' is not valid for specified reduction operation}}
+#pragma omp parallel reduction(& : argc // expected-error {{expected ')'}} expected-note {{to match this '('}} expected-error {{invalid operands to binary expression ('float' and 'float')}}
   foo();
-#pragma omp parallel reduction(| : argc, // expected-error {{expected expression}} expected-error {{expected ')'}} expected-note {{to match this '('}} expected-error {{variable of type 'float' is not valid for specified reduction operation}}
+#pragma omp parallel reduction(| : argc, // expected-error {{expected expression}} expected-error {{expected ')'}} expected-note {{to match this '('}} expected-error {{invalid operands to binary expression ('float' and 'float')}}
   foo();
 #pragma omp parallel reduction(|| : argc ? i : argc) // expected-error 2 {{expected variable name}}
   foo();
@@ -101,7 +101,7 @@ T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
   foo();
 #pragma omp parallel reduction(^ : T) // expected-error {{'T' does not refer to a value}}
   foo();
-#pragma omp parallel reduction(+ : a, b, c, d, f) // expected-error {{reduction variable with incomplete type 'S1'}} expected-error 3 {{const-qualified variable cannot be reduction}}
+#pragma omp parallel reduction(+ : a, b, c, d, f) // expected-error {{reduction variable with incomplete type 'S1'}} expected-error 3 {{const-qualified variable cannot be reduction}} expected-error 3 {{'operator+' is a private member of 'S2'}}
   foo();
 #pragma omp parallel reduction(min : a, b, c, d, f) // expected-error {{reduction variable with incomplete type 'S1'}} expected-error 2 {{arguments of OpenMP clause 'reduction' for 'min' or 'max' must be of arithmetic type}} expected-error 3 {{const-qualified variable cannot be reduction}}
   foo();
@@ -113,7 +113,7 @@ T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
   foo();
 #pragma omp parallel reduction(- : da) // expected-error {{a reduction variable with array type 'const int [5]'}} expected-error {{a reduction variable with array type 'const float [5]'}}
   foo();
-#pragma omp parallel reduction(^ : fl) // expected-error {{variable of type 'float' is not valid for specified reduction operation}}
+#pragma omp parallel reduction(^ : fl) // expected-error {{invalid operands to binary expression ('float' and 'float')}}
   foo();
 #pragma omp parallel reduction(&& : S2::S2s)
   foo();
@@ -121,7 +121,7 @@ T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
   foo();
 #pragma omp parallel reduction(+ : h, k) // expected-error {{threadprivate or thread local variable cannot be reduction}}
   foo();
-#pragma omp parallel reduction(+ : o) // expected-error {{variable of type 'class S6' is not valid for specified reduction operation}}
+#pragma omp parallel reduction(+ : o) // expected-error {{no viable overloaded '='}}
   foo();
 #pragma omp parallel private(i), reduction(+ : j), reduction(+ : q) // expected-error 4 {{argument of OpenMP clause 'reduction' must reference the same object in all threads}}
   foo();
@@ -154,14 +154,14 @@ int main(int argc, char **argv) {
   const int d = 5;       // expected-note 2 {{'d' defined here}}
   const int da[5] = {0}; // expected-note {{'da' defined here}}
   int qa[5] = {0};
-  S4 e(4); // expected-note {{'e' defined here}}
-  S5 g(5); // expected-note {{'g' defined here}}
+  S4 e(4);
+  S5 g(5);
   int i;
   int &j = i;                  // expected-note 2 {{'j' defined here}}
   S3 &p = k;                   // expected-note 2 {{'p' defined here}}
   const int &r = da[i];        // expected-note {{'r' defined here}}
   int &q = qa[i];              // expected-note {{'q' defined here}}
-  float fl;                    // expected-note {{'fl' defined here}}
+  float fl;
 #pragma omp parallel reduction // expected-error {{expected '(' after 'reduction'}}
   foo();
 #pragma omp parallel reduction + // expected-error {{expected '(' after 'reduction'}} expected-warning {{extra tokens at the end of '#pragma omp parallel' are ignored}}
@@ -188,7 +188,7 @@ int main(int argc, char **argv) {
   foo();
 #pragma omp parallel reduction(^ : S1) // expected-error {{'S1' does not refer to a value}}
   foo();
-#pragma omp parallel reduction(+ : a, b, c, d, f) // expected-error {{reduction variable with incomplete type 'S1'}} expected-error 2 {{const-qualified variable cannot be reduction}}
+#pragma omp parallel reduction(+ : a, b, c, d, f) // expected-error {{reduction variable with incomplete type 'S1'}} expected-error 2 {{const-qualified variable cannot be reduction}} expected-error {{'operator+' is a private member of 'S2'}}
   foo();
 #pragma omp parallel reduction(min : a, b, c, d, f) // expected-error {{reduction variable with incomplete type 'S1'}} expected-error 2 {{arguments of OpenMP clause 'reduction' for 'min' or 'max' must be of arithmetic type}} expected-error 2 {{const-qualified variable cannot be reduction}}
   foo();
@@ -200,17 +200,17 @@ int main(int argc, char **argv) {
   foo();
 #pragma omp parallel reduction(- : da) // expected-error {{a reduction variable with array type 'const int [5]'}}
   foo();
-#pragma omp parallel reduction(^ : fl) // expected-error {{variable of type 'float' is not valid for specified reduction operation}}
+#pragma omp parallel reduction(^ : fl) // expected-error {{invalid operands to binary expression ('float' and 'float')}}
   foo();
 #pragma omp parallel reduction(&& : S2::S2s)
   foo();
 #pragma omp parallel reduction(&& : S2::S2sc) // expected-error {{const-qualified variable cannot be reduction}}
   foo();
-#pragma omp parallel reduction(& : e, g) // expected-error {{reduction variable must have an accessible, unambiguous default constructor}} expected-error {{variable of type 'S5' is not valid for specified reduction operation}}
+#pragma omp parallel reduction(& : e, g) // expected-error {{calling a private constructor of class 'S4'}} expected-error {{nvalid operands to binary expression ('S4' and 'S4')}} expected-error {{calling a private constructor of class 'S5'}} expected-error {{invalid operands to binary expression ('S5' and 'S5')}}
   foo();
 #pragma omp parallel reduction(+ : h, k) // expected-error {{threadprivate or thread local variable cannot be reduction}}
   foo();
-#pragma omp parallel reduction(+ : o) // expected-error {{variable of type 'class S6' is not valid for specified reduction operation}}
+#pragma omp parallel reduction(+ : o) // expected-error {{no viable overloaded '='}}
   foo();
 #pragma omp parallel private(i), reduction(+ : j), reduction(+ : q) // expected-error 2 {{argument of OpenMP clause 'reduction' must reference the same object in all threads}}
   foo();

