Reland "[codegen] Store address of indirect arguments on the stack"

The commit was reverted due to a regression in debug information of an
optimized code test in lldb. This has since been addressed by:

1. rGf753e5be8239: [LiveDebugValues] Allow EntryValue with OP_deref
expressions
2. rG055f2f04e658: [mem2reg][debuginfo] Handle op_deref when converting
dbg.declare

Differential Revision: https://reviews.llvm.org/D141381

24 files changed, 278 insertions, 16 deletions

diff --git a/clang/docs/ReleaseNotes.rst b/clang/docs/ReleaseNotes.rst
index e5d09ce..9d9c5c6 100644
--- a/clang/docs/ReleaseNotes.rst
+++ b/clang/docs/ReleaseNotes.rst
@@ -72,6 +72,9 @@ Improvements to Clang's diagnostics
 
 Non-comprehensive list of changes in this release
 -------------------------------------------------
+- Clang now saves the address of ABI-indirect function parameters on the stack,
+  improving the debug information available in programs compiled without
+  optimizations.
 
 New Compiler Flags
 ------------------
diff --git a/clang/lib/CodeGen/CGDebugInfo.cpp b/clang/lib/CodeGen/CGDebugInfo.cpp
index fafc0f8..bb91b51 100644
--- a/clang/lib/CodeGen/CGDebugInfo.cpp
+++ b/clang/lib/CodeGen/CGDebugInfo.cpp
@@ -4840,9 +4840,10 @@ void CGDebugInfo::EmitDeclareOfBlockDeclRefVariable(
 
 llvm::DILocalVariable *
 CGDebugInfo::EmitDeclareOfArgVariable(const VarDecl *VD, llvm::Value *AI,
-                                      unsigned ArgNo, CGBuilderTy &Builder) {
+                                      unsigned ArgNo, CGBuilderTy &Builder,
+                                      bool UsePointerValue) {
   assert(CGM.getCodeGenOpts().hasReducedDebugInfo());
-  return EmitDeclare(VD, AI, ArgNo, Builder);
+  return EmitDeclare(VD, AI, ArgNo, Builder, UsePointerValue);
 }
 
 namespace {
diff --git a/clang/lib/CodeGen/CGDebugInfo.h b/clang/lib/CodeGen/CGDebugInfo.h
index d807fb7..a70b72e 100644
--- a/clang/lib/CodeGen/CGDebugInfo.h
+++ b/clang/lib/CodeGen/CGDebugInfo.h
@@ -489,10 +489,9 @@ public:
 
   /// Emit call to \c llvm.dbg.declare for an argument variable
   /// declaration.
-  llvm::DILocalVariable *EmitDeclareOfArgVariable(const VarDecl *Decl,
-                                                  llvm::Value *AI,
-                                                  unsigned ArgNo,
-                                                  CGBuilderTy &Builder);
+  llvm::DILocalVariable *
+  EmitDeclareOfArgVariable(const VarDecl *Decl, llvm::Value *AI, unsigned ArgNo,
+                           CGBuilderTy &Builder, bool UsePointerValue = false);
 
