Adding support for target in_reduction

[RitanyaB]

Summary

Implementing target in_reduction by wrapping target task with host task with in_reduction and if clause. This is in compliance with OpenMP 5.0 section: 2.19.5.6.
So, this

#pragma omp target in_reduction(+:res)
  for (int i=0; i<N; i++) {
    res = res+i
  }

will become

#pragma omp task in_reduction(+:res) if(0)
#pragma omp target map(res)
for (int i=0; i<N; i++) {
  res = res+i
}

[RitanyaB]

Could someone please review the changes?

Thanks.

[ABataev]

clang/lib/CodeGen/CGStmtOpenMP.cpp

5034–5038

Why integer, not a boolean? Also, this is currently meaningless, since these 2 vars are not visible outside of the if context.

[RitanyaB]

The if condition in the method EmitOMPTargetTaskBasedDirective in CGStmtOpenMP.cpp file was removed since it was unnecessary.

[ABataev]

What about the description? Shall we still emit if(0)?

[RitanyaB]

In D125669#3599134, @ABataev wrote:

What about the description? Shall we still emit if(0)?

Can you please elaborate for more clarity?

clang/lib/CodeGen/CGStmtOpenMP.cpp

5034–5038

Thank you for the suggestion. I have removed the if block.

[ABataev]

In D125669#3604302, @RitanyaB wrote:

In D125669#3599134, @ABataev wrote:

What about the description? Shall we still emit if(0)?

Can you please elaborate for more clarity?

The description of the patch
So, this

#pragma omp target in_reduction(+:res)

for (int i=0; i<N; i++) {
  res = res+i
}

will become

#pragma omp task in_reduction(+:res) if(0)
#pragma omp target map(res)
for (int i=0; i<N; i++) {

res = res+i

}

So, the question is is this still true?

[ssquare08]

In D125669#3604452, @ABataev wrote:

In D125669#3604302, @RitanyaB wrote:

In D125669#3599134, @ABataev wrote:

What about the description? Shall we still emit if(0)?

Can you please elaborate for more clarity?

The description of the patch
So, this

#pragma omp target in_reduction(+:res)

for (int i=0; i<N; i++) {
  res = res+i
}

will become

#pragma omp task in_reduction(+:res) if(0)
#pragma omp target map(res)
for (int i=0; i<N; i++) {

res = res+i

}

So, the question is is this still true?

The idea here is to wrap the target task with a host task and move in_reduction to that new host task with if(0) condition, making it an undeferred task. The approach here is basically using the framework that is in place for the host in_reduction.

[ABataev]

LG

clang/lib/CodeGen/CGOpenMPRuntime.cpp

if (!CGF.HaveInsertPoint())
     return;
 
   const bool OffloadingMandatory = !CGM.getLangOpts().OpenMPIsDevice &&
                                    CGM.getLangOpts().OpenMPOffloadMandatory;
 
   assert((OffloadingMandatory || OutlinedFn) && "Invalid outlined function!");
 
   const bool RequiresOuterTask = D.hasClausesOfKind<OMPDependClause>() ||
-                                 D.hasClausesOfKind<OMPNowaitClause>();
+                                 D.hasClausesOfKind<OMPNowaitClause>() ||
+                                 D.hasClausesOfKind<OMPInReductionClause>();
   llvm::SmallVector<llvm::Value *, 16> CapturedVars;
   const CapturedStmt &CS = *D.getCapturedStmt(OMPD_target);
   auto &&ArgsCodegen = [&CS, &CapturedVars](CodeGenFunction &CGF,
                                             PrePostActionTy &) {
     CGF.GenerateOpenMPCapturedVars(CS, CapturedVars);
   };
   emitInlinedDirective(CGF, OMPD_unknown, ArgsCodegen);
 

clang/lib/CodeGen/CGStmtOpenMP.cpp

for (const auto *C : S.getClausesOfKind<OMPFirstprivateClause>()) {
     for (auto *IInit : C->private_copies()) {
       Data.FirstprivateVars.push_back(*IRef);
       Data.FirstprivateCopies.push_back(IInit);
       Data.FirstprivateInits.push_back(*IElemInitRef);
       ++IRef;
       ++IElemInitRef;
     }
   }
+  SmallVector<const Expr *, 4> LHSs;
+  SmallVector<const Expr *, 4> RHSs;
+  for (const auto *C : S.getClausesOfKind<OMPInReductionClause>()) {
+    Data.ReductionVars.append(C->varlist_begin(), C->varlist_end());
+    Data.ReductionOrigs.append(C->varlist_begin(), C->varlist_end());
+    Data.ReductionCopies.append(C->privates().begin(), C->privates().end());
+    Data.ReductionOps.append(C->reduction_ops().begin(),
+                             C->reduction_ops().end());
+    LHSs.append(C->lhs_exprs().begin(), C->lhs_exprs().end());
+    RHSs.append(C->rhs_exprs().begin(), C->rhs_exprs().end());
+  }
   OMPPrivateScope TargetScope(*this);
   VarDecl *BPVD = nullptr;
   VarDecl *PVD = nullptr;
   VarDecl *SVD = nullptr;
   VarDecl *MVD = nullptr;
   if (InputInfo.NumberOfTargetItems > 0) {
     auto *CD = CapturedDecl::Create(
         getContext(), getContext().getTranslationUnitDecl(), /*NumParams=*/0);
if (!Data.FirstprivateVars.empty()) {
       for (const auto &Pair : PrivatePtrs) {
         Address Replacement(
             CGF.Builder.CreateLoad(Pair.second),
             CGF.ConvertTypeForMem(Pair.first->getType().getNonReferenceType()),
             CGF.getContext().getDeclAlign(Pair.first));
         Scope.addPrivate(Pair.first, Replacement);
       }
     }
-    // Privatize all private variables except for in_reduction items.
-    (void)Scope.Privatize();
+    CGF.processInReduction(S, Data, CGF, CS, Scope);
     if (InputInfo.NumberOfTargetItems > 0) {
       InputInfo.BasePointersArray = CGF.Builder.CreateConstArrayGEP(
           CGF.GetAddrOfLocalVar(BPVD), /*Index=*/0);
       InputInfo.PointersArray = CGF.Builder.CreateConstArrayGEP(
           CGF.GetAddrOfLocalVar(PVD), /*Index=*/0);
       InputInfo.SizesArray = CGF.Builder.CreateConstArrayGEP(
           CGF.GetAddrOfLocalVar(SVD), /*Index=*/0);
       // If MVD is nullptr, the mapper array is not privatized
       if (MVD)
         InputInfo.MappersArray = CGF.Builder.CreateConstArrayGEP(
             CGF.GetAddrOfLocalVar(MVD), /*Index=*/0);
     }
 
     Action.Enter(CGF);
     OMPLexicalScope LexScope(CGF, S, OMPD_task, /*EmitPreInitStmt=*/false);
     BodyGen(CGF);
   };
   llvm::Function *OutlinedFn = CGM.getOpenMPRuntime().emitTaskOutlinedFunction(
       S, *I, *PartId, *TaskT, S.getDirectiveKind(), CodeGen, /*Tied=*/true,
       Data.NumberOfParts);
   llvm::APInt TrueOrFalse(32, S.hasClausesOfKind<OMPNowaitClause>() ? 1 : 0);
   IntegerLiteral IfCond(getContext(), TrueOrFalse,
                         getContext().getIntTypeForBitwidth(32, /*Signed=*/0),
                         SourceLocation());
-
   CGM.getOpenMPRuntime().emitTaskCall(*this, S.getBeginLoc(), S, OutlinedFn,
                                       SharedsTy, CapturedStruct, &IfCond, Data);
 }
 
+void CodeGenFunction::processInReduction(const OMPExecutableDirective &S,
+                                         OMPTaskDataTy &Data,
+                                         CodeGenFunction &CGF,
+                                         const CapturedStmt *CS,
+                                         OMPPrivateScope &Scope) {
+  if (Data.Reductions) {
+    OpenMPDirectiveKind CapturedRegion = S.getDirectiveKind();
+    OMPLexicalScope LexScope(CGF, S, CapturedRegion);
+    ReductionCodeGen RedCG(Data.ReductionVars, Data.ReductionVars,
+                           Data.ReductionCopies, Data.ReductionOps);
+    llvm::Value *ReductionsPtr = CGF.Builder.CreateLoad(
+        CGF.GetAddrOfLocalVar(CS->getCapturedDecl()->getParam(4)));
+    for (unsigned Cnt = 0, E = Data.ReductionVars.size(); Cnt < E; ++Cnt) {
+      RedCG.emitSharedOrigLValue(CGF, Cnt);
+      RedCG.emitAggregateType(CGF, Cnt);
+      // FIXME: This must removed once the runtime library is fixed.
+      // Emit required threadprivate variables for
+      // initializer/combiner/finalizer.
+      CGF.CGM.getOpenMPRuntime().emitTaskReductionFixups(CGF, S.getBeginLoc(),
+                                                         RedCG, Cnt);
+      Address Replacement = CGF.CGM.getOpenMPRuntime().getTaskReductionItem(
+          CGF, S.getBeginLoc(), ReductionsPtr, RedCG.getSharedLValue(Cnt));
+      Replacement =
+          Address(CGF.EmitScalarConversion(
+                      Replacement.getPointer(), CGF.getContext().VoidPtrTy,
+                      CGF.getContext().getPointerType(
+                          Data.ReductionCopies[Cnt]->getType()),
+                      Data.ReductionCopies[Cnt]->getExprLoc()),
+                  CGF.ConvertTypeForMem(Data.ReductionCopies[Cnt]->getType()),
+                  Replacement.getAlignment());
+      Replacement = RedCG.adjustPrivateAddress(CGF, Cnt, Replacement);
+      Scope.addPrivate(RedCG.getBaseDecl(Cnt), Replacement);
+    }
+  }
+  (void)Scope.Privatize();
+  SmallVector<const Expr *, 4> InRedVars;
+  SmallVector<const Expr *, 4> InRedPrivs;
+  SmallVector<const Expr *, 4> InRedOps;
+  SmallVector<const Expr *, 4> TaskgroupDescriptors;
+  for (const auto *C : S.getClausesOfKind<OMPInReductionClause>()) {
+    auto IPriv = C->privates().begin();
+    auto IRed = C->reduction_ops().begin();
+    auto ITD = C->taskgroup_descriptors().begin();
+    for (const Expr *Ref : C->varlists()) {
+      InRedVars.emplace_back(Ref);
+      InRedPrivs.emplace_back(*IPriv);
+      InRedOps.emplace_back(*IRed);
+      TaskgroupDescriptors.emplace_back(*ITD);
+      std::advance(IPriv, 1);
+      std::advance(IRed, 1);
+      std::advance(ITD, 1);
+    }
+  }
+  OMPPrivateScope InRedScope(CGF);
+  if (!InRedVars.empty()) {
+    ReductionCodeGen RedCG(InRedVars, InRedVars, InRedPrivs, InRedOps);
+    for (unsigned Cnt = 0, E = InRedVars.size(); Cnt < E; ++Cnt) {
+      RedCG.emitSharedOrigLValue(CGF, Cnt);
+      RedCG.emitAggregateType(CGF, Cnt);
+      // FIXME: This must removed once the runtime library is fixed.
+      // Emit required threadprivate variables for
+      // initializer/combiner/finalizer.
+      CGF.CGM.getOpenMPRuntime().emitTaskReductionFixups(CGF, S.getBeginLoc(),
+                                                         RedCG, Cnt);
+      llvm::Value *ReductionsPtr;
+      if (const Expr *TRExpr = TaskgroupDescriptors[Cnt]) {
+        ReductionsPtr =
+            CGF.EmitLoadOfScalar(CGF.EmitLValue(TRExpr), TRExpr->getExprLoc());
+      } else {
+        ReductionsPtr = llvm::ConstantPointerNull::get(CGF.VoidPtrTy);
+      }
+      Address Replacement = CGF.CGM.getOpenMPRuntime().getTaskReductionItem(
+          CGF, S.getBeginLoc(), ReductionsPtr, RedCG.getSharedLValue(Cnt));
+      Replacement = Address(
+          CGF.EmitScalarConversion(
+              Replacement.getPointer(), CGF.getContext().VoidPtrTy,
+              CGF.getContext().getPointerType(InRedPrivs[Cnt]->getType()),
+              InRedPrivs[Cnt]->getExprLoc()),
+          CGF.ConvertTypeForMem(InRedPrivs[Cnt]->getType()),
+          Replacement.getAlignment());
+      Replacement = RedCG.adjustPrivateAddress(CGF, Cnt, Replacement);
+      InRedScope.addPrivate(RedCG.getBaseDecl(Cnt), Replacement);
+    }
+  }
+  (void)InRedScope.Privatize();
+}
+
 void CodeGenFunction::EmitOMPTaskDirective(const OMPTaskDirective &S) {
   // Emit outlined function for task construct.
   const CapturedStmt *CS = S.getCapturedStmt(OMPD_task);
   Address CapturedStruct = GenerateCapturedStmtArgument(*CS);
   QualType SharedsTy = getContext().getRecordType(CS->getCapturedRecordDecl());
   const Expr *IfCond = nullptr;
   for (const auto *C : S.getClausesOfKind<OMPIfClause>()) {
     if (C->getNameModifier() == OMPD_unknown ||