clang/test/OpenMP/parallel_sections_reduction_messages.cpp

@@ -11,7 +11,7 @@ struct S1; // expected-note {{declared here}} expected-note 4 {{forward declarat
 extern S1 a;
 class S2 {
   mutable int a;
-  S2 &operator+=(const S2 &arg) { return (*this); }
+  S2 &operator+(const S2 &arg) { return (*this); } // expected-note 4 {{implicitly declared private here}}
 
 public:
   S2() : a(0) {}
@@ -28,17 +28,17 @@ class S3 {
 public:
   S3() : a(0) {}
   S3(const S3 &s3) : a(s3.a) {}
-  S3 operator+=(const S3 &arg1) { return arg1; }
+  S3 operator+(const S3 &arg1) { return arg1; }
 };
-int operator+=(const S3 &arg1, const S3 &arg2) { return 5; }
+int operator+(const S3 &arg1, const S3 &arg2) { return 5; }
 S3 c;               // expected-note 2 {{'c' defined here}}
 const S3 ca[5];     // expected-note 2 {{'ca' defined here}}
 extern const int f; // expected-note 4 {{'f' declared here}}
-class S4 {          // expected-note {{'S4' declared here}}
+class S4 {
   int a;
-  S4();
+  S4(); // expected-note {{implicitly declared private here}}
   S4(const S4 &s4);
-  S4 &operator+=(const S4 &arg) { return (*this); }
+  S4 &operator+(const S4 &arg) { return (*this); }
 
 public:
   S4(int v) : a(v) {}
@@ -46,26 +46,26 @@ public:
 S4 &operator&=(S4 &arg1, S4 &arg2) { return arg1; }
 class S5 {
   int a;
-  S5() : a(0) {}
+  S5() : a(0) {} // expected-note {{implicitly declared private here}}
   S5(const S5 &s5) : a(s5.a) {}
-  S5 &operator+=(const S5 &arg);
+  S5 &operator+(const S5 &arg);
 
 public:
   S5(int v) : a(v) {}
 };
-class S6 {
+class S6 { // expected-note 2 {{candidate function (the implicit copy assignment operator) not viable: no known conversion from 'int' to 'const S6' for 1st argument}}
   int a;
 
 public:
   S6() : a(6) {}
   operator int() { return 6; }
-} o; // expected-note 2 {{'o' defined here}}
+} o;
 
 S3 h, k;
 #pragma omp threadprivate(h) // expected-note 2 {{defined as threadprivate or thread local}}
 