   /// Emit call to \c llvm.dbg.declare for the block-literal argument
   /// to a block invocation function.
diff --git a/clang/lib/CodeGen/CGDecl.cpp b/clang/lib/CodeGen/CGDecl.cpp
index ceaddc4..a70997f 100644
--- a/clang/lib/CodeGen/CGDecl.cpp
+++ b/clang/lib/CodeGen/CGDecl.cpp
@@ -2476,6 +2476,8 @@ void CodeGenFunction::EmitParmDecl(const VarDecl &D, ParamValue Arg,
   Address AllocaPtr = Address::invalid();
   bool DoStore = false;
   bool IsScalar = hasScalarEvaluationKind(Ty);
+  bool UseIndirectDebugAddress = false;
+
   // If we already have a pointer to the argument, reuse the input pointer.
   if (Arg.isIndirect()) {
     // If we have a prettier pointer type at this point, bitcast to that.
@@ -2487,6 +2489,19 @@ void CodeGenFunction::EmitParmDecl(const VarDecl &D, ParamValue Arg,
     auto AllocaAS = CGM.getASTAllocaAddressSpace();
     auto *V = DeclPtr.getPointer();
     AllocaPtr = DeclPtr;
+
+    // For truly ABI indirect arguments -- those that are not `byval` -- store
+    // the address of the argument on the stack to preserve debug information.
+    ABIArgInfo ArgInfo = CurFnInfo->arguments()[ArgNo - 1].info;
+    if (ArgInfo.isIndirect())
+      UseIndirectDebugAddress = !ArgInfo.getIndirectByVal();
+    if (UseIndirectDebugAddress) {
+      auto PtrTy = getContext().getPointerType(Ty);
+      AllocaPtr = CreateMemTemp(PtrTy, getContext().getTypeAlignInChars(PtrTy),
+                                D.getName() + ".indirect_addr");
+      EmitStoreOfScalar(V, AllocaPtr, /* Volatile */ false, PtrTy);
+    }
+
     auto SrcLangAS = getLangOpts().OpenCL ? LangAS::opencl_private : AllocaAS;
     auto DestLangAS =
         getLangOpts().OpenCL ? LangAS::opencl_private : LangAS::Default;
@@ -2603,7 +2618,7 @@ void CodeGenFunction::EmitParmDecl(const VarDecl &D, ParamValue Arg,
     if (CGM.getCodeGenOpts().hasReducedDebugInfo() && !CurFuncIsThunk &&
         !NoDebugInfo) {
       llvm::DILocalVariable *DILocalVar = DI->EmitDeclareOfArgVariable(
-          &D, AllocaPtr.getPointer(), ArgNo, Builder);
+          &D, AllocaPtr.getPointer(), ArgNo, Builder, UseIndirectDebugAddress);
       if (const auto *Var = dyn_cast_or_null<ParmVarDecl>(&D))
         DI->getParamDbgMappings().insert({Var, DILocalVar});
     }
diff --git a/clang/test/CodeGen/aarch64-ls64.c b/clang/test/CodeGen/aarch64-ls64.c
index 341d54d..39def71 100644
--- a/clang/test/CodeGen/aarch64-ls64.c
+++ b/clang/test/CodeGen/aarch64-ls64.c
@@ -94,10 +94,12 @@ EXTERN_C void test_ld64b(void)
 // CHECK-C-LABEL: define {{[^@]+}}@test_st64b(
 // CHECK-C-NEXT:  entry:
 // CHECK-C-NEXT:    [[__ADDR_ADDR_I:%.*]] = alloca ptr, align 8
+// CHECK-C-NEXT:    [[VALUE_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK-C-NEXT:    [[BYVAL_TEMP:%.*]] = alloca [[STRUCT_DATA512_T:%.*]], align 8
 // CHECK-C-NEXT:    [[TMP0:%.*]] = load ptr, ptr @addr, align 8
 // CHECK-C-NEXT:    call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[BYVAL_TEMP]], ptr align 8 @val, i64 64, i1 false)
 // CHECK-C-NEXT:    store ptr [[TMP0]], ptr [[__ADDR_ADDR_I]], align 8
+// CHECK-C-NEXT:    store ptr [[BYVAL_TEMP]], ptr [[VALUE_INDIRECT_ADDR]], align 8
 // CHECK-C-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[__ADDR_ADDR_I]], align 8
 // CHECK-C-NEXT:    [[TMP2:%.*]] = load i64, ptr [[BYVAL_TEMP]], align 8
 // CHECK-C-NEXT:    [[TMP3:%.*]] = getelementptr i64, ptr [[BYVAL_TEMP]], i32 1
@@ -120,10 +122,12 @@ EXTERN_C void test_ld64b(void)
 // CHECK-CXX-LABEL: define {{[^@]+}}@test_st64b(
 // CHECK-CXX-NEXT:  entry:
 // CHECK-CXX-NEXT:    [[__ADDR_ADDR_I:%.*]] = alloca ptr, align 8
+// CHECK-CXX-NEXT:    [[VALUE_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK-CXX-NEXT:    [[AGG_TMP:%.*]] = alloca [[STRUCT_DATA512_T:%.*]], align 8
 // CHECK-CXX-NEXT:    [[TMP0:%.*]] = load ptr, ptr @addr, align 8
 // CHECK-CXX-NEXT:    call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[AGG_TMP]], ptr align 8 @val, i64 64, i1 false)
 // CHECK-CXX-NEXT:    store ptr [[TMP0]], ptr [[__ADDR_ADDR_I]], align 8
+// CHECK-CXX-NEXT:    store ptr [[AGG_TMP]], ptr [[VALUE_INDIRECT_ADDR]], align 8
 // CHECK-CXX-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[__ADDR_ADDR_I]], align 8
 // CHECK-CXX-NEXT:    [[TMP2:%.*]] = load i64, ptr [[AGG_TMP]], align 8
 // CHECK-CXX-NEXT:    [[TMP3:%.*]] = getelementptr i64, ptr [[AGG_TMP]], i32 1
@@ -151,10 +155,12 @@ EXTERN_C void test_st64b(void)
 // CHECK-C-LABEL: define {{[^@]+}}@test_st64bv(
 // CHECK-C-NEXT:  entry:
 // CHECK-C-NEXT:    [[__ADDR_ADDR_I:%.*]] = alloca ptr, align 8
+// CHECK-C-NEXT:    [[VALUE_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK-C-NEXT:    [[BYVAL_TEMP:%.*]] = alloca [[STRUCT_DATA512_T:%.*]], align 8
 // CHECK-C-NEXT:    [[TMP0:%.*]] = load ptr, ptr @addr, align 8
 // CHECK-C-NEXT:    call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[BYVAL_TEMP]], ptr align 8 @val, i64 64, i1 false)
 // CHECK-C-NEXT:    store ptr [[TMP0]], ptr [[__ADDR_ADDR_I]], align 8
+// CHECK-C-NEXT:    store ptr [[BYVAL_TEMP]], ptr [[VALUE_INDIRECT_ADDR]], align 8
 // CHECK-C-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[__ADDR_ADDR_I]], align 8
 // CHECK-C-NEXT:    [[TMP2:%.*]] = load i64, ptr [[BYVAL_TEMP]], align 8
 // CHECK-C-NEXT:    [[TMP3:%.*]] = getelementptr i64, ptr [[BYVAL_TEMP]], i32 1
@@ -178,10 +184,12 @@ EXTERN_C void test_st64b(void)
 // CHECK-CXX-LABEL: define {{[^@]+}}@test_st64bv(
 // CHECK-CXX-NEXT:  entry:
 // CHECK-CXX-NEXT:    [[__ADDR_ADDR_I:%.*]] = alloca ptr, align 8
+// CHECK-CXX-NEXT:    [[VALUE_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK-CXX-NEXT:    [[AGG_TMP:%.*]] = alloca [[STRUCT_DATA512_T:%.*]], align 8
 // CHECK-CXX-NEXT:    [[TMP0:%.*]] = load ptr, ptr @addr, align 8
 // CHECK-CXX-NEXT:    call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[AGG_TMP]], ptr align 8 @val, i64 64, i1 false)
 // CHECK-CXX-NEXT:    store ptr [[TMP0]], ptr [[__ADDR_ADDR_I]], align 8
+// CHECK-CXX-NEXT:    store ptr [[AGG_TMP]], ptr [[VALUE_INDIRECT_ADDR]], align 8
 // CHECK-CXX-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[__ADDR_ADDR_I]], align 8
 // CHECK-CXX-NEXT:    [[TMP2:%.*]] = load i64, ptr [[AGG_TMP]], align 8
 // CHECK-CXX-NEXT:    [[TMP3:%.*]] = getelementptr i64, ptr [[AGG_TMP]], i32 1
@@ -210,10 +218,12 @@ EXTERN_C void test_st64bv(void)
 // CHECK-C-LABEL: define {{[^@]+}}@test_st64bv0(
 // CHECK-C-NEXT:  entry:
 // CHECK-C-NEXT:    [[__ADDR_ADDR_I:%.*]] = alloca ptr, align 8
+// CHECK-C-NEXT:    [[VALUE_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK-C-NEXT:    [[BYVAL_TEMP:%.*]] = alloca [[STRUCT_DATA512_T:%.*]], align 8
 // CHECK-C-NEXT:    [[TMP0:%.*]] = load ptr, ptr @addr, align 8
 // CHECK-C-NEXT:    call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[BYVAL_TEMP]], ptr align 8 @val, i64 64, i1 false)
 // CHECK-C-NEXT:    store ptr [[TMP0]], ptr [[__ADDR_ADDR_I]], align 8
+// CHECK-C-NEXT:    store ptr [[BYVAL_TEMP]], ptr [[VALUE_INDIRECT_ADDR]], align 8
 // CHECK-C-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[__ADDR_ADDR_I]], align 8
 // CHECK-C-NEXT:    [[TMP2:%.*]] = load i64, ptr [[BYVAL_TEMP]], align 8
 // CHECK-C-NEXT:    [[TMP3:%.*]] = getelementptr i64, ptr [[BYVAL_TEMP]], i32 1
@@ -237,10 +247,12 @@ EXTERN_C void test_st64bv(void)
 // CHECK-CXX-LABEL: define {{[^@]+}}@test_st64bv0(
 // CHECK-CXX-NEXT:  entry:
 // CHECK-CXX-NEXT:    [[__ADDR_ADDR_I:%.*]] = alloca ptr, align 8
+// CHECK-CXX-NEXT:    [[VALUE_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK-CXX-NEXT:    [[AGG_TMP:%.*]] = alloca [[STRUCT_DATA512_T:%.*]], align 8
 // CHECK-CXX-NEXT:    [[TMP0:%.*]] = load ptr, ptr @addr, align 8
 // CHECK-CXX-NEXT:    call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[AGG_TMP]], ptr align 8 @val, i64 64, i1 false)
 // CHECK-CXX-NEXT:    store ptr [[TMP0]], ptr [[__ADDR_ADDR_I]], align 8
+// CHECK-CXX-NEXT:    store ptr [[AGG_TMP]], ptr [[VALUE_INDIRECT_ADDR]], align 8
 // CHECK-CXX-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[__ADDR_ADDR_I]], align 8
 // CHECK-CXX-NEXT:    [[TMP2:%.*]] = load i64, ptr [[AGG_TMP]], align 8
 // CHECK-CXX-NEXT:    [[TMP3:%.*]] = getelementptr i64, ptr [[AGG_TMP]], i32 1
diff --git a/clang/test/CodeGen/atomic-arm64.c b/clang/test/CodeGen/atomic-arm64.c
index 077d9ca..ab01f72 100644
--- a/clang/test/CodeGen/atomic-arm64.c
+++ b/clang/test/CodeGen/atomic-arm64.c
@@ -59,7 +59,10 @@ void test3(pointer_pair_t pair) {
 }
 