clang/lib/CodeGen/CodeGenFunction.h

OMPTargetDataInfo(Address BasePointersArray, Address PointersArray,
                       unsigned NumberOfTargetItems)
         : BasePointersArray(BasePointersArray), PointersArray(PointersArray),
           SizesArray(SizesArray), MappersArray(MappersArray),
           NumberOfTargetItems(NumberOfTargetItems) {}
   };
   void EmitOMPTargetTaskBasedDirective(const OMPExecutableDirective &S,
                                        const RegionCodeGenTy &BodyGen,
                                        OMPTargetDataInfo &InputInfo);
-
+  void processInReduction(const OMPExecutableDirective &S,
+                          OMPTaskDataTy &Data,
+                          CodeGenFunction &CGF,
+                          const CapturedStmt *CS,
+                          OMPPrivateScope &Scope);
   void EmitOMPMetaDirective(const OMPMetaDirective &S);
   void EmitOMPParallelDirective(const OMPParallelDirective &S);
   void EmitOMPSimdDirective(const OMPSimdDirective &S);
   void EmitOMPTileDirective(const OMPTileDirective &S);
   void EmitOMPUnrollDirective(const OMPUnrollDirective &S);
   void EmitOMPForDirective(const OMPForDirective &S);
   void EmitOMPForSimdDirective(const OMPForSimdDirective &S);
   void EmitOMPSectionsDirective(const OMPSectionsDirective &S);

clang/lib/Sema/SemaOpenMP.cpp

StmtResult Sema::ActOnOpenMPRegionEnd(StmtResult S,
   getOpenMPCaptureRegions(CaptureRegions, DSAStack->getCurrentDirective());
   OMPOrderedClause *OC = nullptr;
   OMPScheduleClause *SC = nullptr;
   SmallVector<const OMPLinearClause *, 4> LCs;
   SmallVector<const OMPClauseWithPreInit *, 4> PICs;
   // This is required for proper codegen.
   for (OMPClause *Clause : Clauses) {
     if (!LangOpts.OpenMPSimd &&
-        isOpenMPTaskingDirective(DSAStack->getCurrentDirective()) &&
+        (isOpenMPTaskingDirective(DSAStack->getCurrentDirective()) ||
+         DSAStack->getCurrentDirective() == OMPD_target) &&
         Clause->getClauseKind() == OMPC_in_reduction) {
       // Capture taskgroup task_reduction descriptors inside the tasking regions
       // with the corresponding in_reduction items.
       auto *IRC = cast<OMPInReductionClause>(Clause);
       for (Expr *E : IRC->taskgroup_descriptors())
         if (E)
           MarkDeclarationsReferencedInExpr(E);
     }