 template <class T>       // expected-note {{declared here}}
-T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
+T tmain(T argc) {
   const T d = T();       // expected-note 4 {{'d' defined here}}
   const T da[5] = {T()}; // expected-note 2 {{'da' defined here}}
   T qa[5] = {T()};
@@ -74,7 +74,7 @@ T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
   S3 &p = k;                            // expected-note 2 {{'p' defined here}}
   const T &r = da[(int)i];              // expected-note 2 {{'r' defined here}}
   T &q = qa[(int)i];                    // expected-note 2 {{'q' defined here}}
-  T fl;                                 // expected-note {{'fl' defined here}}
+  T fl;
 #pragma omp parallel sections reduction // expected-error {{expected '(' after 'reduction'}}
   {
     foo();
@@ -103,11 +103,11 @@ T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
   {
     foo();
   }
-#pragma omp parallel sections reduction(& : argc // expected-error {{expected ')'}} expected-note {{to match this '('}} expected-error {{variable of type 'float' is not valid for specified reduction operation}}
+#pragma omp parallel sections reduction(& : argc // expected-error {{expected ')'}} expected-note {{to match this '('}} expected-error {{invalid operands to binary expression ('float' and 'float')}}
   {
     foo();
   }
-#pragma omp parallel sections reduction(| : argc, // expected-error {{expected expression}} expected-error {{expected ')'}} expected-note {{to match this '('}} expected-error {{variable of type 'float' is not valid for specified reduction operation}}
+#pragma omp parallel sections reduction(| : argc, // expected-error {{expected expression}} expected-error {{expected ')'}} expected-note {{to match this '('}} expected-error {{invalid operands to binary expression ('float' and 'float')}}
   {
     foo();
   }
@@ -127,7 +127,7 @@ T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
   {
     foo();
   }
-#pragma omp parallel sections reduction(+ : a, b, c, d, f) // expected-error {{reduction variable with incomplete type 'S1'}} expected-error 3 {{const-qualified variable cannot be reduction}}
+#pragma omp parallel sections reduction(+ : a, b, c, d, f) // expected-error {{reduction variable with incomplete type 'S1'}} expected-error 3 {{const-qualified variable cannot be reduction}} expected-error 3 {{'operator+' is a private member of 'S2'}}
   {
     foo();
   }
@@ -151,7 +151,7 @@ T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
   {
     foo();
   }
-#pragma omp parallel sections reduction(^ : fl) // expected-error {{variable of type 'float' is not valid for specified reduction operation}}
+#pragma omp parallel sections reduction(^ : fl) // expected-error {{invalid operands to binary expression ('float' and 'float')}}
   {
     foo();
   }
@@ -167,7 +167,7 @@ T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
   {
     foo();
   }
-#pragma omp parallel sections reduction(+ : o) // expected-error {{variable of type 'class S6' is not valid for specified reduction operation}}
+#pragma omp parallel sections reduction(+ : o) // expected-error {{no viable overloaded '='}}
   {
     foo();
   }
@@ -212,14 +212,14 @@ int main(int argc, char **argv) {
   const int d = 5;       // expected-note 2 {{'d' defined here}}
   const int da[5] = {0}; // expected-note {{'da' defined here}}
   int qa[5] = {0};
-  S4 e(4); // expected-note {{'e' defined here}}
-  S5 g(5); // expected-note {{'g' defined here}}
+  S4 e(4);
+  S5 g(5);
   int i;
   int &j = i;                           // expected-note 2 {{'j' defined here}}
   S3 &p = k;                            // expected-note 2 {{'p' defined here}}
   const int &r = da[i];                 // expected-note {{'r' defined here}}
   int &q = qa[i];                       // expected-note {{'q' defined here}}
-  float fl;                             // expected-note {{'fl' defined here}}
+  float fl;
 #pragma omp parallel sections reduction // expected-error {{expected '(' after 'reduction'}}
   {
     foo();
@@ -272,7 +272,7 @@ int main(int argc, char **argv) {
   {
     foo();
   }
-#pragma omp parallel sections reduction(+ : a, b, c, d, f) // expected-error {{reduction variable with incomplete type 'S1'}} expected-error 2 {{const-qualified variable cannot be reduction}}
+#pragma omp parallel sections reduction(+ : a, b, c, d, f) // expected-error {{reduction variable with incomplete type 'S1'}} expected-error 2 {{const-qualified variable cannot be reduction}} expected-error {{'operator+' is a private member of 'S2'}}
   {
     foo();
   }
@@ -296,7 +296,7 @@ int main(int argc, char **argv) {
   {
     foo();
   }
-#pragma omp parallel sections reduction(^ : fl) // expected-error {{variable of type 'float' is not valid for specified reduction operation}}
+#pragma omp parallel sections reduction(^ : fl) // expected-error {{invalid operands to binary expression ('float' and 'float')}}
   {
     foo();
   }
@@ -308,7 +308,7 @@ int main(int argc, char **argv) {
   {
     foo();
   }
-#pragma omp parallel sections reduction(& : e, g) // expected-error {{reduction variable must have an accessible, unambiguous default constructor}} expected-error {{variable of type 'S5' is not valid for specified reduction operation}}
+#pragma omp parallel sections reduction(& : e, g) // expected-error {{calling a private constructor of class 'S4'}} expected-error {{invalid operands to binary expression ('S4' and 'S4')}} expected-error {{calling a private constructor of class 'S5'}} expected-error {{invalid operands to binary expression ('S5' and 'S5')}}
   {
     foo();
   }
@@ -316,7 +316,7 @@ int main(int argc, char **argv) {
   {
     foo();
   }
-#pragma omp parallel sections reduction(+ : o) // expected-error {{variable of type 'class S6' is not valid for specified reduction operation}}
+#pragma omp parallel sections reduction(+ : o) // expected-error {{no viable overloaded '='}}
   {
     foo();
   }

clang/test/OpenMP/sections_reduction_messages.cpp

@@ -1,4 +1,4 @@
-// RUN: %clang_cc1 -verify -fopenmp=libiomp5 -ferror-limit 100 -o - %s
+// RUN: %clang_cc1 -verify -fopenmp=libiomp5 -ferror-limit 150 -o - %s
 
 void foo() {
 }
@@ -11,7 +11,7 @@ struct S1; // expected-note {{declared here}} expected-note 4 {{forward declarat
 extern S1 a;
 class S2 {
   mutable int a;
-  S2 &operator+=(const S2 &arg) { return (*this); }
+  S2 &operator+(const S2 &arg) { return (*this); } // expected-note 4 {{implicitly declared private here}}
 
 public:
   S2() : a(0) {}
@@ -28,17 +28,17 @@ class S3 {
 public:
   S3() : a(0) {}
   S3(const S3 &s3) : a(s3.a) {}
-  S3 operator+=(const S3 &arg1) { return arg1; }
+  S3 operator+(const S3 &arg1) { return arg1; }
 };
-int operator+=(const S3 &arg1, const S3 &arg2) { return 5; }
+int operator+(const S3 &arg1, const S3 &arg2) { return 5; }
 S3 c;               // expected-note 2 {{'c' defined here}}
 const S3 ca[5];     // expected-note 2 {{'ca' defined here}}
 extern const int f; // expected-note 4 {{'f' declared here}}
-class S4 {          // expected-note {{'S4' declared here}}
+class S4 {
   int a;
-  S4();
+  S4(); // expected-note {{implicitly declared private here}}
   S4(const S4 &s4);
-  S4 &operator+=(const S4 &arg) { return (*this); }
+  S4 &operator+(const S4 &arg) { return (*this); }
 
 public:
   S4(int v) : a(v) {}
@@ -46,26 +46,26 @@ public:
 S4 &operator&=(S4 &arg1, S4 &arg2) { return arg1; }
 class S5 {
   int a;
-  S5() : a(0) {}
+  S5() : a(0) {} // expected-note {{implicitly declared private here}}
   S5(const S5 &s5) : a(s5.a) {}
-  S5 &operator+=(const S5 &arg);
+  S5 &operator+(const S5 &arg);
 
 public:
   S5(int v) : a(v) {}
 };
-class S6 {
+class S6 { // expected-note 2 {{candidate function (the implicit copy assignment operator) not viable: no known conversion from 'int' to 'const S6' for 1st argument}}
   int a;
 
 public:
   S6() : a(6) {}
   operator int() { return 6; }
-} o; // expected-note 2 {{'o' defined here}}
+} o;
 