 // CHECK-LABEL:define{{.*}} void @test4(
-// CHECK:      [[TEMP:%.*]] = alloca [[QUAD_T:%.*]], align 8
+// CHECK-SAME: ptr noundef [[QUAD:%.*]])
+// CHECK:      [[QUAD_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
+// CHECK-NEXT: [[TEMP:%.*]] = alloca [[QUAD_T:%.*]], align 8
+// CHECK-NEXT: store ptr [[QUAD]], ptr [[QUAD_INDIRECT_ADDR]]
 // CHECK-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[TEMP]], ptr align 8 {{%.*}}, i64 32, i1 false)
 // CHECK-NEXT: call void @__atomic_store(i64 noundef 32, ptr noundef @a_pointer_quad, ptr noundef [[TEMP]], i32 noundef 5)
 void test4(pointer_quad_t quad) {
diff --git a/clang/test/CodeGenCXX/amdgcn-func-arg.cpp b/clang/test/CodeGenCXX/amdgcn-func-arg.cpp
index f6b75d5..67c4788 100644
--- a/clang/test/CodeGenCXX/amdgcn-func-arg.cpp
+++ b/clang/test/CodeGenCXX/amdgcn-func-arg.cpp
@@ -19,9 +19,13 @@ void func_with_ref_arg(A &a);
 void func_with_ref_arg(B &b);
 
 // CHECK-LABEL: @_Z22func_with_indirect_arg1A(
+// CHECK-SAME:  ptr addrspace(5) noundef [[ARG:%.*]])
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[INDIRECT_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
 // CHECK-NEXT:    [[P:%.*]] = alloca ptr, align 8, addrspace(5)
+// CHECK-NEXT:    [[INDIRECT_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[INDIRECT_ADDR]] to ptr
 // CHECK-NEXT:    [[P_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[P]] to ptr
+// CHECK-NEXT:    store ptr addrspace(5) [[ARG]], ptr [[INDIRECT_ADDR_ASCAST]]
 // CHECK-NEXT:    [[A_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[A:%.*]] to ptr
 // CHECK-NEXT:    store ptr [[A_ASCAST]], ptr [[P_ASCAST]], align 8
 // CHECK-NEXT:    ret void
diff --git a/clang/test/CodeGenCXX/debug-info.cpp b/clang/test/CodeGenCXX/debug-info.cpp
index 128b56e..89ab488 100644
--- a/clang/test/CodeGenCXX/debug-info.cpp
+++ b/clang/test/CodeGenCXX/debug-info.cpp
@@ -4,7 +4,13 @@
 // CHECK: @_ZN6pr96081xE ={{.*}} global ptr null, align 8, !dbg [[X:![0-9]+]]
 
 // CHECK: define{{.*}} void @_ZN7pr147634funcENS_3fooE
-// CHECK: call void @llvm.dbg.declare({{.*}}, metadata ![[F:[0-9]+]], metadata !DIExpression())
+// CHECK-SAME: ptr noundef [[param:%.*]])
+// CHECK-NEXT: entry:
+// CHECK-NEXT:   alloca ptr, align 8
+// CHECK-NEXT:   [[param_addr_storage:%.*]] = alloca ptr, align 8
+// CHECK-NEXT:   store
+// CHECK-NEXT:   store ptr [[param]], ptr [[param_addr_storage]], align 8
+// CHECK-NEXT:   call void @llvm.dbg.declare(metadata ptr [[param_addr_storage]], metadata ![[F:[0-9]+]], metadata !DIExpression(DW_OP_deref))
 
 // !llvm.dbg.cu pulls in globals and their types first.
 // CHECK-NOT: !DIGlobalVariable(name: "c"
diff --git a/clang/test/CodeGenCXX/derived-to-base-conv.cpp b/clang/test/CodeGenCXX/derived-to-base-conv.cpp
index 9025f66..8ba9802 100644
--- a/clang/test/CodeGenCXX/derived-to-base-conv.cpp
+++ b/clang/test/CodeGenCXX/derived-to-base-conv.cpp
@@ -32,7 +32,10 @@ void test0_helper(A);
 void test0(X x) {
   test0_helper(x);
   // CHECK-LABEL:    define{{.*}} void @_Z5test01X(
-  // CHECK:      [[TMP:%.*]] = alloca [[A:%.*]], align
+  // CHECK-SAME: ptr noundef [[ARG:%.*]])
+  // CHECK:      [[ARG_ADDR:%.*]] = alloca ptr
+  // CHECK-NEXT: [[TMP:%.*]] = alloca [[A:%.*]], align
+  // CHECK-NEXT: store ptr [[ARG]], ptr [[ARG_ADDR]]
   // CHECK-NEXT: [[T0:%.*]] = call noundef nonnull align {{[0-9]+}} dereferenceable({{[0-9]+}}) ptr @_ZN1XcvR1BEv(
   // CHECK-NEXT: call void @_ZN1AC1ERKS_(ptr {{[^,]*}} [[TMP]], ptr noundef nonnull align {{[0-9]+}} dereferenceable({{[0-9]+}}) [[T0]])
   // CHECK-NEXT: call void @_Z12test0_helper1A(ptr noundef [[TMP]])
diff --git a/clang/test/CodeGenCoroutines/coro-params-exp-namespace.cpp b/clang/test/CodeGenCoroutines/coro-params-exp-namespace.cpp
index 183ba3a..84c3e67 100644
--- a/clang/test/CodeGenCoroutines/coro-params-exp-namespace.cpp
+++ b/clang/test/CodeGenCoroutines/coro-params-exp-namespace.cpp
@@ -64,8 +64,8 @@ void consume(int, int, int) noexcept;
 // TODO: Add support for CopyOnly params
 // CHECK: define{{.*}} void @_Z1fi8MoveOnly11MoveAndCopy(i32 noundef %val, %struct.MoveOnly* noundef %[[MoParam:.+]], %struct.MoveAndCopy* noundef %[[McParam:.+]]) #0 personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*
 void f(int val, MoveOnly moParam, MoveAndCopy mcParam) {
-  // CHECK: %[[MoCopy:.+]] = alloca %struct.MoveOnly
-  // CHECK: %[[McCopy:.+]] = alloca %struct.MoveAndCopy
+  // CHECK: %[[MoCopy:.+]] = alloca %struct.MoveOnly,
+  // CHECK: %[[McCopy:.+]] = alloca %struct.MoveAndCopy,
   // CHECK: store i32 %val, i32* %[[ValAddr:.+]]
 