clang/test/OpenMP/target_in_reduction_codegen.cpp

+// RUN: %clang_cc1 -no-opaque-pointers -no-enable-noundef-analysis -verify -triple x86_64-apple-darwin10 -fopenmp -x c++ -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK1
+// RUN: %clang_cc1 -no-opaque-pointers -no-enable-noundef-analysis -fopenmp -x c++ -triple x86_64-apple-darwin10 -emit-pch -o %t %s
+// RUN: %clang_cc1 -no-opaque-pointers -no-enable-noundef-analysis -fopenmp -x c++ -triple x86_64-apple-darwin10 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix=CHECK1
+// expected-no-diagnostics
+#ifndef HEADER
+#define HEADER
+
+struct S {
+  int a;
+  S() : a(0) {}
+  S(const S &) {}
+  S &operator=(const S &) { return *this; }
+  ~S() {}
+  friend S operator+(const S &a, const S &b) { return a; }
+};
+
+int main(int argc, char **argv) {
+  int a;
+  float b;
+  S c[5];
+  short d[argc];
+#pragma omp taskgroup task_reduction(+ \
+                                     : a, b, argc)
+  {
+#pragma omp taskgroup task_reduction(- \
+                                     : c, d)
+#pragma omp parallel
+#pragma omp target in_reduction(+ \
+                                : a)
+    for (int i = 0; i < 5; i++)
+      a += d[a];
+  }
+  return 0;
+}
+
+#endif
+// CHECK1-LABEL: define {{[^@]+}}@main
+// CHECK1-SAME: (i32 [[ARGC:%.*]], i8** [[ARGV:%.*]]) #[[ATTR0:[0-9]+]] {
+// CHECK1-NEXT:  entry:
+// CHECK1-NEXT:    [[RETVAL:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[ARGC_ADDR:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[ARGV_ADDR:%.*]] = alloca i8**, align 8
+// CHECK1-NEXT:    [[A:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[B:%.*]] = alloca float, align 4
+// CHECK1-NEXT:    [[C:%.*]] = alloca [5 x %struct.S], align 16
+// CHECK1-NEXT:    [[SAVED_STACK:%.*]] = alloca i8*, align 8
+// CHECK1-NEXT:    [[__VLA_EXPR0:%.*]] = alloca i64, align 8
+// CHECK1-NEXT:    [[DOTRD_INPUT_:%.*]] = alloca [3 x %struct.kmp_taskred_input_t], align 8
+// CHECK1-NEXT:    [[DOTTASK_RED_:%.*]] = alloca i8*, align 8
+// CHECK1-NEXT:    [[DOTRD_INPUT_3:%.*]] = alloca [2 x %struct.kmp_taskred_input_t.0], align 8
+// CHECK1-NEXT:    [[DOTTASK_RED_6:%.*]] = alloca i8*, align 8
+// CHECK1-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1:[0-9]+]])
+// CHECK1-NEXT:    store i32 0, i32* [[RETVAL]], align 4
+// CHECK1-NEXT:    store i32 [[ARGC]], i32* [[ARGC_ADDR]], align 4
+// CHECK1-NEXT:    store i8** [[ARGV]], i8*** [[ARGV_ADDR]], align 8
+// CHECK1-NEXT:    [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [5 x %struct.S], [5 x %struct.S]* [[C]], i32 0, i32 0
+// CHECK1-NEXT:    [[ARRAYCTOR_END:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAY_BEGIN]], i64 5
+// CHECK1-NEXT:    br label [[ARRAYCTOR_LOOP:%.*]]
+// CHECK1:       arrayctor.loop:
+// CHECK1-NEXT:    [[ARRAYCTOR_CUR:%.*]] = phi %struct.S* [ [[ARRAY_BEGIN]], [[ENTRY:%.*]] ], [ [[ARRAYCTOR_NEXT:%.*]], [[ARRAYCTOR_LOOP]] ]
+// CHECK1-NEXT:    call void @_ZN1SC1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYCTOR_CUR]])
+// CHECK1-NEXT:    [[ARRAYCTOR_NEXT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYCTOR_CUR]], i64 1
+// CHECK1-NEXT:    [[ARRAYCTOR_DONE:%.*]] = icmp eq %struct.S* [[ARRAYCTOR_NEXT]], [[ARRAYCTOR_END]]
+// CHECK1-NEXT:    br i1 [[ARRAYCTOR_DONE]], label [[ARRAYCTOR_CONT:%.*]], label [[ARRAYCTOR_LOOP]]
+// CHECK1:       arrayctor.cont:
+// CHECK1-NEXT:    [[TMP1:%.*]] = load i32, i32* [[ARGC_ADDR]], align 4
+// CHECK1-NEXT:    [[TMP2:%.*]] = zext i32 [[TMP1]] to i64
+// CHECK1-NEXT:    [[TMP3:%.*]] = call i8* @llvm.stacksave()
+// CHECK1-NEXT:    store i8* [[TMP3]], i8** [[SAVED_STACK]], align 8
+// CHECK1-NEXT:    [[VLA:%.*]] = alloca i16, i64 [[TMP2]], align 16
+// CHECK1-NEXT:    store i64 [[TMP2]], i64* [[__VLA_EXPR0]], align 8
+// CHECK1-NEXT:    call void @__kmpc_taskgroup(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+// CHECK1-NEXT:    [[DOTRD_INPUT_GEP_:%.*]] = getelementptr inbounds [3 x %struct.kmp_taskred_input_t], [3 x %struct.kmp_taskred_input_t]* [[DOTRD_INPUT_]], i64 0, i64 0
+// CHECK1-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T:%.*]], %struct.kmp_taskred_input_t* [[DOTRD_INPUT_GEP_]], i32 0, i32 0
+// CHECK1-NEXT:    [[TMP5:%.*]] = bitcast i32* [[A]] to i8*
+// CHECK1-NEXT:    store i8* [[TMP5]], i8** [[TMP4]], align 8
+// CHECK1-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], %struct.kmp_taskred_input_t* [[DOTRD_INPUT_GEP_]], i32 0, i32 1
+// CHECK1-NEXT:    [[TMP7:%.*]] = bitcast i32* [[A]] to i8*
+// CHECK1-NEXT:    store i8* [[TMP7]], i8** [[TMP6]], align 8
+// CHECK1-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], %struct.kmp_taskred_input_t* [[DOTRD_INPUT_GEP_]], i32 0, i32 2
+// CHECK1-NEXT:    store i64 4, i64* [[TMP8]], align 8
+// CHECK1-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], %struct.kmp_taskred_input_t* [[DOTRD_INPUT_GEP_]], i32 0, i32 3
+// CHECK1-NEXT:    store i8* bitcast (void (i8*, i8*)* @.red_init. to i8*), i8** [[TMP9]], align 8
+// CHECK1-NEXT:    [[TMP10:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], %struct.kmp_taskred_input_t* [[DOTRD_INPUT_GEP_]], i32 0, i32 4
+// CHECK1-NEXT:    store i8* null, i8** [[TMP10]], align 8
+// CHECK1-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], %struct.kmp_taskred_input_t* [[DOTRD_INPUT_GEP_]], i32 0, i32 5
+// CHECK1-NEXT:    store i8* bitcast (void (i8*, i8*)* @.red_comb. to i8*), i8** [[TMP11]], align 8
+// CHECK1-NEXT:    [[TMP12:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], %struct.kmp_taskred_input_t* [[DOTRD_INPUT_GEP_]], i32 0, i32 6
+// CHECK1-NEXT:    [[TMP13:%.*]] = bitcast i32* [[TMP12]] to i8*
+// CHECK1-NEXT:    call void @llvm.memset.p0i8.i64(i8* align 8 [[TMP13]], i8 0, i64 4, i1 false)
+// CHECK1-NEXT:    [[DOTRD_INPUT_GEP_1:%.*]] = getelementptr inbounds [3 x %struct.kmp_taskred_input_t], [3 x %struct.kmp_taskred_input_t]* [[DOTRD_INPUT_]], i64 0, i64 1
+// CHECK1-NEXT:    [[TMP14:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], %struct.kmp_taskred_input_t* [[DOTRD_INPUT_GEP_1]], i32 0, i32 0
+// CHECK1-NEXT:    [[TMP15:%.*]] = bitcast float* [[B]] to i8*
+// CHECK1-NEXT:    store i8* [[TMP15]], i8** [[TMP14]], align 8
+// CHECK1-NEXT:    [[TMP16:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], %struct.kmp_taskred_input_t* [[DOTRD_INPUT_GEP_1]], i32 0, i32 1
+// CHECK1-NEXT:    [[TMP17:%.*]] = bitcast float* [[B]] to i8*
+// CHECK1-NEXT:    store i8* [[TMP17]], i8** [[TMP16]], align 8
+// CHECK1-NEXT:    [[TMP18:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], %struct.kmp_taskred_input_t* [[DOTRD_INPUT_GEP_1]], i32 0, i32 2
+// CHECK1-NEXT:    store i64 4, i64* [[TMP18]], align 8
+// CHECK1-NEXT:    [[TMP19:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], %struct.kmp_taskred_input_t* [[DOTRD_INPUT_GEP_1]], i32 0, i32 3
+// CHECK1-NEXT:    store i8* bitcast (void (i8*, i8*)* @.red_init..1 to i8*), i8** [[TMP19]], align 8
+// CHECK1-NEXT:    [[TMP20:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], %struct.kmp_taskred_input_t* [[DOTRD_INPUT_GEP_1]], i32 0, i32 4
+// CHECK1-NEXT:    store i8* null, i8** [[TMP20]], align 8
+// CHECK1-NEXT:    [[TMP21:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], %struct.kmp_taskred_input_t* [[DOTRD_INPUT_GEP_1]], i32 0, i32 5
+// CHECK1-NEXT:    store i8* bitcast (void (i8*, i8*)* @.red_comb..2 to i8*), i8** [[TMP21]], align 8
+// CHECK1-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], %struct.kmp_taskred_input_t* [[DOTRD_INPUT_GEP_1]], i32 0, i32 6
+// CHECK1-NEXT:    [[TMP23:%.*]] = bitcast i32* [[TMP22]] to i8*
+// CHECK1-NEXT:    call void @llvm.memset.p0i8.i64(i8* align 8 [[TMP23]], i8 0, i64 4, i1 false)
+// CHECK1-NEXT:    [[DOTRD_INPUT_GEP_2:%.*]] = getelementptr inbounds [3 x %struct.kmp_taskred_input_t], [3 x %struct.kmp_taskred_input_t]* [[DOTRD_INPUT_]], i64 0, i64 2
+// CHECK1-NEXT:    [[TMP24:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], %struct.kmp_taskred_input_t* [[DOTRD_INPUT_GEP_2]], i32 0, i32 0
+// CHECK1-NEXT:    [[TMP25:%.*]] = bitcast i32* [[ARGC_ADDR]] to i8*
+// CHECK1-NEXT:    store i8* [[TMP25]], i8** [[TMP24]], align 8
+// CHECK1-NEXT:    [[TMP26:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], %struct.kmp_taskred_input_t* [[DOTRD_INPUT_GEP_2]], i32 0, i32 1
+// CHECK1-NEXT:    [[TMP27:%.*]] = bitcast i32* [[ARGC_ADDR]] to i8*
+// CHECK1-NEXT:    store i8* [[TMP27]], i8** [[TMP26]], align 8