 S3 h, k;
 #pragma omp threadprivate(h) // expected-note 2 {{defined as threadprivate or thread local}}
 
 template <class T>       // expected-note {{declared here}}
-T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
+T tmain(T argc) {
   const T d = T();       // expected-note 4 {{'d' defined here}}
   const T da[5] = {T()}; // expected-note 2 {{'da' defined here}}
   T qa[5] = {T()};
@@ -74,7 +74,7 @@ T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
   S3 &p = k;               // expected-note 2 {{'p' defined here}}
   const T &r = da[(int)i]; // expected-note 2 {{'r' defined here}}
   T &q = qa[(int)i];       // expected-note 2 {{'q' defined here}}
-  T fl;                    // expected-note {{'fl' defined here}}
+  T fl;
 #pragma omp parallel
 #pragma omp sections reduction // expected-error {{expected '(' after 'reduction'}}
   {
@@ -111,12 +111,12 @@ T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
     foo();
   }
 #pragma omp parallel
-#pragma omp sections reduction(& : argc // expected-error {{expected ')'}} expected-note {{to match this '('}} expected-error {{variable of type 'float' is not valid for specified reduction operation}}
+#pragma omp sections reduction(& : argc // expected-error {{expected ')'}} expected-note {{to match this '('}} expected-error {{invalid operands to binary expression ('float' and 'float')}}
   {
     foo();
   }
 #pragma omp parallel
-#pragma omp sections reduction(| : argc, // expected-error {{expected expression}} expected-error {{expected ')'}} expected-note {{to match this '('}} expected-error {{variable of type 'float' is not valid for specified reduction operation}}
+#pragma omp sections reduction(| : argc, // expected-error {{expected expression}} expected-error {{expected ')'}} expected-note {{to match this '('}} expected-error {{invalid operands to binary expression ('float' and 'float')}}
   {
     foo();
   }
@@ -141,7 +141,7 @@ T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
     foo();
   }
 #pragma omp parallel
-#pragma omp sections reduction(+ : a, b, c, d, f) // expected-error {{reduction variable with incomplete type 'S1'}} expected-error 3 {{const-qualified variable cannot be reduction}}
+#pragma omp sections reduction(+ : a, b, c, d, f) // expected-error {{reduction variable with incomplete type 'S1'}} expected-error 3 {{const-qualified variable cannot be reduction}} expected-error 3 {{'operator+' is a private member of 'S2'}}
   {
     foo();
   }
@@ -171,7 +171,7 @@ T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
     foo();
   }
 #pragma omp parallel
-#pragma omp sections reduction(^ : fl) // expected-error {{variable of type 'float' is not valid for specified reduction operation}}
+#pragma omp sections reduction(^ : fl) // expected-error {{invalid operands to binary expression ('float' and 'float')}}
   {
     foo();
   }
@@ -191,7 +191,7 @@ T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
     foo();
   }
 #pragma omp parallel
-#pragma omp sections reduction(+ : o) // expected-error {{variable of type 'class S6' is not valid for specified reduction operation}}
+#pragma omp sections reduction(+ : o) // expected-error {{no viable overloaded '='}}
   {
     foo();
   }
@@ -239,14 +239,14 @@ int main(int argc, char **argv) {
   const int d = 5;       // expected-note 2 {{'d' defined here}}
   const int da[5] = {0}; // expected-note {{'da' defined here}}
   int qa[5] = {0};
-  S4 e(4); // expected-note {{'e' defined here}}
-  S5 g(5); // expected-note {{'g' defined here}}
+  S4 e(4);
+  S5 g(5);
   int i;
   int &j = i;           // expected-note 2 {{'j' defined here}}
   S3 &p = k;            // expected-note 2 {{'p' defined here}}
   const int &r = da[i]; // expected-note {{'r' defined here}}
   int &q = qa[i];       // expected-note {{'q' defined here}}
-  float fl;             // expected-note {{'fl' defined here}}
+  float fl;
 #pragma omp parallel
 #pragma omp sections reduction // expected-error {{expected '(' after 'reduction'}}
   {
@@ -313,7 +313,7 @@ int main(int argc, char **argv) {
     foo();
   }
 #pragma omp parallel
-#pragma omp sections reduction(+ : a, b, c, d, f) // expected-error {{reduction variable with incomplete type 'S1'}} expected-error 2 {{const-qualified variable cannot be reduction}}
+#pragma omp sections reduction(+ : a, b, c, d, f) // expected-error {{reduction variable with incomplete type 'S1'}} expected-error 2 {{const-qualified variable cannot be reduction}} expected-error {{'operator+' is a private member of 'S2'}}
   {
     foo();
   }
@@ -343,7 +343,7 @@ int main(int argc, char **argv) {
     foo();
   }
 #pragma omp parallel
-#pragma omp sections reduction(^ : fl) // expected-error {{variable of type 'float' is not valid for specified reduction operation}}
+#pragma omp sections reduction(^ : fl) // expected-error {{invalid operands to binary expression ('float' and 'float')}}
   {
     foo();
   }
@@ -358,7 +358,7 @@ int main(int argc, char **argv) {
     foo();
   }
 #pragma omp parallel
-#pragma omp sections reduction(& : e, g) // expected-error {{reduction variable must have an accessible, unambiguous default constructor}} expected-error {{variable of type 'S5' is not valid for specified reduction operation}}
+#pragma omp sections reduction(& : e, g) // expected-error {{calling a private constructor of class 'S4'}} expected-error {{invalid operands to binary expression ('S4' and 'S4')}} expected-error {{calling a private constructor of class 'S5'}} expected-error {{invalid operands to binary expression ('S5' and 'S5')}}
   {
     foo();
   }
@@ -368,7 +368,7 @@ int main(int argc, char **argv) {
     foo();
   }
 #pragma omp parallel
-#pragma omp sections reduction(+ : o) // expected-error {{variable of type 'class S6' is not valid for specified reduction operation}}
+#pragma omp sections reduction(+ : o) // expected-error {{no viable overloaded '='}}
   {
     foo();
   }