   // CHECK: call i8* @llvm.coro.begin(
@@ -110,7 +110,7 @@ void f(int val, MoveOnly moParam, MoveAndCopy mcParam) {
 // CHECK-LABEL: void @_Z16dependent_paramsI1A1BEvT_T0_S3_(%struct.A* noundef %x, %struct.B* noundef %0, %struct.B* noundef %y)
 template <typename T, typename U>
 void dependent_params(T x, U, U y) {
-  // CHECK: %[[x_copy:.+]] = alloca %struct.A
+  // CHECK: %[[x_copy:.+]] = alloca %struct.A,
   // CHECK-NEXT: %[[unnamed_copy:.+]] = alloca %struct.B
   // CHECK-NEXT: %[[y_copy:.+]] = alloca %struct.B
 
diff --git a/clang/test/CodeGenCoroutines/coro-params.cpp b/clang/test/CodeGenCoroutines/coro-params.cpp
index 86911fd..44753c6 100644
--- a/clang/test/CodeGenCoroutines/coro-params.cpp
+++ b/clang/test/CodeGenCoroutines/coro-params.cpp
@@ -64,8 +64,8 @@ void consume(int,int,int) noexcept;
 // TODO: Add support for CopyOnly params
 // CHECK: define{{.*}} void @_Z1fi8MoveOnly11MoveAndCopy(i32 noundef %val, %struct.MoveOnly* noundef %[[MoParam:.+]], %struct.MoveAndCopy* noundef %[[McParam:.+]]) #0 personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*
 void f(int val, MoveOnly moParam, MoveAndCopy mcParam) {
-  // CHECK: %[[MoCopy:.+]] = alloca %struct.MoveOnly
-  // CHECK: %[[McCopy:.+]] = alloca %struct.MoveAndCopy
+  // CHECK: %[[MoCopy:.+]] = alloca %struct.MoveOnly,
+  // CHECK: %[[McCopy:.+]] = alloca %struct.MoveAndCopy,
   // CHECK: store i32 %val, i32* %[[ValAddr:.+]]
 
   // CHECK: call i8* @llvm.coro.begin(
@@ -110,7 +110,7 @@ void f(int val, MoveOnly moParam, MoveAndCopy mcParam) {
 // CHECK-LABEL: void @_Z16dependent_paramsI1A1BEvT_T0_S3_(%struct.A* noundef %x, %struct.B* noundef %0, %struct.B* noundef %y)
 template <typename T, typename U>
 void dependent_params(T x, U, U y) {
-  // CHECK: %[[x_copy:.+]] = alloca %struct.A
+  // CHECK: %[[x_copy:.+]] = alloca %struct.A,
   // CHECK-NEXT: %[[unnamed_copy:.+]] = alloca %struct.B
   // CHECK-NEXT: %[[y_copy:.+]] = alloca %struct.B
 