+// CHECK1-NEXT:    [[TMP28:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], %struct.kmp_taskred_input_t* [[DOTRD_INPUT_GEP_2]], i32 0, i32 2
+// CHECK1-NEXT:    store i64 4, i64* [[TMP28]], align 8
+// CHECK1-NEXT:    [[TMP29:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], %struct.kmp_taskred_input_t* [[DOTRD_INPUT_GEP_2]], i32 0, i32 3
+// CHECK1-NEXT:    store i8* bitcast (void (i8*, i8*)* @.red_init..3 to i8*), i8** [[TMP29]], align 8
+// CHECK1-NEXT:    [[TMP30:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], %struct.kmp_taskred_input_t* [[DOTRD_INPUT_GEP_2]], i32 0, i32 4
+// CHECK1-NEXT:    store i8* null, i8** [[TMP30]], align 8
+// CHECK1-NEXT:    [[TMP31:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], %struct.kmp_taskred_input_t* [[DOTRD_INPUT_GEP_2]], i32 0, i32 5
+// CHECK1-NEXT:    store i8* bitcast (void (i8*, i8*)* @.red_comb..4 to i8*), i8** [[TMP31]], align 8
+// CHECK1-NEXT:    [[TMP32:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], %struct.kmp_taskred_input_t* [[DOTRD_INPUT_GEP_2]], i32 0, i32 6
+// CHECK1-NEXT:    [[TMP33:%.*]] = bitcast i32* [[TMP32]] to i8*
+// CHECK1-NEXT:    call void @llvm.memset.p0i8.i64(i8* align 8 [[TMP33]], i8 0, i64 4, i1 false)
+// CHECK1-NEXT:    [[TMP34:%.*]] = bitcast [3 x %struct.kmp_taskred_input_t]* [[DOTRD_INPUT_]] to i8*
+// CHECK1-NEXT:    [[TMP35:%.*]] = call i8* @__kmpc_taskred_init(i32 [[TMP0]], i32 3, i8* [[TMP34]])
+// CHECK1-NEXT:    store i8* [[TMP35]], i8** [[DOTTASK_RED_]], align 8
+// CHECK1-NEXT:    call void @__kmpc_taskgroup(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+// CHECK1-NEXT:    [[DOTRD_INPUT_GEP_4:%.*]] = getelementptr inbounds [2 x %struct.kmp_taskred_input_t.0], [2 x %struct.kmp_taskred_input_t.0]* [[DOTRD_INPUT_3]], i64 0, i64 0
+// CHECK1-NEXT:    [[TMP36:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0:%.*]], %struct.kmp_taskred_input_t.0* [[DOTRD_INPUT_GEP_4]], i32 0, i32 0
+// CHECK1-NEXT:    [[TMP37:%.*]] = bitcast [5 x %struct.S]* [[C]] to i8*
+// CHECK1-NEXT:    store i8* [[TMP37]], i8** [[TMP36]], align 8
+// CHECK1-NEXT:    [[TMP38:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0]], %struct.kmp_taskred_input_t.0* [[DOTRD_INPUT_GEP_4]], i32 0, i32 1
+// CHECK1-NEXT:    [[TMP39:%.*]] = bitcast [5 x %struct.S]* [[C]] to i8*
+// CHECK1-NEXT:    store i8* [[TMP39]], i8** [[TMP38]], align 8
+// CHECK1-NEXT:    [[TMP40:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0]], %struct.kmp_taskred_input_t.0* [[DOTRD_INPUT_GEP_4]], i32 0, i32 2
+// CHECK1-NEXT:    store i64 20, i64* [[TMP40]], align 8
+// CHECK1-NEXT:    [[TMP41:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0]], %struct.kmp_taskred_input_t.0* [[DOTRD_INPUT_GEP_4]], i32 0, i32 3
+// CHECK1-NEXT:    store i8* bitcast (void (i8*, i8*)* @.red_init..5 to i8*), i8** [[TMP41]], align 8
+// CHECK1-NEXT:    [[TMP42:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0]], %struct.kmp_taskred_input_t.0* [[DOTRD_INPUT_GEP_4]], i32 0, i32 4
+// CHECK1-NEXT:    store i8* bitcast (void (i8*)* @.red_fini. to i8*), i8** [[TMP42]], align 8
+// CHECK1-NEXT:    [[TMP43:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0]], %struct.kmp_taskred_input_t.0* [[DOTRD_INPUT_GEP_4]], i32 0, i32 5
+// CHECK1-NEXT:    store i8* bitcast (void (i8*, i8*)* @.red_comb..6 to i8*), i8** [[TMP43]], align 8
+// CHECK1-NEXT:    [[TMP44:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0]], %struct.kmp_taskred_input_t.0* [[DOTRD_INPUT_GEP_4]], i32 0, i32 6
+// CHECK1-NEXT:    [[TMP45:%.*]] = bitcast i32* [[TMP44]] to i8*
+// CHECK1-NEXT:    call void @llvm.memset.p0i8.i64(i8* align 8 [[TMP45]], i8 0, i64 4, i1 false)
+// CHECK1-NEXT:    [[DOTRD_INPUT_GEP_5:%.*]] = getelementptr inbounds [2 x %struct.kmp_taskred_input_t.0], [2 x %struct.kmp_taskred_input_t.0]* [[DOTRD_INPUT_3]], i64 0, i64 1
+// CHECK1-NEXT:    [[TMP46:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0]], %struct.kmp_taskred_input_t.0* [[DOTRD_INPUT_GEP_5]], i32 0, i32 0
+// CHECK1-NEXT:    [[TMP47:%.*]] = bitcast i16* [[VLA]] to i8*
+// CHECK1-NEXT:    store i8* [[TMP47]], i8** [[TMP46]], align 8
+// CHECK1-NEXT:    [[TMP48:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0]], %struct.kmp_taskred_input_t.0* [[DOTRD_INPUT_GEP_5]], i32 0, i32 1
+// CHECK1-NEXT:    [[TMP49:%.*]] = bitcast i16* [[VLA]] to i8*
+// CHECK1-NEXT:    store i8* [[TMP49]], i8** [[TMP48]], align 8
+// CHECK1-NEXT:    [[TMP50:%.*]] = mul nuw i64 [[TMP2]], 2
+// CHECK1-NEXT:    [[TMP51:%.*]] = udiv exact i64 [[TMP50]], ptrtoint (i16* getelementptr (i16, i16* null, i32 1) to i64)
+// CHECK1-NEXT:    [[TMP52:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0]], %struct.kmp_taskred_input_t.0* [[DOTRD_INPUT_GEP_5]], i32 0, i32 2
+// CHECK1-NEXT:    store i64 [[TMP50]], i64* [[TMP52]], align 8
+// CHECK1-NEXT:    [[TMP53:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0]], %struct.kmp_taskred_input_t.0* [[DOTRD_INPUT_GEP_5]], i32 0, i32 3
+// CHECK1-NEXT:    store i8* bitcast (void (i8*, i8*)* @.red_init..7 to i8*), i8** [[TMP53]], align 8
+// CHECK1-NEXT:    [[TMP54:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0]], %struct.kmp_taskred_input_t.0* [[DOTRD_INPUT_GEP_5]], i32 0, i32 4
+// CHECK1-NEXT:    store i8* null, i8** [[TMP54]], align 8
+// CHECK1-NEXT:    [[TMP55:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0]], %struct.kmp_taskred_input_t.0* [[DOTRD_INPUT_GEP_5]], i32 0, i32 5
+// CHECK1-NEXT:    store i8* bitcast (void (i8*, i8*)* @.red_comb..8 to i8*), i8** [[TMP55]], align 8
+// CHECK1-NEXT:    [[TMP56:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0]], %struct.kmp_taskred_input_t.0* [[DOTRD_INPUT_GEP_5]], i32 0, i32 6
+// CHECK1-NEXT:    store i32 1, i32* [[TMP56]], align 8
+// CHECK1-NEXT:    [[TMP57:%.*]] = bitcast [2 x %struct.kmp_taskred_input_t.0]* [[DOTRD_INPUT_3]] to i8*
+// CHECK1-NEXT:    [[TMP58:%.*]] = call i8* @__kmpc_taskred_init(i32 [[TMP0]], i32 2, i8* [[TMP57]])
+// CHECK1-NEXT:    store i8* [[TMP58]], i8** [[DOTTASK_RED_6]], align 8
+// CHECK1-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB1]], i32 4, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32*, i64, i16*, i8**)* @.omp_outlined. to void (i32*, i32*, ...)*), i32* [[A]], i64 [[TMP2]], i16* [[VLA]], i8** [[DOTTASK_RED_]])
+// CHECK1-NEXT:    call void @__kmpc_end_taskgroup(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+// CHECK1-NEXT:    call void @__kmpc_end_taskgroup(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+// CHECK1-NEXT:    store i32 0, i32* [[RETVAL]], align 4
+// CHECK1-NEXT:    [[TMP59:%.*]] = load i8*, i8** [[SAVED_STACK]], align 8
+// CHECK1-NEXT:    call void @llvm.stackrestore(i8* [[TMP59]])
+// CHECK1-NEXT:    [[ARRAY_BEGIN7:%.*]] = getelementptr inbounds [5 x %struct.S], [5 x %struct.S]* [[C]], i32 0, i32 0
+// CHECK1-NEXT:    [[TMP60:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAY_BEGIN7]], i64 5
+// CHECK1-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
+// CHECK1:       arraydestroy.body:
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ [[TMP60]], [[ARRAYCTOR_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK1-NEXT:    call void @_ZN1SD1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR3:[0-9]+]]
+// CHECK1-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN7]]
+// CHECK1-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE8:%.*]], label [[ARRAYDESTROY_BODY]]
+// CHECK1:       arraydestroy.done8:
+// CHECK1-NEXT:    [[TMP61:%.*]] = load i32, i32* [[RETVAL]], align 4
+// CHECK1-NEXT:    ret i32 [[TMP61]]
+//
+//
+// CHECK1-LABEL: define {{[^@]+}}@_ZN1SC1Ev
+// CHECK1-SAME: (%struct.S* nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR1:[0-9]+]] align 2 {
+// CHECK1-NEXT:  entry:
+// CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca %struct.S*, align 8
+// CHECK1-NEXT:    store %struct.S* [[THIS]], %struct.S** [[THIS_ADDR]], align 8
+// CHECK1-NEXT:    [[THIS1:%.*]] = load %struct.S*, %struct.S** [[THIS_ADDR]], align 8
+// CHECK1-NEXT:    call void @_ZN1SC2Ev(%struct.S* nonnull align 4 dereferenceable(4) [[THIS1]])
+// CHECK1-NEXT:    ret void
+//
+//
+// CHECK1-LABEL: define {{[^@]+}}@.red_init.
+// CHECK1-SAME: (i8* noalias [[TMP0:%.*]], i8* noalias [[TMP1:%.*]]) #[[ATTR5:[0-9]+]] {
+// CHECK1-NEXT:  entry:
+// CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
+// CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca i8*, align 8
+// CHECK1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