clang/test/OpenMP/simd_reduction_messages.cpp

@@ -11,7 +11,7 @@ struct S1; // expected-note {{declared here}} expected-note 4 {{forward declarat
 extern S1 a;
 class S2 {
   mutable int a;
-  S2 &operator+=(const S2 &arg) { return (*this); }
+  S2 &operator+(const S2 &arg) { return (*this); } // expected-note 4 {{implicitly declared private here}}
 
 public:
   S2() : a(0) {}
@@ -28,17 +28,17 @@ class S3 {
 public:
   S3() : a(0) {}
   S3(const S3 &s3) : a(s3.a) {}
-  S3 operator+=(const S3 &arg1) { return arg1; }
+  S3 operator+(const S3 &arg1) { return arg1; }
 };
-int operator+=(const S3 &arg1, const S3 &arg2) { return 5; }
+int operator+(const S3 &arg1, const S3 &arg2) { return 5; }
 S3 c;               // expected-note 2 {{'c' defined here}}
 const S3 ca[5];     // expected-note 2 {{'ca' defined here}}
 extern const int f; // expected-note 4 {{'f' declared here}}
-class S4 {          // expected-note {{'S4' declared here}}
+class S4 {
   int a;
-  S4();
+  S4(); // expected-note {{implicitly declared private here}}
   S4(const S4 &s4);
-  S4 &operator+=(const S4 &arg) { return (*this); }
+  S4 &operator+(const S4 &arg) { return (*this); }
 
 public:
   S4(int v) : a(v) {}
@@ -46,26 +46,26 @@ public:
 S4 &operator&=(S4 &arg1, S4 &arg2) { return arg1; }
 class S5 {
   int a;
-  S5() : a(0) {}
+  S5() : a(0) {} // expected-note {{implicitly declared private here}}
   S5(const S5 &s5) : a(s5.a) {}
-  S5 &operator+=(const S5 &arg);
+  S5 &operator+(const S5 &arg);
 
 public:
   S5(int v) : a(v) {}
 };
-class S6 {
+class S6 { // expected-note 2 {{candidate function (the implicit copy assignment operator) not viable: no known conversion from 'int' to 'const S6' for 1st argument}}
   int a;
 
 public:
   S6() : a(6) {}
   operator int() { return 6; }
-} o; // expected-note 2 {{'o' defined here}}
+} o;
 
 S3 h, k;
 #pragma omp threadprivate(h) // expected-note 2 {{defined as threadprivate or thread local}}
 