diff --git a/clang/test/OpenMP/for_firstprivate_codegen.cpp b/clang/test/OpenMP/for_firstprivate_codegen.cpp
index 9428ef5..0fd8343 100644
--- a/clang/test/OpenMP/for_firstprivate_codegen.cpp
+++ b/clang/test/OpenMP/for_firstprivate_codegen.cpp
@@ -391,8 +391,10 @@ int main() {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
 // CHECK1-NEXT:    call void @_ZN1SIfEC2ERKS0_2St(ptr noundef nonnull align 4 dereferenceable(4) [[THIS1]], ptr noundef nonnull align 4 dereferenceable(4) [[TMP0]], ptr noundef [[T]])
@@ -462,8 +464,10 @@ int main() {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
@@ -649,8 +653,10 @@ int main() {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
 // CHECK1-NEXT:    call void @_ZN1SIiEC2ERKS0_2St(ptr noundef nonnull align 4 dereferenceable(4) [[THIS1]], ptr noundef nonnull align 4 dereferenceable(4) [[TMP0]], ptr noundef [[T]])
@@ -700,8 +706,10 @@ int main() {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S_0:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
diff --git a/clang/test/OpenMP/parallel_firstprivate_codegen.cpp b/clang/test/OpenMP/parallel_firstprivate_codegen.cpp
index eeed9e4..2c70e05 100644
--- a/clang/test/OpenMP/parallel_firstprivate_codegen.cpp
+++ b/clang/test/OpenMP/parallel_firstprivate_codegen.cpp
@@ -408,8 +408,10 @@ void array_func(float a[3], St s[2], int n, long double vla1[n]) {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
 // CHECK1-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 4
+// CHECK1-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 4
 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
 // CHECK1-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 4
+// CHECK1-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 4
 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 4
 // CHECK1-NEXT:    call void @_ZN1SIfEC2ERKS0_2St(ptr nonnull align 4 dereferenceable(4) [[THIS1]], ptr nonnull align 4 dereferenceable(4) [[TMP0]], ptr [[T]])
@@ -642,8 +644,10 @@ void array_func(float a[3], St s[2], int n, long double vla1[n]) {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
 // CHECK1-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 4
+// CHECK1-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 4
 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
 // CHECK1-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 4
+// CHECK1-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 4
 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
 // CHECK1-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 4
@@ -778,8 +782,10 @@ void array_func(float a[3], St s[2], int n, long double vla1[n]) {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
 // CHECK1-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 4
+// CHECK1-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 4
 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
 // CHECK1-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 4
+// CHECK1-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 4
 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 4
 // CHECK1-NEXT:    call void @_ZN1SIiEC2ERKS0_2St(ptr nonnull align 4 dereferenceable(4) [[THIS1]], ptr nonnull align 4 dereferenceable(4) [[TMP0]], ptr [[T]])
@@ -886,8 +892,10 @@ void array_func(float a[3], St s[2], int n, long double vla1[n]) {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
 // CHECK1-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 4
+// CHECK1-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 4
 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
 // CHECK1-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 4
+// CHECK1-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 4
 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
 // CHECK1-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S_0:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 4
@@ -1608,8 +1616,10 @@ void array_func(float a[3], St s[2], int n, long double vla1[n]) {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK9-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK9-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK9-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
 // CHECK9-NEXT:    call void @_ZN1SIfEC2ERKS0_2St(ptr nonnull align 4 dereferenceable(4) [[THIS1]], ptr nonnull align 4 dereferenceable(4) [[TMP0]], ptr [[T]])
@@ -1842,8 +1852,10 @@ void array_func(float a[3], St s[2], int n, long double vla1[n]) {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK9-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK9-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK9-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK9-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK9-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
@@ -1978,8 +1990,10 @@ void array_func(float a[3], St s[2], int n, long double vla1[n]) {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK9-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK9-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK9-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
 // CHECK9-NEXT:    call void @_ZN1SIiEC2ERKS0_2St(ptr nonnull align 4 dereferenceable(4) [[THIS1]], ptr nonnull align 4 dereferenceable(4) [[TMP0]], ptr [[T]])
@@ -2086,8 +2100,10 @@ void array_func(float a[3], St s[2], int n, long double vla1[n]) {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK9-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK9-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK9-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK9-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S_0:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK9-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
diff --git a/clang/test/OpenMP/sections_firstprivate_codegen.cpp b/clang/test/OpenMP/sections_firstprivate_codegen.cpp
index 0c2e507..15344fc 100644
--- a/clang/test/OpenMP/sections_firstprivate_codegen.cpp
+++ b/clang/test/OpenMP/sections_firstprivate_codegen.cpp
@@ -370,8 +370,10 @@ int main() {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
 // CHECK1-NEXT:    call void @_ZN1SIfEC2ERKS0_2St(ptr noundef nonnull align 4 dereferenceable(4) [[THIS1]], ptr noundef nonnull align 4 dereferenceable(4) [[TMP0]], ptr noundef [[T]])
@@ -441,8 +443,10 @@ int main() {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
@@ -606,8 +610,10 @@ int main() {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
 // CHECK1-NEXT:    call void @_ZN1SIiEC2ERKS0_2St(ptr noundef nonnull align 4 dereferenceable(4) [[THIS1]], ptr noundef nonnull align 4 dereferenceable(4) [[TMP0]], ptr noundef [[T]])
@@ -657,8 +663,10 @@ int main() {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S_0:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
diff --git a/clang/test/OpenMP/single_firstprivate_codegen.cpp b/clang/test/OpenMP/single_firstprivate_codegen.cpp
index 02d1d8d..9369899 100644
--- a/clang/test/OpenMP/single_firstprivate_codegen.cpp
+++ b/clang/test/OpenMP/single_firstprivate_codegen.cpp
@@ -325,8 +325,10 @@ int main() {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
 // CHECK1-NEXT:    call void @_ZN1SIfEC2ERKS0_2St(ptr nonnull align 4 dereferenceable(4) [[THIS1]], ptr nonnull align 4 dereferenceable(4) [[TMP0]], ptr [[T]])
@@ -396,8 +398,10 @@ int main() {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
@@ -523,8 +527,10 @@ int main() {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
 // CHECK1-NEXT:    call void @_ZN1SIiEC2ERKS0_2St(ptr nonnull align 4 dereferenceable(4) [[THIS1]], ptr nonnull align 4 dereferenceable(4) [[TMP0]], ptr [[T]])
@@ -574,8 +580,10 @@ int main() {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S_0:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
diff --git a/clang/test/OpenMP/target_teams_distribute_firstprivate_codegen.cpp b/clang/test/OpenMP/target_teams_distribute_firstprivate_codegen.cpp
index 93a869a..c19e2d4 100644
--- a/clang/test/OpenMP/target_teams_distribute_firstprivate_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_distribute_firstprivate_codegen.cpp
@@ -507,8 +507,10 @@ int main() {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
 // CHECK1-NEXT:    call void @_ZN1SIfEC2ERKS0_2St(ptr noundef nonnull align 4 dereferenceable(4) [[THIS1]], ptr noundef nonnull align 4 dereferenceable(4) [[TMP0]], ptr noundef [[T]])
@@ -648,8 +650,10 @@ int main() {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
@@ -850,8 +854,10 @@ int main() {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
 // CHECK1-NEXT:    call void @_ZN1SIiEC2ERKS0_2St(ptr noundef nonnull align 4 dereferenceable(4) [[THIS1]], ptr noundef nonnull align 4 dereferenceable(4) [[TMP0]], ptr noundef [[T]])
@@ -901,8 +907,10 @@ int main() {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S_0:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
@@ -1306,8 +1314,10 @@ int main() {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 4
+// CHECK3-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 4
+// CHECK3-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 4
 // CHECK3-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 4
 // CHECK3-NEXT:    call void @_ZN1SIfEC2ERKS0_2St(ptr noundef nonnull align 4 dereferenceable(4) [[THIS1]], ptr noundef nonnull align 4 dereferenceable(4) [[TMP0]], ptr noundef [[T]])
@@ -1447,8 +1457,10 @@ int main() {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 4