+// CHECK1-NEXT:    store i8* [[TMP1]], i8** [[DOTADDR1]], align 8
+// CHECK1-NEXT:    [[TMP2:%.*]] = bitcast i8** [[DOTADDR]] to i32**
+// CHECK1-NEXT:    [[TMP3:%.*]] = load i32*, i32** [[TMP2]], align 8
+// CHECK1-NEXT:    store i32 0, i32* [[TMP3]], align 4
+// CHECK1-NEXT:    ret void
+//
+//
+// CHECK1-LABEL: define {{[^@]+}}@.red_comb.
+// CHECK1-SAME: (i8* [[TMP0:%.*]], i8* [[TMP1:%.*]]) #[[ATTR5]] {
+// CHECK1-NEXT:  entry:
+// CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
+// CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca i8*, align 8
+// CHECK1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
+// CHECK1-NEXT:    store i8* [[TMP1]], i8** [[DOTADDR1]], align 8
+// CHECK1-NEXT:    [[TMP2:%.*]] = bitcast i8** [[DOTADDR]] to i32**
+// CHECK1-NEXT:    [[TMP3:%.*]] = load i32*, i32** [[TMP2]], align 8
+// CHECK1-NEXT:    [[TMP4:%.*]] = bitcast i8** [[DOTADDR1]] to i32**
+// CHECK1-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[TMP4]], align 8
+// CHECK1-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP3]], align 4
+// CHECK1-NEXT:    [[TMP7:%.*]] = load i32, i32* [[TMP5]], align 4
+// CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP6]], [[TMP7]]
+// CHECK1-NEXT:    store i32 [[ADD]], i32* [[TMP3]], align 4
+// CHECK1-NEXT:    ret void
+//
+//
+// CHECK1-LABEL: define {{[^@]+}}@.red_init..1
+// CHECK1-SAME: (i8* noalias [[TMP0:%.*]], i8* noalias [[TMP1:%.*]]) #[[ATTR5]] {
+// CHECK1-NEXT:  entry:
+// CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
+// CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca i8*, align 8
+// CHECK1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
+// CHECK1-NEXT:    store i8* [[TMP1]], i8** [[DOTADDR1]], align 8
+// CHECK1-NEXT:    [[TMP2:%.*]] = bitcast i8** [[DOTADDR]] to float**
+// CHECK1-NEXT:    [[TMP3:%.*]] = load float*, float** [[TMP2]], align 8
+// CHECK1-NEXT:    store float 0.000000e+00, float* [[TMP3]], align 4
+// CHECK1-NEXT:    ret void
+//
+//
+// CHECK1-LABEL: define {{[^@]+}}@.red_comb..2
+// CHECK1-SAME: (i8* [[TMP0:%.*]], i8* [[TMP1:%.*]]) #[[ATTR5]] {
+// CHECK1-NEXT:  entry:
+// CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
+// CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca i8*, align 8
+// CHECK1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
+// CHECK1-NEXT:    store i8* [[TMP1]], i8** [[DOTADDR1]], align 8
+// CHECK1-NEXT:    [[TMP2:%.*]] = bitcast i8** [[DOTADDR]] to float**
+// CHECK1-NEXT:    [[TMP3:%.*]] = load float*, float** [[TMP2]], align 8
+// CHECK1-NEXT:    [[TMP4:%.*]] = bitcast i8** [[DOTADDR1]] to float**
+// CHECK1-NEXT:    [[TMP5:%.*]] = load float*, float** [[TMP4]], align 8
+// CHECK1-NEXT:    [[TMP6:%.*]] = load float, float* [[TMP3]], align 4
+// CHECK1-NEXT:    [[TMP7:%.*]] = load float, float* [[TMP5]], align 4
+// CHECK1-NEXT:    [[ADD:%.*]] = fadd float [[TMP6]], [[TMP7]]
+// CHECK1-NEXT:    store float [[ADD]], float* [[TMP3]], align 4
+// CHECK1-NEXT:    ret void
+//
+//
+// CHECK1-LABEL: define {{[^@]+}}@.red_init..3
+// CHECK1-SAME: (i8* noalias [[TMP0:%.*]], i8* noalias [[TMP1:%.*]]) #[[ATTR5]] {
+// CHECK1-NEXT:  entry:
+// CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
+// CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca i8*, align 8
+// CHECK1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
+// CHECK1-NEXT:    store i8* [[TMP1]], i8** [[DOTADDR1]], align 8
+// CHECK1-NEXT:    [[TMP2:%.*]] = bitcast i8** [[DOTADDR]] to i32**
+// CHECK1-NEXT:    [[TMP3:%.*]] = load i32*, i32** [[TMP2]], align 8
+// CHECK1-NEXT:    store i32 0, i32* [[TMP3]], align 4
+// CHECK1-NEXT:    ret void
+//
+//
+// CHECK1-LABEL: define {{[^@]+}}@.red_comb..4
+// CHECK1-SAME: (i8* [[TMP0:%.*]], i8* [[TMP1:%.*]]) #[[ATTR5]] {
+// CHECK1-NEXT:  entry:
+// CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
+// CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca i8*, align 8
+// CHECK1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
+// CHECK1-NEXT:    store i8* [[TMP1]], i8** [[DOTADDR1]], align 8
+// CHECK1-NEXT:    [[TMP2:%.*]] = bitcast i8** [[DOTADDR]] to i32**
+// CHECK1-NEXT:    [[TMP3:%.*]] = load i32*, i32** [[TMP2]], align 8
+// CHECK1-NEXT:    [[TMP4:%.*]] = bitcast i8** [[DOTADDR1]] to i32**
+// CHECK1-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[TMP4]], align 8
+// CHECK1-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP3]], align 4
+// CHECK1-NEXT:    [[TMP7:%.*]] = load i32, i32* [[TMP5]], align 4
+// CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP6]], [[TMP7]]
+// CHECK1-NEXT:    store i32 [[ADD]], i32* [[TMP3]], align 4
+// CHECK1-NEXT:    ret void
+//
+//
+// CHECK1-LABEL: define {{[^@]+}}@.red_init..5
+// CHECK1-SAME: (i8* noalias [[TMP0:%.*]], i8* noalias [[TMP1:%.*]]) #[[ATTR5]] {
+// CHECK1-NEXT:  entry:
+// CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
+// CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca i8*, align 8
+// CHECK1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
+// CHECK1-NEXT:    store i8* [[TMP1]], i8** [[DOTADDR1]], align 8
+// CHECK1-NEXT:    [[TMP2:%.*]] = bitcast i8** [[DOTADDR]] to [5 x %struct.S]**
+// CHECK1-NEXT:    [[TMP3:%.*]] = load [5 x %struct.S]*, [5 x %struct.S]** [[TMP2]], align 8
+// CHECK1-NEXT:    [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [5 x %struct.S], [5 x %struct.S]* [[TMP3]], i32 0, i32 0
+// CHECK1-NEXT:    [[TMP4:%.*]] = getelementptr [[STRUCT_S:%.*]], %struct.S* [[ARRAY_BEGIN]], i64 5
+// CHECK1-NEXT:    [[OMP_ARRAYINIT_ISEMPTY:%.*]] = icmp eq %struct.S* [[ARRAY_BEGIN]], [[TMP4]]
+// CHECK1-NEXT:    br i1 [[OMP_ARRAYINIT_ISEMPTY]], label [[OMP_ARRAYINIT_DONE:%.*]], label [[OMP_ARRAYINIT_BODY:%.*]]
+// CHECK1:       omp.arrayinit.body:
+// CHECK1-NEXT:    [[OMP_ARRAYCPY_DESTELEMENTPAST:%.*]] = phi %struct.S* [ [[ARRAY_BEGIN]], [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_DEST_ELEMENT:%.*]], [[OMP_ARRAYINIT_BODY]] ]
+// CHECK1-NEXT:    call void @_ZN1SC1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[OMP_ARRAYCPY_DESTELEMENTPAST]])
+// CHECK1-NEXT:    [[OMP_ARRAYCPY_DEST_ELEMENT]] = getelementptr [[STRUCT_S]], %struct.S* [[OMP_ARRAYCPY_DESTELEMENTPAST]], i32 1
+// CHECK1-NEXT:    [[OMP_ARRAYCPY_DONE:%.*]] = icmp eq %struct.S* [[OMP_ARRAYCPY_DEST_ELEMENT]], [[TMP4]]
+// CHECK1-NEXT:    br i1 [[OMP_ARRAYCPY_DONE]], label [[OMP_ARRAYINIT_DONE]], label [[OMP_ARRAYINIT_BODY]]
+// CHECK1:       omp.arrayinit.done:
+// CHECK1-NEXT:    ret void
+//
+//
+// CHECK1-LABEL: define {{[^@]+}}@.red_fini.
+// CHECK1-SAME: (i8* [[TMP0:%.*]]) #[[ATTR5]] {
+// CHECK1-NEXT:  entry:
+// CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
+// CHECK1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
+// CHECK1-NEXT:    [[TMP1:%.*]] = load i8*, i8** [[DOTADDR]], align 8
+// CHECK1-NEXT:    [[TMP2:%.*]] = bitcast i8* [[TMP1]] to [5 x %struct.S]*
+// CHECK1-NEXT:    [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [5 x %struct.S], [5 x %struct.S]* [[TMP2]], i32 0, i32 0
+// CHECK1-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAY_BEGIN]], i64 5
+// CHECK1-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
+// CHECK1:       arraydestroy.body:
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ [[TMP3]], [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK1-NEXT:    call void @_ZN1SD1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #3
+// CHECK1-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
+// CHECK1-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
+// CHECK1:       arraydestroy.done1:
+// CHECK1-NEXT:    ret void
+//
+//
+// CHECK1-LABEL: define {{[^@]+}}@_ZN1SD1Ev
+// CHECK1-SAME: (%struct.S* nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] align 2 {
+// CHECK1-NEXT:  entry:
+// CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca %struct.S*, align 8
+// CHECK1-NEXT:    store %struct.S* [[THIS]], %struct.S** [[THIS_ADDR]], align 8
+// CHECK1-NEXT:    [[THIS1:%.*]] = load %struct.S*, %struct.S** [[THIS_ADDR]], align 8
+// CHECK1-NEXT:    call void @_ZN1SD2Ev(%struct.S* nonnull align 4 dereferenceable(4) [[THIS1]]) #3
+// CHECK1-NEXT:    ret void
+//
+//
+// CHECK1-LABEL: define {{[^@]+}}@.red_comb..6
+// CHECK1-SAME: (i8* [[TMP0:%.*]], i8* [[TMP1:%.*]]) #[[ATTR5]] {
+// CHECK1-NEXT:  entry:
+// CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
+// CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca i8*, align 8