 template <class T>       // expected-note {{declared here}}
-T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
+T tmain(T argc) {
   const T d = T();       // expected-note 4 {{'d' defined here}}
   const T da[5] = {T()}; // expected-note 2 {{'da' defined here}}
   T qa[5] = {T()};
@@ -74,7 +74,7 @@ T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
   S3 &p = k;                       // expected-note 2 {{'p' defined here}}
   const T &r = da[(int)i];         // expected-note 2 {{'r' defined here}}
   T &q = qa[(int)i];               // expected-note 2 {{'q' defined here}}
-  T fl;                            // expected-note {{'fl' defined here}}
+  T fl;
 #pragma omp simd reduction // expected-error {{expected '(' after 'reduction'}}
   for (int i = 0; i < 10; ++i)
     foo();
@@ -96,10 +96,10 @@ T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
 #pragma omp simd reduction(\) // expected-error {{expected unqualified-id}} expected-warning {{missing ':' after reduction identifier - ignoring}}
   for (int i = 0; i < 10; ++i)
     foo();
-#pragma omp simd reduction(& : argc // expected-error {{expected ')'}} expected-note {{to match this '('}} expected-error {{variable of type 'float' is not valid for specified reduction operation}}
+#pragma omp simd reduction(& : argc // expected-error {{expected ')'}} expected-note {{to match this '('}} expected-error {{invalid operands to binary expression ('float' and 'float')}}
   for (int i = 0; i < 10; ++i)
     foo();
-#pragma omp simd reduction(| : argc, // expected-error {{expected expression}} expected-error {{expected ')'}} expected-note {{to match this '('}} expected-error {{variable of type 'float' is not valid for specified reduction operation}}
+#pragma omp simd reduction(| : argc, // expected-error {{expected expression}} expected-error {{expected ')'}} expected-note {{to match this '('}} expected-error {{invalid operands to binary expression ('float' and 'float')}}
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp simd reduction(|| : argc ? i : argc) // expected-error 2 {{expected variable name}}
@@ -114,7 +114,7 @@ T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
 #pragma omp simd reduction(^ : T) // expected-error {{'T' does not refer to a value}}
   for (int i = 0; i < 10; ++i)
     foo();
-#pragma omp simd reduction(+ : a, b, c, d, f) // expected-error {{reduction variable with incomplete type 'S1'}} expected-error 3 {{const-qualified variable cannot be reduction}}
+#pragma omp simd reduction(+ : a, b, c, d, f) // expected-error {{reduction variable with incomplete type 'S1'}} expected-error 3 {{const-qualified variable cannot be reduction}} expected-error 3 {{'operator+' is a private member of 'S2'}}
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp simd reduction(min : a, b, c, d, f) // expected-error {{reduction variable with incomplete type 'S1'}} expected-error 2 {{arguments of OpenMP clause 'reduction' for 'min' or 'max' must be of arithmetic type}} expected-error 3 {{const-qualified variable cannot be reduction}}
@@ -132,7 +132,7 @@ T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
 #pragma omp simd reduction(- : da) // expected-error {{a reduction variable with array type 'const int [5]'}} expected-error {{a reduction variable with array type 'const float [5]'}}
   for (int i = 0; i < 10; ++i)
     foo();
-#pragma omp simd reduction(^ : fl) // expected-error {{variable of type 'float' is not valid for specified reduction operation}}
+#pragma omp simd reduction(^ : fl) // expected-error {{invalid operands to binary expression ('float' and 'float')}}
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp simd reduction(&& : S2::S2s) // expected-error {{shared variable cannot be reduction}}
@@ -144,7 +144,7 @@ T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
 #pragma omp simd reduction(+ : h, k) // expected-error {{threadprivate or thread local variable cannot be reduction}}
   for (int i = 0; i < 10; ++i)
     foo();
-#pragma omp simd reduction(+ : o) // expected-error {{variable of type 'class S6' is not valid for specified reduction operation}}
+#pragma omp simd reduction(+ : o) // expected-error {{no viable overloaded '='}}
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp simd private(i), reduction(+ : j), reduction(+ : q) // expected-error 4 {{argument of OpenMP clause 'reduction' must reference the same object in all threads}}
@@ -184,14 +184,14 @@ int main(int argc, char **argv) {
   const int d = 5;       // expected-note 2 {{'d' defined here}}
   const int da[5] = {0}; // expected-note {{'da' defined here}}
   int qa[5] = {0};
-  S4 e(4); // expected-note {{'e' defined here}}
-  S5 g(5); // expected-note {{'g' defined here}}
+  S4 e(4);
+  S5 g(5);
   int i;
   int &j = i;                      // expected-note 2 {{'j' defined here}}
   S3 &p = k;                       // expected-note 2 {{'p' defined here}}
   const int &r = da[i];            // expected-note {{'r' defined here}}
   int &q = qa[i];                  // expected-note {{'q' defined here}}
-  float fl;                        // expected-note {{'fl' defined here}}
+  float fl;
 #pragma omp simd reduction // expected-error {{expected '(' after 'reduction'}}
   for (int i = 0; i < 10; ++i)
     foo();
@@ -231,7 +231,7 @@ int main(int argc, char **argv) {
 #pragma omp simd reduction(^ : S1) // expected-error {{'S1' does not refer to a value}}
   for (int i = 0; i < 10; ++i)
     foo();
-#pragma omp simd reduction(+ : a, b, c, d, f) // expected-error {{reduction variable with incomplete type 'S1'}} expected-error 2 {{const-qualified variable cannot be reduction}}
+#pragma omp simd reduction(+ : a, b, c, d, f) // expected-error {{reduction variable with incomplete type 'S1'}} expected-error 2 {{const-qualified variable cannot be reduction}} expected-error {{'operator+' is a private member of 'S2'}}
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp simd reduction(min : a, b, c, d, f) // expected-error {{reduction variable with incomplete type 'S1'}} expected-error 2 {{arguments of OpenMP clause 'reduction' for 'min' or 'max' must be of arithmetic type}} expected-error 2 {{const-qualified variable cannot be reduction}}
@@ -249,7 +249,7 @@ int main(int argc, char **argv) {
 #pragma omp simd reduction(- : da) // expected-error {{a reduction variable with array type 'const int [5]'}}
   for (int i = 0; i < 10; ++i)
     foo();
-#pragma omp simd reduction(^ : fl) // expected-error {{variable of type 'float' is not valid for specified reduction operation}}
+#pragma omp simd reduction(^ : fl) // expected-error {{invalid operands to binary expression ('float' and 'float')}}
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp simd reduction(&& : S2::S2s) // expected-error {{shared variable cannot be reduction}}
@@ -258,13 +258,13 @@ int main(int argc, char **argv) {
 #pragma omp simd reduction(&& : S2::S2sc) // expected-error {{const-qualified variable cannot be reduction}}
   for (int i = 0; i < 10; ++i)
     foo();
-#pragma omp simd reduction(& : e, g) // expected-error {{reduction variable must have an accessible, unambiguous default constructor}} expected-error {{variable of type 'S5' is not valid for specified reduction operation}}
+#pragma omp simd reduction(& : e, g) // expected-error {{calling a private constructor of class 'S4'}} expected-error {{invalid operands to binary expression ('S4' and 'S4')}} expected-error {{calling a private constructor of class 'S5'}} expected-error {{invalid operands to binary expression ('S5' and 'S5')}}
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp simd reduction(+ : h, k) // expected-error {{threadprivate or thread local variable cannot be reduction}}
   for (int i = 0; i < 10; ++i)
     foo();
-#pragma omp simd reduction(+ : o) // expected-error {{variable of type 'class S6' is not valid for specified reduction operation}}
+#pragma omp simd reduction(+ : o) // expected-error {{no viable overloaded '='}}
   for (int i = 0; i < 10; ++i)
     foo();
 #pragma omp simd private(i), reduction(+ : j), reduction(+ : q) // expected-error 2 {{argument of OpenMP clause 'reduction' must reference the same object in all threads}}