+// CHECK3-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 4
+// CHECK3-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 4
 // CHECK3-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 4
@@ -1647,8 +1659,10 @@ int main() {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 4
+// CHECK3-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 4
+// CHECK3-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 4
 // CHECK3-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 4
 // CHECK3-NEXT:    call void @_ZN1SIiEC2ERKS0_2St(ptr noundef nonnull align 4 dereferenceable(4) [[THIS1]], ptr noundef nonnull align 4 dereferenceable(4) [[TMP0]], ptr noundef [[T]])
@@ -1698,8 +1712,10 @@ int main() {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 4
+// CHECK3-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 4
+// CHECK3-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 4
 // CHECK3-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S_0:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 4
diff --git a/clang/test/OpenMP/target_teams_distribute_parallel_for_firstprivate_codegen.cpp b/clang/test/OpenMP/target_teams_distribute_parallel_for_firstprivate_codegen.cpp
index a01d42a..c9eee578 100644
--- a/clang/test/OpenMP/target_teams_distribute_parallel_for_firstprivate_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_distribute_parallel_for_firstprivate_codegen.cpp
@@ -563,8 +563,10 @@ int main() {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
 // CHECK1-NEXT:    call void @_ZN1SIfEC2ERKS0_2St(ptr noundef nonnull align 4 dereferenceable(4) [[THIS1]], ptr noundef nonnull align 4 dereferenceable(4) [[TMP0]], ptr noundef [[T]])
@@ -839,8 +841,10 @@ int main() {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
@@ -1036,8 +1040,10 @@ int main() {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
 // CHECK1-NEXT:    call void @_ZN1SIiEC2ERKS0_2St(ptr noundef nonnull align 4 dereferenceable(4) [[THIS1]], ptr noundef nonnull align 4 dereferenceable(4) [[TMP0]], ptr noundef [[T]])
@@ -1222,8 +1228,10 @@ int main() {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S_0:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
@@ -1622,8 +1630,10 @@ int main() {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 4
+// CHECK3-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 4
+// CHECK3-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 4
 // CHECK3-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 4
 // CHECK3-NEXT:    call void @_ZN1SIfEC2ERKS0_2St(ptr noundef nonnull align 4 dereferenceable(4) [[THIS1]], ptr noundef nonnull align 4 dereferenceable(4) [[TMP0]], ptr noundef [[T]])
@@ -1894,8 +1904,10 @@ int main() {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 4
+// CHECK3-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 4
+// CHECK3-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 4
 // CHECK3-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 4
@@ -2089,8 +2101,10 @@ int main() {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 4
+// CHECK3-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 4
+// CHECK3-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 4
 // CHECK3-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 4
 // CHECK3-NEXT:    call void @_ZN1SIiEC2ERKS0_2St(ptr noundef nonnull align 4 dereferenceable(4) [[THIS1]], ptr noundef nonnull align 4 dereferenceable(4) [[TMP0]], ptr noundef [[T]])
@@ -2271,8 +2285,10 @@ int main() {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 4
+// CHECK3-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 4
+// CHECK3-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 4
 // CHECK3-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S_0:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 4
@@ -2812,8 +2828,10 @@ int main() {
 // CHECK13-NEXT:  entry:
 // CHECK13-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK13-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK13-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK13-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK13-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK13-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK13-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK13-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
 // CHECK13-NEXT:    call void @_ZN1SIfEC2ERKS0_2St(ptr noundef nonnull align 4 dereferenceable(4) [[THIS1]], ptr noundef nonnull align 4 dereferenceable(4) [[TMP0]], ptr noundef [[T]]) #[[ATTR4]]
@@ -3125,8 +3143,10 @@ int main() {
 // CHECK13-NEXT:  entry:
 // CHECK13-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK13-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK13-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK13-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK13-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK13-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK13-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK13-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
 // CHECK13-NEXT:    call void @_ZN1SIiEC2ERKS0_2St(ptr noundef nonnull align 4 dereferenceable(4) [[THIS1]], ptr noundef nonnull align 4 dereferenceable(4) [[TMP0]], ptr noundef [[T]]) #[[ATTR4]]
@@ -3314,8 +3334,10 @@ int main() {
 // CHECK13-NEXT:  entry:
 // CHECK13-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK13-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK13-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK13-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK13-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK13-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK13-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK13-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK13-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
@@ -3343,8 +3365,10 @@ int main() {
 // CHECK13-NEXT:  entry:
 // CHECK13-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK13-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK13-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK13-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK13-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK13-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK13-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK13-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S_0:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK13-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
@@ -3518,8 +3542,10 @@ int main() {
 // CHECK15-NEXT:  entry:
 // CHECK15-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
 // CHECK15-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 4
+// CHECK15-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 4
 // CHECK15-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
 // CHECK15-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 4
+// CHECK15-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 4
 // CHECK15-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
 // CHECK15-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 4
 // CHECK15-NEXT:    call void @_ZN1SIfEC2ERKS0_2St(ptr noundef nonnull align 4 dereferenceable(4) [[THIS1]], ptr noundef nonnull align 4 dereferenceable(4) [[TMP0]], ptr noundef [[T]]) #[[ATTR4]]
@@ -3825,8 +3851,10 @@ int main() {
 // CHECK15-NEXT:  entry:
 // CHECK15-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
 // CHECK15-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 4
+// CHECK15-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 4
 // CHECK15-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
 // CHECK15-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 4
+// CHECK15-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 4
 // CHECK15-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
 // CHECK15-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 4
 // CHECK15-NEXT:    call void @_ZN1SIiEC2ERKS0_2St(ptr noundef nonnull align 4 dereferenceable(4) [[THIS1]], ptr noundef nonnull align 4 dereferenceable(4) [[TMP0]], ptr noundef [[T]]) #[[ATTR4]]
@@ -4010,8 +4038,10 @@ int main() {
 // CHECK15-NEXT:  entry:
 // CHECK15-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
 // CHECK15-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 4
+// CHECK15-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 4
 // CHECK15-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
 // CHECK15-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 4
+// CHECK15-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 4
 // CHECK15-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
 // CHECK15-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK15-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 4
@@ -4039,8 +4069,10 @@ int main() {
 // CHECK15-NEXT:  entry:
 // CHECK15-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
 // CHECK15-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 4
+// CHECK15-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 4
 // CHECK15-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
 // CHECK15-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 4
+// CHECK15-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 4
 // CHECK15-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
 // CHECK15-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S_0:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK15-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 4