+// CHECK1-NEXT:    [[REF_TMP:%.*]] = alloca [[STRUCT_S:%.*]], align 4
+// CHECK1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
+// CHECK1-NEXT:    store i8* [[TMP1]], i8** [[DOTADDR1]], align 8
+// CHECK1-NEXT:    [[TMP2:%.*]] = bitcast i8** [[DOTADDR]] to %struct.S**
+// CHECK1-NEXT:    [[TMP3:%.*]] = load %struct.S*, %struct.S** [[TMP2]], align 8
+// CHECK1-NEXT:    [[TMP4:%.*]] = bitcast i8** [[DOTADDR1]] to %struct.S**
+// CHECK1-NEXT:    [[TMP5:%.*]] = load %struct.S*, %struct.S** [[TMP4]], align 8
+// CHECK1-NEXT:    [[TMP6:%.*]] = getelementptr [[STRUCT_S]], %struct.S* [[TMP3]], i64 5
+// CHECK1-NEXT:    [[OMP_ARRAYCPY_ISEMPTY:%.*]] = icmp eq %struct.S* [[TMP3]], [[TMP6]]
+// CHECK1-NEXT:    br i1 [[OMP_ARRAYCPY_ISEMPTY]], label [[OMP_ARRAYCPY_DONE2:%.*]], label [[OMP_ARRAYCPY_BODY:%.*]]
+// CHECK1:       omp.arraycpy.body:
+// CHECK1-NEXT:    [[OMP_ARRAYCPY_SRCELEMENTPAST:%.*]] = phi %struct.S* [ [[TMP5]], [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_SRC_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]
+// CHECK1-NEXT:    [[OMP_ARRAYCPY_DESTELEMENTPAST:%.*]] = phi %struct.S* [ [[TMP3]], [[ENTRY]] ], [ [[OMP_ARRAYCPY_DEST_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]
+// CHECK1-NEXT:    call void @_ZplRK1SS1_(%struct.S* sret([[STRUCT_S]]) align 4 [[REF_TMP]], %struct.S* nonnull align 4 dereferenceable(4) [[OMP_ARRAYCPY_DESTELEMENTPAST]], %struct.S* nonnull align 4 dereferenceable(4) [[OMP_ARRAYCPY_SRCELEMENTPAST]])
+// CHECK1-NEXT:    [[CALL:%.*]] = call nonnull align 4 dereferenceable(4) %struct.S* @_ZN1SaSERKS_(%struct.S* nonnull align 4 dereferenceable(4) [[OMP_ARRAYCPY_DESTELEMENTPAST]], %struct.S* nonnull align 4 dereferenceable(4) [[REF_TMP]])
+// CHECK1-NEXT:    call void @_ZN1SD1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[REF_TMP]]) #3
+// CHECK1-NEXT:    [[OMP_ARRAYCPY_DEST_ELEMENT]] = getelementptr [[STRUCT_S]], %struct.S* [[OMP_ARRAYCPY_DESTELEMENTPAST]], i32 1
+// CHECK1-NEXT:    [[OMP_ARRAYCPY_SRC_ELEMENT]] = getelementptr [[STRUCT_S]], %struct.S* [[OMP_ARRAYCPY_SRCELEMENTPAST]], i32 1
+// CHECK1-NEXT:    [[OMP_ARRAYCPY_DONE:%.*]] = icmp eq %struct.S* [[OMP_ARRAYCPY_DEST_ELEMENT]], [[TMP6]]
+// CHECK1-NEXT:    br i1 [[OMP_ARRAYCPY_DONE]], label [[OMP_ARRAYCPY_DONE2]], label [[OMP_ARRAYCPY_BODY]]
+// CHECK1:       omp.arraycpy.done2:
+// CHECK1-NEXT:    ret void
+//
+//
+// CHECK1-LABEL: define {{[^@]+}}@_ZplRK1SS1_
+// CHECK1-SAME: (%struct.S* noalias sret([[STRUCT_S:%.*]]) align 4 [[AGG_RESULT:%.*]], %struct.S* nonnull align 4 dereferenceable(4) [[A:%.*]], %struct.S* nonnull align 4 dereferenceable(4) [[B:%.*]]) #[[ATTR7:[0-9]+]] {
+// CHECK1-NEXT:  entry:
+// CHECK1-NEXT:    [[RESULT_PTR:%.*]] = alloca i8*, align 8
+// CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca %struct.S*, align 8
+// CHECK1-NEXT:    [[B_ADDR:%.*]] = alloca %struct.S*, align 8
+// CHECK1-NEXT:    [[TMP0:%.*]] = bitcast %struct.S* [[AGG_RESULT]] to i8*
+// CHECK1-NEXT:    store i8* [[TMP0]], i8** [[RESULT_PTR]], align 8
+// CHECK1-NEXT:    store %struct.S* [[A]], %struct.S** [[A_ADDR]], align 8
+// CHECK1-NEXT:    store %struct.S* [[B]], %struct.S** [[B_ADDR]], align 8
+// CHECK1-NEXT:    [[TMP1:%.*]] = load %struct.S*, %struct.S** [[A_ADDR]], align 8
+// CHECK1-NEXT:    call void @_ZN1SC1ERKS_(%struct.S* nonnull align 4 dereferenceable(4) [[AGG_RESULT]], %struct.S* nonnull align 4 dereferenceable(4) [[TMP1]])
+// CHECK1-NEXT:    ret void
+//
+//
+// CHECK1-LABEL: define {{[^@]+}}@_ZN1SaSERKS_
+// CHECK1-SAME: (%struct.S* nonnull align 4 dereferenceable(4) [[THIS:%.*]], %struct.S* nonnull align 4 dereferenceable(4) [[TMP0:%.*]]) #[[ATTR7]] align 2 {
+// CHECK1-NEXT:  entry:
+// CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca %struct.S*, align 8
+// CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca %struct.S*, align 8
+// CHECK1-NEXT:    store %struct.S* [[THIS]], %struct.S** [[THIS_ADDR]], align 8
+// CHECK1-NEXT:    store %struct.S* [[TMP0]], %struct.S** [[DOTADDR]], align 8
+// CHECK1-NEXT:    [[THIS1:%.*]] = load %struct.S*, %struct.S** [[THIS_ADDR]], align 8
+// CHECK1-NEXT:    ret %struct.S* [[THIS1]]
+//
+//
+// CHECK1-LABEL: define {{[^@]+}}@.red_init..7
+// CHECK1-SAME: (i8* noalias [[TMP0:%.*]], i8* noalias [[TMP1:%.*]]) #[[ATTR5]] {
+// CHECK1-NEXT:  entry:
+// CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
+// CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca i8*, align 8
+// CHECK1-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
+// CHECK1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
+// CHECK1-NEXT:    store i8* [[TMP1]], i8** [[DOTADDR1]], align 8
+// CHECK1-NEXT:    [[TMP3:%.*]] = bitcast i8** [[DOTADDR]] to i16**
+// CHECK1-NEXT:    [[TMP4:%.*]] = load i16*, i16** [[TMP3]], align 8
+// CHECK1-NEXT:    [[TMP5:%.*]] = call i8* @__kmpc_threadprivate_cached(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]], i8* bitcast (i64* @{{reduction_size[.].+[.]}})
+// CHECK1-NEXT:    [[TMP6:%.*]] = bitcast i8* [[TMP5]] to i64*
+// CHECK1-NEXT:    [[TMP7:%.*]] = load i64, i64* [[TMP6]], align 8
+// CHECK1-NEXT:    [[TMP8:%.*]] = getelementptr i16, i16* [[TMP4]], i64 [[TMP7]]
+// CHECK1-NEXT:    [[OMP_ARRAYINIT_ISEMPTY:%.*]] = icmp eq i16* [[TMP4]], [[TMP8]]
+// CHECK1-NEXT:    br i1 [[OMP_ARRAYINIT_ISEMPTY]], label [[OMP_ARRAYINIT_DONE:%.*]], label [[OMP_ARRAYINIT_BODY:%.*]]
+// CHECK1:       omp.arrayinit.body:
+// CHECK1-NEXT:    [[OMP_ARRAYCPY_DESTELEMENTPAST:%.*]] = phi i16* [ [[TMP4]], [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_DEST_ELEMENT:%.*]], [[OMP_ARRAYINIT_BODY]] ]
+// CHECK1-NEXT:    store i16 0, i16* [[OMP_ARRAYCPY_DESTELEMENTPAST]], align 2
+// CHECK1-NEXT:    [[OMP_ARRAYCPY_DEST_ELEMENT]] = getelementptr i16, i16* [[OMP_ARRAYCPY_DESTELEMENTPAST]], i32 1
+// CHECK1-NEXT:    [[OMP_ARRAYCPY_DONE:%.*]] = icmp eq i16* [[OMP_ARRAYCPY_DEST_ELEMENT]], [[TMP8]]
+// CHECK1-NEXT:    br i1 [[OMP_ARRAYCPY_DONE]], label [[OMP_ARRAYINIT_DONE]], label [[OMP_ARRAYINIT_BODY]]
+// CHECK1:       omp.arrayinit.done:
+// CHECK1-NEXT:    ret void
+//
+//
+// CHECK1-LABEL: define {{[^@]+}}@.red_comb..8
+// CHECK1-SAME: (i8* [[TMP0:%.*]], i8* [[TMP1:%.*]]) #[[ATTR5]] {
+// CHECK1-NEXT:  entry:
+// CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
+// CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca i8*, align 8
+// CHECK1-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
+// CHECK1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
+// CHECK1-NEXT:    store i8* [[TMP1]], i8** [[DOTADDR1]], align 8
+// CHECK1-NEXT:    [[TMP3:%.*]] = call i8* @__kmpc_threadprivate_cached(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]], i8* bitcast (i64* @{{reduction_size[.].+[.]}})
+// CHECK1-NEXT:    [[TMP4:%.*]] = bitcast i8* [[TMP3]] to i64*
+// CHECK1-NEXT:    [[TMP5:%.*]] = load i64, i64* [[TMP4]], align 8
+// CHECK1-NEXT:    [[TMP6:%.*]] = bitcast i8** [[DOTADDR]] to i16**
+// CHECK1-NEXT:    [[TMP7:%.*]] = load i16*, i16** [[TMP6]], align 8
+// CHECK1-NEXT:    [[TMP8:%.*]] = bitcast i8** [[DOTADDR1]] to i16**
+// CHECK1-NEXT:    [[TMP9:%.*]] = load i16*, i16** [[TMP8]], align 8
+// CHECK1-NEXT:    [[TMP10:%.*]] = getelementptr i16, i16* [[TMP7]], i64 [[TMP5]]
+// CHECK1-NEXT:    [[OMP_ARRAYCPY_ISEMPTY:%.*]] = icmp eq i16* [[TMP7]], [[TMP10]]
+// CHECK1-NEXT:    br i1 [[OMP_ARRAYCPY_ISEMPTY]], label [[OMP_ARRAYCPY_DONE4:%.*]], label [[OMP_ARRAYCPY_BODY:%.*]]
+// CHECK1:       omp.arraycpy.body:
+// CHECK1-NEXT:    [[OMP_ARRAYCPY_SRCELEMENTPAST:%.*]] = phi i16* [ [[TMP9]], [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_SRC_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]
+// CHECK1-NEXT:    [[OMP_ARRAYCPY_DESTELEMENTPAST:%.*]] = phi i16* [ [[TMP7]], [[ENTRY]] ], [ [[OMP_ARRAYCPY_DEST_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]
+// CHECK1-NEXT:    [[TMP11:%.*]] = load i16, i16* [[OMP_ARRAYCPY_DESTELEMENTPAST]], align 2
+// CHECK1-NEXT:    [[CONV:%.*]] = sext i16 [[TMP11]] to i32
+// CHECK1-NEXT:    [[TMP12:%.*]] = load i16, i16* [[OMP_ARRAYCPY_SRCELEMENTPAST]], align 2
+// CHECK1-NEXT:    [[CONV2:%.*]] = sext i16 [[TMP12]] to i32
+// CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 [[CONV]], [[CONV2]]
+// CHECK1-NEXT:    [[CONV3:%.*]] = trunc i32 [[ADD]] to i16
+// CHECK1-NEXT:    store i16 [[CONV3]], i16* [[OMP_ARRAYCPY_DESTELEMENTPAST]], align 2
+// CHECK1-NEXT:    [[OMP_ARRAYCPY_DEST_ELEMENT]] = getelementptr i16, i16* [[OMP_ARRAYCPY_DESTELEMENTPAST]], i32 1