clang/test/OpenMP/teams_reduction_messages.cpp

@@ -11,7 +11,7 @@ struct S1; // expected-note {{declared here}} expected-note 4 {{forward declarat
 extern S1 a;
 class S2 {
   mutable int a;
-  S2 &operator+=(const S2 &arg) { return (*this); }
+  S2 &operator+(const S2 &arg) { return (*this); } // expected-note 4 {{implicitly declared private here}}
 
 public:
   S2() : a(0) {}
@@ -28,17 +28,17 @@ class S3 {
 public:
   S3() : a(0) {}
   S3(const S3 &s3) : a(s3.a) {}
-  S3 operator+=(const S3 &arg1) { return arg1; }
+  S3 operator+(const S3 &arg1) { return arg1; }
 };
-int operator+=(const S3 &arg1, const S3 &arg2) { return 5; }
+int operator+(const S3 &arg1, const S3 &arg2) { return 5; }
 S3 c;               // expected-note 2 {{'c' defined here}}
 const S3 ca[5];     // expected-note 2 {{'ca' defined here}}
 extern const int f; // expected-note 4 {{'f' declared here}}
-class S4 {          // expected-note {{'S4' declared here}}
+class S4 {
   int a;
-  S4();
+  S4(); // expected-note {{implicitly declared private here}}
   S4(const S4 &s4);
-  S4 &operator+=(const S4 &arg) { return (*this); }
+  S4 &operator+(const S4 &arg) { return (*this); }
 
 public:
   S4(int v) : a(v) {}
@@ -46,26 +46,26 @@ public:
 S4 &operator&=(S4 &arg1, S4 &arg2) { return arg1; }
 class S5 {
   int a;
-  S5() : a(0) {}
+  S5() : a(0) {} // expected-note {{implicitly declared private here}}
   S5(const S5 &s5) : a(s5.a) {}
-  S5 &operator+=(const S5 &arg);
+  S5 &operator+(const S5 &arg);
 
 public:
   S5(int v) : a(v) {}
 };
-class S6 {
+class S6 { // expected-note 2 {{candidate function (the implicit copy assignment operator) not viable: no known conversion from 'int' to 'const S6' for 1st argument}}
   int a;
 
 public:
   S6() : a(6) {}
   operator int() { return 6; }
-} o; // expected-note 2 {{'o' defined here}}
+} o;
 
 S3 h, k;
 #pragma omp threadprivate(h) // expected-note 2 {{defined as threadprivate or thread local}}
 