diff --git a/clang/test/OpenMP/target_teams_distribute_parallel_for_simd_firstprivate_codegen.cpp b/clang/test/OpenMP/target_teams_distribute_parallel_for_simd_firstprivate_codegen.cpp
index 35bd6fa..bc272f8 100644
--- a/clang/test/OpenMP/target_teams_distribute_parallel_for_simd_firstprivate_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_distribute_parallel_for_simd_firstprivate_codegen.cpp
@@ -568,8 +568,10 @@ int main() {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
 // CHECK1-NEXT:    call void @_ZN1SIfEC2ERKS0_2St(ptr noundef nonnull align 4 dereferenceable(4) [[THIS1]], ptr noundef nonnull align 4 dereferenceable(4) [[TMP0]], ptr noundef [[T]])
@@ -851,8 +853,10 @@ int main() {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
@@ -1055,8 +1059,10 @@ int main() {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
 // CHECK1-NEXT:    call void @_ZN1SIiEC2ERKS0_2St(ptr noundef nonnull align 4 dereferenceable(4) [[THIS1]], ptr noundef nonnull align 4 dereferenceable(4) [[TMP0]], ptr noundef [[T]])
@@ -1248,8 +1254,10 @@ int main() {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S_0:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
@@ -1655,8 +1663,10 @@ int main() {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 4
+// CHECK3-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 4
+// CHECK3-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 4
 // CHECK3-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 4
 // CHECK3-NEXT:    call void @_ZN1SIfEC2ERKS0_2St(ptr noundef nonnull align 4 dereferenceable(4) [[THIS1]], ptr noundef nonnull align 4 dereferenceable(4) [[TMP0]], ptr noundef [[T]])
@@ -1934,8 +1944,10 @@ int main() {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 4
+// CHECK3-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 4
+// CHECK3-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 4
 // CHECK3-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 4
@@ -2136,8 +2148,10 @@ int main() {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 4
+// CHECK3-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 4
+// CHECK3-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 4
 // CHECK3-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 4
 // CHECK3-NEXT:    call void @_ZN1SIiEC2ERKS0_2St(ptr noundef nonnull align 4 dereferenceable(4) [[THIS1]], ptr noundef nonnull align 4 dereferenceable(4) [[TMP0]], ptr noundef [[T]])
@@ -2325,8 +2339,10 @@ int main() {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 4
+// CHECK3-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 4
+// CHECK3-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 4
 // CHECK3-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S_0:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 4
@@ -3661,8 +3677,10 @@ int main() {
 // CHECK13-NEXT:  entry:
 // CHECK13-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK13-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK13-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK13-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK13-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK13-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK13-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK13-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
 // CHECK13-NEXT:    call void @_ZN1SIfEC2ERKS0_2St(ptr noundef nonnull align 4 dereferenceable(4) [[THIS1]], ptr noundef nonnull align 4 dereferenceable(4) [[TMP0]], ptr noundef [[T]]) #[[ATTR4]]
@@ -3988,8 +4006,10 @@ int main() {
 // CHECK13-NEXT:  entry:
 // CHECK13-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK13-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK13-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK13-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK13-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK13-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK13-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK13-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
 // CHECK13-NEXT:    call void @_ZN1SIiEC2ERKS0_2St(ptr noundef nonnull align 4 dereferenceable(4) [[THIS1]], ptr noundef nonnull align 4 dereferenceable(4) [[TMP0]], ptr noundef [[T]]) #[[ATTR4]]
@@ -4184,8 +4204,10 @@ int main() {
 // CHECK13-NEXT:  entry:
 // CHECK13-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK13-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK13-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK13-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK13-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK13-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK13-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK13-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK13-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
@@ -4213,8 +4235,10 @@ int main() {
 // CHECK13-NEXT:  entry:
 // CHECK13-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK13-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK13-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK13-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK13-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK13-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK13-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK13-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S_0:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK13-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
@@ -4395,8 +4419,10 @@ int main() {
 // CHECK15-NEXT:  entry:
 // CHECK15-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
 // CHECK15-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 4
+// CHECK15-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 4
 // CHECK15-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
 // CHECK15-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 4
+// CHECK15-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 4
 // CHECK15-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
 // CHECK15-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 4
 // CHECK15-NEXT:    call void @_ZN1SIfEC2ERKS0_2St(ptr noundef nonnull align 4 dereferenceable(4) [[THIS1]], ptr noundef nonnull align 4 dereferenceable(4) [[TMP0]], ptr noundef [[T]]) #[[ATTR4]]
@@ -4716,8 +4742,10 @@ int main() {
 // CHECK15-NEXT:  entry:
 // CHECK15-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
 // CHECK15-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 4
+// CHECK15-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 4
 // CHECK15-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
 // CHECK15-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 4
+// CHECK15-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 4
 // CHECK15-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
 // CHECK15-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 4
 // CHECK15-NEXT:    call void @_ZN1SIiEC2ERKS0_2St(ptr noundef nonnull align 4 dereferenceable(4) [[THIS1]], ptr noundef nonnull align 4 dereferenceable(4) [[TMP0]], ptr noundef [[T]]) #[[ATTR4]]
@@ -4908,8 +4936,10 @@ int main() {
 // CHECK15-NEXT:  entry:
 // CHECK15-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
 // CHECK15-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 4
+// CHECK15-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 4
 // CHECK15-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
 // CHECK15-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 4
+// CHECK15-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 4
 // CHECK15-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
 // CHECK15-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK15-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 4
@@ -4937,8 +4967,10 @@ int main() {
 // CHECK15-NEXT:  entry:
 // CHECK15-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
 // CHECK15-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 4
+// CHECK15-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 4
 // CHECK15-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
 // CHECK15-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 4
+// CHECK15-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 4
 // CHECK15-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
 // CHECK15-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S_0:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK15-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 4
diff --git a/clang/test/OpenMP/target_teams_distribute_simd_firstprivate_codegen.cpp b/clang/test/OpenMP/target_teams_distribute_simd_firstprivate_codegen.cpp
index 3fc376b..20e0185 100644
--- a/clang/test/OpenMP/target_teams_distribute_simd_firstprivate_codegen.cpp
+++ b/clang/test/OpenMP/target_teams_distribute_simd_firstprivate_codegen.cpp
@@ -514,8 +514,10 @@ int main() {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
 // CHECK1-NEXT:    call void @_ZN1SIfEC2ERKS0_2St(ptr noundef nonnull align 4 dereferenceable(4) [[THIS1]], ptr noundef nonnull align 4 dereferenceable(4) [[TMP0]], ptr noundef [[T]])
@@ -655,8 +657,10 @@ int main() {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
@@ -864,8 +868,10 @@ int main() {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
 // CHECK1-NEXT:    call void @_ZN1SIiEC2ERKS0_2St(ptr noundef nonnull align 4 dereferenceable(4) [[THIS1]], ptr noundef nonnull align 4 dereferenceable(4) [[TMP0]], ptr noundef [[T]])
@@ -915,8 +921,10 @@ int main() {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S_0:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