+// CHECK1-NEXT:    [[OMP_ARRAYCPY_SRC_ELEMENT]] = getelementptr i16, i16* [[OMP_ARRAYCPY_SRCELEMENTPAST]], i32 1
+// CHECK1-NEXT:    [[OMP_ARRAYCPY_DONE:%.*]] = icmp eq i16* [[OMP_ARRAYCPY_DEST_ELEMENT]], [[TMP10]]
+// CHECK1-NEXT:    br i1 [[OMP_ARRAYCPY_DONE]], label [[OMP_ARRAYCPY_DONE4]], label [[OMP_ARRAYCPY_BODY]]
+// CHECK1:       omp.arraycpy.done4:
+// CHECK1-NEXT:    ret void
+//
+//
+// CHECK1-LABEL: define {{[^@]+}}@.omp_outlined.
+// CHECK1-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32* nonnull align 4 dereferenceable(4) [[A:%.*]], i64 [[VLA:%.*]], i16* nonnull align 2 dereferenceable(2) [[D:%.*]], i8** nonnull align 8 dereferenceable(8) [[DOTTASK_RED_:%.*]]) #[[ATTR8:[0-9]+]] {
+// CHECK1-NEXT:  entry:
+// CHECK1-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+// CHECK1-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+// CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca i32*, align 8
+// CHECK1-NEXT:    [[VLA_ADDR:%.*]] = alloca i64, align 8
+// CHECK1-NEXT:    [[D_ADDR:%.*]] = alloca i16*, align 8
+// CHECK1-NEXT:    [[DOTTASK_RED__ADDR:%.*]] = alloca i8**, align 8
+// CHECK1-NEXT:    [[AGG_CAPTURED:%.*]] = alloca [[STRUCT_ANON:%.*]], align 8
+// CHECK1-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
+// CHECK1-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
+// CHECK1-NEXT:    store i32* [[A]], i32** [[A_ADDR]], align 8
+// CHECK1-NEXT:    store i64 [[VLA]], i64* [[VLA_ADDR]], align 8
+// CHECK1-NEXT:    store i16* [[D]], i16** [[D_ADDR]], align 8
+// CHECK1-NEXT:    store i8** [[DOTTASK_RED_]], i8*** [[DOTTASK_RED__ADDR]], align 8
+// CHECK1-NEXT:    [[TMP0:%.*]] = load i32*, i32** [[A_ADDR]], align 8
+// CHECK1-NEXT:    [[TMP1:%.*]] = load i64, i64* [[VLA_ADDR]], align 8
+// CHECK1-NEXT:    [[TMP2:%.*]] = load i16*, i16** [[D_ADDR]], align 8
+// CHECK1-NEXT:    [[TMP3:%.*]] = load i8**, i8*** [[DOTTASK_RED__ADDR]], align 8
+// CHECK1-NEXT:    [[TMP4:%.*]] = load i8*, i8** [[TMP3]], align 8
+// CHECK1-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [[STRUCT_ANON]], %struct.anon* [[AGG_CAPTURED]], i32 0, i32 0
+// CHECK1-NEXT:    store i32* [[TMP0]], i32** [[TMP5]], align 8
+// CHECK1-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [[STRUCT_ANON]], %struct.anon* [[AGG_CAPTURED]], i32 0, i32 1
+// CHECK1-NEXT:    store i64 [[TMP1]], i64* [[TMP6]], align 8
+// CHECK1-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [[STRUCT_ANON]], %struct.anon* [[AGG_CAPTURED]], i32 0, i32 2
+// CHECK1-NEXT:    store i16* [[TMP2]], i16** [[TMP7]], align 8
+// CHECK1-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [[STRUCT_ANON]], %struct.anon* [[AGG_CAPTURED]], i32 0, i32 3
+// CHECK1-NEXT:    [[TMP9:%.*]] = load i8*, i8** [[TMP3]], align 8
+// CHECK1-NEXT:    store i8* [[TMP9]], i8** [[TMP8]], align 8
+// CHECK1-NEXT:    [[TMP10:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
+// CHECK1-NEXT:    [[TMP11:%.*]] = load i32, i32* [[TMP10]], align 4
+// CHECK1-NEXT:    [[TMP12:%.*]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* @[[GLOB1]], i32 [[TMP11]], i32 1, i64 48, i64 32, i32 (i32, i8*)* bitcast (i32 (i32, %struct.kmp_task_t_with_privates*)* @.omp_task_entry. to i32 (i32, i8*)*))
+// CHECK1-NEXT:    [[TMP13:%.*]] = bitcast i8* [[TMP12]] to %struct.kmp_task_t_with_privates*
+// CHECK1-NEXT:    [[TMP14:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T_WITH_PRIVATES:%.*]], %struct.kmp_task_t_with_privates* [[TMP13]], i32 0, i32 0
+// CHECK1-NEXT:    [[TMP15:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T:%.*]], %struct.kmp_task_t* [[TMP14]], i32 0, i32 0
+// CHECK1-NEXT:    [[TMP16:%.*]] = load i8*, i8** [[TMP15]], align 8
+// CHECK1-NEXT:    [[TMP17:%.*]] = bitcast %struct.anon* [[AGG_CAPTURED]] to i8*
+// CHECK1-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 8 [[TMP16]], i8* align 8 [[TMP17]], i64 32, i1 false)
+// CHECK1-NEXT:    [[TMP18:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T_WITH_PRIVATES]], %struct.kmp_task_t_with_privates* [[TMP13]], i32 0, i32 1
+// CHECK1-NEXT:    [[TMP19:%.*]] = bitcast i8* [[TMP16]] to %struct.anon*
+// CHECK1-NEXT:    [[TMP20:%.*]] = getelementptr inbounds [[STRUCT__KMP_PRIVATES_T:%.*]], %struct..kmp_privates.t* [[TMP18]], i32 0, i32 0
+// CHECK1-NEXT:    [[TMP21:%.*]] = load i8*, i8** [[TMP3]], align 8
+// CHECK1-NEXT:    store i8* [[TMP21]], i8** [[TMP20]], align 8
+// CHECK1-NEXT:    call void @__kmpc_omp_task_begin_if0(%struct.ident_t* @[[GLOB1]], i32 [[TMP11]], i8* [[TMP12]])
+// CHECK1-NEXT:    [[TMP22:%.*]] = call i32 @.omp_task_entry.(i32 [[TMP11]], %struct.kmp_task_t_with_privates* [[TMP13]]) #[[ATTR3]]
+// CHECK1-NEXT:    call void @__kmpc_omp_task_complete_if0(%struct.ident_t* @[[GLOB1]], i32 [[TMP11]], i8* [[TMP12]])
+// CHECK1-NEXT:    ret void
+//
+//
+// CHECK1-LABEL: define {{[^@]+}}@__omp_offloading_{{.*}}_main_l{{[0-9]+}}
+// CHECK1-SAME: (i32* nonnull align 4 dereferenceable(4) [[A:%.*]], i64 [[VLA:%.*]], i16* nonnull align 2 dereferenceable(2) [[D:%.*]], i8* [[DOTTASK_RED_:%.*]]) #[[ATTR9:[0-9]+]] {
+// CHECK1-NEXT:  entry:
+// CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca i32*, align 8
+// CHECK1-NEXT:    [[VLA_ADDR:%.*]] = alloca i64, align 8
+// CHECK1-NEXT:    [[D_ADDR:%.*]] = alloca i16*, align 8
+// CHECK1-NEXT:    [[DOTTASK_RED__ADDR:%.*]] = alloca i8*, align 8
+// CHECK1-NEXT:    [[I:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    store i32* [[A]], i32** [[A_ADDR]], align 8
+// CHECK1-NEXT:    store i64 [[VLA]], i64* [[VLA_ADDR]], align 8
+// CHECK1-NEXT:    store i16* [[D]], i16** [[D_ADDR]], align 8
+// CHECK1-NEXT:    store i8* [[DOTTASK_RED_]], i8** [[DOTTASK_RED__ADDR]], align 8
+// CHECK1-NEXT:    [[TMP0:%.*]] = load i32*, i32** [[A_ADDR]], align 8
+// CHECK1-NEXT:    [[TMP1:%.*]] = load i64, i64* [[VLA_ADDR]], align 8
+// CHECK1-NEXT:    [[TMP2:%.*]] = load i16*, i16** [[D_ADDR]], align 8
+// CHECK1-NEXT:    store i32 0, i32* [[I]], align 4
+// CHECK1-NEXT:    br label [[FOR_COND:%.*]]
+// CHECK1:       for.cond:
+// CHECK1-NEXT:    [[TMP3:%.*]] = load i32, i32* [[I]], align 4
+// CHECK1-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP3]], 5
+// CHECK1-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
+// CHECK1:       for.body:
+// CHECK1-NEXT:    [[TMP4:%.*]] = load i32, i32* [[TMP0]], align 4
+// CHECK1-NEXT:    [[IDXPROM_I:%.*]] = sext i32 [[TMP4]] to i64
+// CHECK1-NEXT:    [[ARRAYIDX_I:%.*]] = getelementptr inbounds i16, i16* [[TMP2]], i64 [[IDXPROM_I]]
+// CHECK1-NEXT:    [[TMP5:%.*]] = load i16, i16* [[ARRAYIDX_I]], align 2
+// CHECK1-NEXT:    [[CONV:%.*]] = sext i16 [[TMP5]] to i32
+// CHECK1-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP0]], align 4
+// CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP6]], [[CONV]]
+// CHECK1-NEXT:    store i32 [[ADD]], i32* [[TMP0]], align 4
+// CHECK1-NEXT:    br label [[FOR_INC:%.*]]
+// CHECK1:       for.inc:
+// CHECK1-NEXT:    [[TMP7:%.*]] = load i32, i32* [[I]], align 4
+// CHECK1-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP7]], 1
+// CHECK1-NEXT:    store i32 [[INC]], i32* [[I]], align 4
+// CHECK1-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP3:![0-9]+]]
+// CHECK1:       for.end
+// CHECK1-NEXT:    ret void
+//
+//
+// CHECK1-LABEL: define {{[^@]+}}@.omp_task_privates_map.
+// CHECK1-SAME: (%struct..kmp_privates.t* noalias [[TMP0:%.*]], i8*** noalias [[TMP1:%.*]]) #[[ATTR9:[0-9]+]] {
+// CHECK1-NEXT:  entry:
+// CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca %struct..kmp_privates.t*, align 8
+// CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca i8***, align 8
+// CHECK1-NEXT:    store %struct..kmp_privates.t* [[TMP0]], %struct..kmp_privates.t** [[DOTADDR]], align 8
+// CHECK1-NEXT:    store i8*** [[TMP1]], i8**** [[DOTADDR1]], align 8
+// CHECK1-NEXT:    [[TMP2:%.*]] = load %struct..kmp_privates.t*, %struct..kmp_privates.t** [[DOTADDR]], align 8
+// CHECK1-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [[STRUCT__KMP_PRIVATES_T:%.*]], %struct..kmp_privates.t* [[TMP2]], i32 0, i32 0
+// CHECK1-NEXT:    [[TMP4:%.*]] = load i8***, i8**** [[DOTADDR1]], align 8
+// CHECK1-NEXT:    store i8** [[TMP3]], i8*** [[TMP4]], align 8
+// CHECK1-NEXT:    ret void
+//
+//
+// CHECK1-LABEL: define {{[^@]+}}@.omp_task_entry.
+// CHECK1-SAME: (i32 [[TMP0:%.*]], %struct.kmp_task_t_with_privates* noalias [[TMP1:%.*]]) #[[ATTR5]] {
+// CHECK1-NEXT:  entry:
+// CHECK1-NEXT:    [[DOTGLOBAL_TID__ADDR_I:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[DOTPART_ID__ADDR_I:%.*]] = alloca i32*, align 8
+// CHECK1-NEXT:    [[DOTPRIVATES__ADDR_I:%.*]] = alloca i8*, align 8
+// CHECK1-NEXT:    [[DOTCOPY_FN__ADDR_I:%.*]] = alloca void (i8*, ...)*, align 8
+// CHECK1-NEXT:    [[DOTTASK_T__ADDR_I:%.*]] = alloca i8*, align 8
+// CHECK1-NEXT:    [[__CONTEXT_ADDR_I:%.*]] = alloca %struct.anon*, align 8
+// CHECK1-NEXT:    [[DOTFIRSTPRIV_PTR_ADDR_I:%.*]] = alloca i8**, align 8
+// CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i32, align 4
+// CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca %struct.kmp_task_t_with_privates*, align 8
+// CHECK1-NEXT:    store i32 [[TMP0]], i32* [[DOTADDR]], align 4
+// CHECK1-NEXT:    store %struct.kmp_task_t_with_privates* [[TMP1]], %struct.kmp_task_t_with_privates** [[DOTADDR1]], align 8
+// CHECK1-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTADDR]], align 4
+// CHECK1-NEXT:    [[TMP3:%.*]] = load %struct.kmp_task_t_with_privates*, %struct.kmp_task_t_with_privates** [[DOTADDR1]], align 8
+// CHECK1-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T_WITH_PRIVATES:%.*]], %struct.kmp_task_t_with_privates* [[TMP3]], i32 0, i32 0
+// CHECK1-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T:%.*]], %struct.kmp_task_t* [[TMP4]], i32 0, i32 2
+// CHECK1-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T]], %struct.kmp_task_t* [[TMP4]], i32 0, i32 0
+// CHECK1-NEXT:    [[TMP7:%.*]] = load i8*, i8** [[TMP6]], align 8
+// CHECK1-NEXT:    [[TMP8:%.*]] = bitcast i8* [[TMP7]] to %struct.anon*
+// CHECK1-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T_WITH_PRIVATES]], %struct.kmp_task_t_with_privates* [[TMP3]], i32 0, i32 1
+// CHECK1-NEXT:    [[TMP10:%.*]] = bitcast %struct..kmp_privates.t* [[TMP9]] to i8*
+// CHECK1-NEXT:    [[TMP11:%.*]] = bitcast %struct.kmp_task_t_with_privates* [[TMP3]] to i8*
+// CHECK1-NEXT:    call void @llvm.experimental.noalias.scope.decl(metadata [[META3:![0-9]+]])
+// CHECK1-NEXT:    call void @llvm.experimental.noalias.scope.decl(metadata [[META8:![0-9]+]])
+// CHECK1-NEXT:    call void @llvm.experimental.noalias.scope.decl(metadata [[META10:![0-9]+]])
+// CHECK1-NEXT:    call void @llvm.experimental.noalias.scope.decl(metadata [[META12:![0-9]+]])
+// CHECK1-NEXT:    store i32 [[TMP2]], i32* [[DOTGLOBAL_TID__ADDR_I]], align 4, !noalias !14
+// CHECK1-NEXT:    store i32* [[TMP5]], i32** [[DOTPART_ID__ADDR_I]], align 8, !noalias !14
+// CHECK1-NEXT:    store i8* [[TMP10]], i8** [[DOTPRIVATES__ADDR_I]], align 8, !noalias !14
+// CHECK1-NEXT:    store void (i8*, ...)* bitcast (void (%struct..kmp_privates.t*, i8***)* @.omp_task_privates_map. to void (i8*, ...)*), void (i8*, ...)** [[DOTCOPY_FN__ADDR_I]], align 8, !noalias !14
+// CHECK1-NEXT:    store i8* [[TMP11]], i8** [[DOTTASK_T__ADDR_I]], align 8, !noalias !14
+// CHECK1-NEXT:    store %struct.anon* [[TMP8]], %struct.anon** [[__CONTEXT_ADDR_I]], align 8, !noalias !14
+// CHECK1-NEXT:    [[TMP12:%.*]] = load %struct.anon*, %struct.anon** [[__CONTEXT_ADDR_I]], align 8, !noalias !14
+// CHECK1-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [[STRUCT_ANON:%.*]], %struct.anon* [[TMP12]], i32 0, i32 1
+// CHECK1-NEXT:    [[TMP14:%.*]] = load i64, i64* [[TMP13]], align 8
+// CHECK1-NEXT:    [[TMP15:%.*]] = load void (i8*, ...)*, void (i8*, ...)** [[DOTCOPY_FN__ADDR_I]], align 8, !noalias !14
+// CHECK1-NEXT:    [[TMP16:%.*]] = load i8*, i8** [[DOTPRIVATES__ADDR_I]], align 8, !noalias !14
+// CHECK1-NEXT:    [[TMP17:%.*]] = bitcast void (i8*, ...)* [[TMP15]] to void (i8*, i8***)*
+// CHECK1-NEXT:    call void [[TMP17]](i8* [[TMP16]], i8*** [[DOTFIRSTPRIV_PTR_ADDR_I]]) #[[ATTR3]]
+// CHECK1-NEXT:    [[TMP18:%.*]] = load i8**, i8*** [[DOTFIRSTPRIV_PTR_ADDR_I]], align 8, !noalias !14
+// CHECK1-NEXT:    [[TMP19:%.*]] = getelementptr inbounds [[STRUCT_ANON]], %struct.anon* [[TMP12]], i32 0, i32 0
+// CHECK1-NEXT:    [[TMP20:%.*]] = load i32*, i32** [[TMP19]], align 8
+// CHECK1-NEXT:    [[TMP21:%.*]] = load i8*, i8** [[TMP18]], align 8
+// CHECK1-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTGLOBAL_TID__ADDR_I]], align 4, !noalias !14
+// CHECK1-NEXT:    [[TMP23:%.*]] = bitcast i32* [[TMP20]] to i8*
+// CHECK1-NEXT:    [[TMP24:%.*]] = call i8* @__kmpc_task_reduction_get_th_data(i32 [[TMP22]], i8* [[TMP21]], i8* [[TMP23]]) #[[ATTR3]]
+// CHECK1-NEXT:    [[CONV_I:%.*]] = bitcast i8* [[TMP24]] to i32*
+// CHECK1-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [[STRUCT_ANON]], %struct.anon* [[TMP12]], i32 0, i32 0
+// CHECK1-NEXT:    [[TMP26:%.*]] = load i32*, i32** [[TMP25]], align 8
+// CHECK1-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [[STRUCT_ANON]], %struct.anon* [[TMP12]], i32 0, i32 2
+// CHECK1-NEXT:    [[TMP28:%.*]] = load i16*, i16** [[TMP27]], align 8
+// CHECK1-NEXT:    [[TMP29:%.*]] = load i8*, i8** [[TMP18]], align 8
+// CHECK1-NEXT:    call void @__omp_offloading_{{.*}}_main_l{{[0-9]+}}(i32* [[TMP26]], i64 [[TMP14]], i16* [[TMP28]], i8* [[TMP29]]) #[[ATTR3]]
+// CHECK1-NEXT:    ret i32 0
+//
+//
+// CHECK1-LABEL: define {{[^@]+}}@_ZN1SC2Ev
+// CHECK1-SAME: (%struct.S* nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] align 2 {
+// CHECK1-NEXT:  entry:
+// CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca %struct.S*, align 8
+// CHECK1-NEXT:    store %struct.S* [[THIS]], %struct.S** [[THIS_ADDR]], align 8
+// CHECK1-NEXT:    [[THIS1:%.*]] = load %struct.S*, %struct.S** [[THIS_ADDR]], align 8
+// CHECK1-NEXT:    [[A:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[THIS1]], i32 0, i32 0
+// CHECK1-NEXT:    store i32 0, i32* [[A]], align 4
+// CHECK1-NEXT:    ret void
+//
+//
+// CHECK1-LABEL: define {{[^@]+}}@_ZN1SD2Ev
+// CHECK1-SAME: (%struct.S* nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] align 2 {
+// CHECK1-NEXT:  entry:
+// CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca %struct.S*, align 8
+// CHECK1-NEXT:    store %struct.S* [[THIS]], %struct.S** [[THIS_ADDR]], align 8
+// CHECK1-NEXT:    [[THIS1:%.*]] = load %struct.S*, %struct.S** [[THIS_ADDR]], align 8
+// CHECK1-NEXT:    ret void
+//
+//
+// CHECK1-LABEL: define {{[^@]+}}@_ZN1SC1ERKS_
+// CHECK1-SAME: (%struct.S* nonnull align 4 dereferenceable(4) [[THIS:%.*]], %struct.S* nonnull align 4 dereferenceable(4) [[TMP0:%.*]]) unnamed_addr #[[ATTR1]] align 2 {
+// CHECK1-NEXT:  entry:
+// CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca %struct.S*, align 8
+// CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca %struct.S*, align 8
+// CHECK1-NEXT:    store %struct.S* [[THIS]], %struct.S** [[THIS_ADDR]], align 8
+// CHECK1-NEXT:    store %struct.S* [[TMP0]], %struct.S** [[DOTADDR]], align 8
+// CHECK1-NEXT:    [[THIS1:%.*]] = load %struct.S*, %struct.S** [[THIS_ADDR]], align 8
+// CHECK1-NEXT:    [[TMP1:%.*]] = load %struct.S*, %struct.S** [[DOTADDR]], align 8
+// CHECK1-NEXT:    call void @_ZN1SC2ERKS_(%struct.S* nonnull align 4 dereferenceable(4) [[THIS1]], %struct.S* nonnull align 4 dereferenceable(4) [[TMP1]])
+// CHECK1-NEXT:    ret void
+//
+//
+// CHECK1-LABEL: define {{[^@]+}}@_ZN1SC2ERKS_
+// CHECK1-SAME: (%struct.S* nonnull align 4 dereferenceable(4) [[THIS:%.*]], %struct.S* nonnull align 4 dereferenceable(4) [[TMP0:%.*]]) unnamed_addr #[[ATTR1]] align 2 {
+// CHECK1-NEXT:  entry:
+// CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca %struct.S*, align 8
+// CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca %struct.S*, align 8
+// CHECK1-NEXT:    store %struct.S* [[THIS]], %struct.S** [[THIS_ADDR]], align 8
+// CHECK1-NEXT:    store %struct.S* [[TMP0]], %struct.S** [[DOTADDR]], align 8
+// CHECK1-NEXT:    [[THIS1:%.*]] = load %struct.S*, %struct.S** [[THIS_ADDR]], align 8
+// CHECK1-NEXT:    ret void
+//

llvm/include/llvm/Frontend/OpenMP/OMP.td

let allowedClauses = [
     VersionedClause<OMPC_If>,
     VersionedClause<OMPC_Map>,
     VersionedClause<OMPC_Private>,
     VersionedClause<OMPC_Depend>,
     VersionedClause<OMPC_FirstPrivate>,
     VersionedClause<OMPC_IsDevicePtr>,
     VersionedClause<OMPC_HasDeviceAddr, 51>,
     VersionedClause<OMPC_Reduction>,
+    VersionedClause<OMPC_InReduction, 50>,
     VersionedClause<OMPC_Allocate>,
     VersionedClause<OMPC_UsesAllocators, 50>
   ];
   let allowedOnceClauses = [
     VersionedClause<OMPC_Device>,
     VersionedClause<OMPC_DefaultMap>,
     VersionedClause<OMPC_NoWait>
   ];