 template <class T>       // expected-note {{declared here}}
-T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
+T tmain(T argc) {
   const T d = T();       // expected-note 4 {{'d' defined here}}
   const T da[5] = {T()}; // expected-note 2 {{'da' defined here}}
   T qa[5] = {T()};
@@ -74,7 +74,7 @@ T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
   S3 &p = k;                   // expected-note 2 {{'p' defined here}}
   const T &r = da[(int)i];     // expected-note 2 {{'r' defined here}}
   T &q = qa[(int)i];           // expected-note 2 {{'q' defined here}}
-  T fl;                        // expected-note {{'fl' defined here}}
+  T fl;
 #pragma omp target
 #pragma omp teams reduction // expected-error {{expected '(' after 'reduction'}}
   foo();
@@ -97,10 +97,10 @@ T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
 #pragma omp teams reduction(\) // expected-error {{expected unqualified-id}} expected-warning {{missing ':' after reduction identifier - ignoring}}
   foo();
 #pragma omp target
-#pragma omp teams reduction(& : argc // expected-error {{expected ')'}} expected-note {{to match this '('}} expected-error {{variable of type 'float' is not valid for specified reduction operation}}
+#pragma omp teams reduction(& : argc // expected-error {{expected ')'}} expected-note {{to match this '('}} expected-error {{invalid operands to binary expression ('float' and 'float')}}
   foo();
 #pragma omp target
-#pragma omp teams reduction(| : argc, // expected-error {{expected expression}} expected-error {{expected ')'}} expected-note {{to match this '('}} expected-error {{variable of type 'float' is not valid for specified reduction operation}}
+#pragma omp teams reduction(| : argc, // expected-error {{expected expression}} expected-error {{expected ')'}} expected-note {{to match this '('}} expected-error {{invalid operands to binary expression ('float' and 'float')}}
   foo();
 #pragma omp target
 #pragma omp teams reduction(|| : argc ? i : argc) // expected-error 2 {{expected variable name}}
@@ -115,7 +115,7 @@ T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
 #pragma omp teams reduction(^ : T) // expected-error {{'T' does not refer to a value}}
   foo();
 #pragma omp target
-#pragma omp teams reduction(+ : a, b, c, d, f) // expected-error {{reduction variable with incomplete type 'S1'}} expected-error 3 {{const-qualified variable cannot be reduction}}
+#pragma omp teams reduction(+ : a, b, c, d, f) // expected-error {{reduction variable with incomplete type 'S1'}} expected-error 3 {{const-qualified variable cannot be reduction}} expected-error 3 {{'operator+' is a private member of 'S2'}}
   foo();
 #pragma omp target
 #pragma omp teams reduction(min : a, b, c, d, f) // expected-error {{reduction variable with incomplete type 'S1'}} expected-error 2 {{arguments of OpenMP clause 'reduction' for 'min' or 'max' must be of arithmetic type}} expected-error 3 {{const-qualified variable cannot be reduction}}
@@ -133,7 +133,7 @@ T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
 #pragma omp teams reduction(- : da) // expected-error {{a reduction variable with array type 'const int [5]'}} expected-error {{a reduction variable with array type 'const float [5]'}}
   foo();
 #pragma omp target
-#pragma omp teams reduction(^ : fl) // expected-error {{variable of type 'float' is not valid for specified reduction operation}}
+#pragma omp teams reduction(^ : fl) // expected-error {{invalid operands to binary expression ('float' and 'float')}}
   foo();
 #pragma omp target
 #pragma omp teams reduction(&& : S2::S2s)
@@ -145,7 +145,7 @@ T tmain(T argc) {        // expected-note 2 {{'argc' defined here}}
 #pragma omp teams reduction(+ : h, k) // expected-error {{threadprivate or thread local variable cannot be reduction}}
   foo();
 #pragma omp target
-#pragma omp teams reduction(+ : o) // expected-error {{variable of type 'class S6' is not valid for specified reduction operation}}
+#pragma omp teams reduction(+ : o) // expected-error {{no viable overloaded '='}}
   foo();
 #pragma omp target
 #pragma omp teams private(i), reduction(+ : j), reduction(+ : q) // expected-error 4 {{argument of OpenMP clause 'reduction' must reference the same object in all threads}}
@@ -187,14 +187,14 @@ int main(int argc, char **argv) {
   const int d = 5;       // expected-note 2 {{'d' defined here}}
   const int da[5] = {0}; // expected-note {{'da' defined here}}
   int qa[5] = {0};
-  S4 e(4); // expected-note {{'e' defined here}}
-  S5 g(5); // expected-note {{'g' defined here}}
+  S4 e(4);
+  S5 g(5);
   int i;
   int &j = i;                  // expected-note 2 {{'j' defined here}}
   S3 &p = k;                   // expected-note 2 {{'p' defined here}}
   const int &r = da[i];        // expected-note {{'r' defined here}}
   int &q = qa[i];              // expected-note {{'q' defined here}}
-  float fl;                    // expected-note {{'fl' defined here}}
+  float fl;
 #pragma omp target
 #pragma omp teams reduction // expected-error {{expected '(' after 'reduction'}}
   foo();
@@ -235,7 +235,7 @@ int main(int argc, char **argv) {
 #pragma omp teams reduction(^ : S1) // expected-error {{'S1' does not refer to a value}}
   foo();
 #pragma omp target
-#pragma omp teams reduction(+ : a, b, c, d, f) // expected-error {{reduction variable with incomplete type 'S1'}} expected-error 2 {{const-qualified variable cannot be reduction}}
+#pragma omp teams reduction(+ : a, b, c, d, f) // expected-error {{reduction variable with incomplete type 'S1'}} expected-error 2 {{const-qualified variable cannot be reduction}} expected-error {{'operator+' is a private member of 'S2'}}
   foo();
 #pragma omp target
 #pragma omp teams reduction(min : a, b, c, d, f) // expected-error {{reduction variable with incomplete type 'S1'}} expected-error 2 {{arguments of OpenMP clause 'reduction' for 'min' or 'max' must be of arithmetic type}} expected-error 2 {{const-qualified variable cannot be reduction}}
@@ -253,7 +253,7 @@ int main(int argc, char **argv) {
 #pragma omp teams reduction(- : da) // expected-error {{a reduction variable with array type 'const int [5]'}}
   foo();
 #pragma omp target
-#pragma omp teams reduction(^ : fl) // expected-error {{variable of type 'float' is not valid for specified reduction operation}}
+#pragma omp teams reduction(^ : fl) // expected-error {{invalid operands to binary expression ('float' and 'float')}}
   foo();
 #pragma omp target
 #pragma omp teams reduction(&& : S2::S2s)
@@ -262,13 +262,13 @@ int main(int argc, char **argv) {
 #pragma omp teams reduction(&& : S2::S2sc) // expected-error {{const-qualified variable cannot be reduction}}
   foo();
 #pragma omp target
-#pragma omp teams reduction(& : e, g) // expected-error {{reduction variable must have an accessible, unambiguous default constructor}} expected-error {{variable of type 'S5' is not valid for specified reduction operation}}
+#pragma omp teams reduction(& : e, g) // expected-error {{calling a private constructor of class 'S4'}} expected-error {{invalid operands to binary expression ('S4' and 'S4')}} expected-error {{calling a private constructor of class 'S5'}} expected-error {{invalid operands to binary expression ('S5' and 'S5')}}
   foo();
 #pragma omp target
 #pragma omp teams reduction(+ : h, k) // expected-error {{threadprivate or thread local variable cannot be reduction}}
   foo();
 #pragma omp target
-#pragma omp teams reduction(+ : o) // expected-error {{variable of type 'class S6' is not valid for specified reduction operation}}
+#pragma omp teams reduction(+ : o) // expected-error {{no viable overloaded '='}}
   foo();
 #pragma omp target
 #pragma omp teams private(i), reduction(+ : j), reduction(+ : q) // expected-error 2 {{argument of OpenMP clause 'reduction' must reference the same object in all threads}}

clang/tools/libclang/CIndex.cpp

@@ -2028,6 +2028,15 @@ void OMPClauseEnqueue::VisitOMPSharedClause(const OMPSharedClause *C) {
 }
 void OMPClauseEnqueue::VisitOMPReductionClause(const OMPReductionClause *C) {
   VisitOMPClauseList(C);
+  for (auto *E : C->lhs_exprs()) {
+    Visitor->AddStmt(E);
+  }
+  for (auto *E : C->rhs_exprs()) {
+    Visitor->AddStmt(E);
+  }
+  for (auto *E : C->reduction_ops()) {
+    Visitor->AddStmt(E);
+  }
 }
 void OMPClauseEnqueue::VisitOMPLinearClause(const OMPLinearClause *C) {
   VisitOMPClauseList(C);