@@ -1327,8 +1335,10 @@ int main() {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 4
+// CHECK3-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 4
+// CHECK3-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 4
 // CHECK3-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 4
 // CHECK3-NEXT:    call void @_ZN1SIfEC2ERKS0_2St(ptr noundef nonnull align 4 dereferenceable(4) [[THIS1]], ptr noundef nonnull align 4 dereferenceable(4) [[TMP0]], ptr noundef [[T]])
@@ -1468,8 +1478,10 @@ int main() {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 4
+// CHECK3-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 4
+// CHECK3-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 4
 // CHECK3-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 4
@@ -1675,8 +1687,10 @@ int main() {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 4
+// CHECK3-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 4
+// CHECK3-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 4
 // CHECK3-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 4
 // CHECK3-NEXT:    call void @_ZN1SIiEC2ERKS0_2St(ptr noundef nonnull align 4 dereferenceable(4) [[THIS1]], ptr noundef nonnull align 4 dereferenceable(4) [[TMP0]], ptr noundef [[T]])
@@ -1726,8 +1740,10 @@ int main() {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 4
+// CHECK3-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 4
+// CHECK3-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 4
 // CHECK3-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S_0:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 4
diff --git a/clang/test/OpenMP/teams_distribute_firstprivate_codegen.cpp b/clang/test/OpenMP/teams_distribute_firstprivate_codegen.cpp
index 9c7ead5..3f07941 100644
--- a/clang/test/OpenMP/teams_distribute_firstprivate_codegen.cpp
+++ b/clang/test/OpenMP/teams_distribute_firstprivate_codegen.cpp
@@ -510,8 +510,10 @@ int main() {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
 // CHECK1-NEXT:    call void @_ZN1SIfEC2ERKS0_2St(ptr noundef nonnull align 4 dereferenceable(4) [[THIS1]], ptr noundef nonnull align 4 dereferenceable(4) [[TMP0]], ptr noundef [[T]])
@@ -653,8 +655,10 @@ int main() {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
@@ -855,8 +859,10 @@ int main() {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
 // CHECK1-NEXT:    call void @_ZN1SIiEC2ERKS0_2St(ptr noundef nonnull align 4 dereferenceable(4) [[THIS1]], ptr noundef nonnull align 4 dereferenceable(4) [[TMP0]], ptr noundef [[T]])
@@ -906,8 +912,10 @@ int main() {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S_0:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
@@ -1311,8 +1319,10 @@ int main() {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 4
+// CHECK3-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 4
+// CHECK3-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 4
 // CHECK3-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 4
 // CHECK3-NEXT:    call void @_ZN1SIfEC2ERKS0_2St(ptr noundef nonnull align 4 dereferenceable(4) [[THIS1]], ptr noundef nonnull align 4 dereferenceable(4) [[TMP0]], ptr noundef [[T]])
@@ -1454,8 +1464,10 @@ int main() {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 4
+// CHECK3-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 4
+// CHECK3-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 4
 // CHECK3-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 4
@@ -1654,8 +1666,10 @@ int main() {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 4
+// CHECK3-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 4
+// CHECK3-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 4
 // CHECK3-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 4
 // CHECK3-NEXT:    call void @_ZN1SIiEC2ERKS0_2St(ptr noundef nonnull align 4 dereferenceable(4) [[THIS1]], ptr noundef nonnull align 4 dereferenceable(4) [[TMP0]], ptr noundef [[T]])
@@ -1705,8 +1719,10 @@ int main() {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 4
+// CHECK3-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 4
+// CHECK3-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 4
 // CHECK3-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S_0:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 4
diff --git a/clang/test/OpenMP/teams_distribute_parallel_for_firstprivate_codegen.cpp b/clang/test/OpenMP/teams_distribute_parallel_for_firstprivate_codegen.cpp
index 107f178..1884ea8 100644
--- a/clang/test/OpenMP/teams_distribute_parallel_for_firstprivate_codegen.cpp
+++ b/clang/test/OpenMP/teams_distribute_parallel_for_firstprivate_codegen.cpp
@@ -539,8 +539,10 @@ int main() {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
 // CHECK1-NEXT:    call void @_ZN1SIfEC2ERKS0_2St(ptr noundef nonnull align 4 dereferenceable(4) [[THIS1]], ptr noundef nonnull align 4 dereferenceable(4) [[TMP0]], ptr noundef [[T]])
@@ -817,8 +819,10 @@ int main() {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
@@ -1014,8 +1018,10 @@ int main() {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
 // CHECK1-NEXT:    call void @_ZN1SIiEC2ERKS0_2St(ptr noundef nonnull align 4 dereferenceable(4) [[THIS1]], ptr noundef nonnull align 4 dereferenceable(4) [[TMP0]], ptr noundef [[T]])
@@ -1200,8 +1206,10 @@ int main() {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S_0:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
@@ -1600,8 +1608,10 @@ int main() {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 4
+// CHECK3-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 4
+// CHECK3-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 4
 // CHECK3-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 4
 // CHECK3-NEXT:    call void @_ZN1SIfEC2ERKS0_2St(ptr noundef nonnull align 4 dereferenceable(4) [[THIS1]], ptr noundef nonnull align 4 dereferenceable(4) [[TMP0]], ptr noundef [[T]])
@@ -1874,8 +1884,10 @@ int main() {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 4
+// CHECK3-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 4
+// CHECK3-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 4
 // CHECK3-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 4
@@ -2069,8 +2081,10 @@ int main() {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 4
+// CHECK3-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 4
+// CHECK3-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 4
 // CHECK3-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 4
 // CHECK3-NEXT:    call void @_ZN1SIiEC2ERKS0_2St(ptr noundef nonnull align 4 dereferenceable(4) [[THIS1]], ptr noundef nonnull align 4 dereferenceable(4) [[TMP0]], ptr noundef [[T]])
@@ -2251,8 +2265,10 @@ int main() {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 4
+// CHECK3-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 4
+// CHECK3-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 4
 // CHECK3-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S_0:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 4
diff --git a/clang/test/OpenMP/teams_distribute_parallel_for_simd_firstprivate_codegen.cpp b/clang/test/OpenMP/teams_distribute_parallel_for_simd_firstprivate_codegen.cpp
index c36ba04..5dc93d7 100644
--- a/clang/test/OpenMP/teams_distribute_parallel_for_simd_firstprivate_codegen.cpp
+++ b/clang/test/OpenMP/teams_distribute_parallel_for_simd_firstprivate_codegen.cpp
@@ -549,8 +549,10 @@ int main() {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
 // CHECK1-NEXT:    call void @_ZN1SIfEC2ERKS0_2St(ptr noundef nonnull align 4 dereferenceable(4) [[THIS1]], ptr noundef nonnull align 4 dereferenceable(4) [[TMP0]], ptr noundef [[T]])
@@ -834,8 +836,10 @@ int main() {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
@@ -1038,8 +1042,10 @@ int main() {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
 // CHECK1-NEXT:    call void @_ZN1SIiEC2ERKS0_2St(ptr noundef nonnull align 4 dereferenceable(4) [[THIS1]], ptr noundef nonnull align 4 dereferenceable(4) [[TMP0]], ptr noundef [[T]])
@@ -1231,8 +1237,10 @@ int main() {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S_0:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
@@ -1638,8 +1646,10 @@ int main() {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 4
+// CHECK3-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 4
+// CHECK3-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 4
 // CHECK3-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 4
 // CHECK3-NEXT:    call void @_ZN1SIfEC2ERKS0_2St(ptr noundef nonnull align 4 dereferenceable(4) [[THIS1]], ptr noundef nonnull align 4 dereferenceable(4) [[TMP0]], ptr noundef [[T]])
@@ -1919,8 +1929,10 @@ int main() {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 4
+// CHECK3-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 4
+// CHECK3-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 4
 // CHECK3-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 4
@@ -2121,8 +2133,10 @@ int main() {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 4
+// CHECK3-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 4
+// CHECK3-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 4
 // CHECK3-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 4
 // CHECK3-NEXT:    call void @_ZN1SIiEC2ERKS0_2St(ptr noundef nonnull align 4 dereferenceable(4) [[THIS1]], ptr noundef nonnull align 4 dereferenceable(4) [[TMP0]], ptr noundef [[T]])
@@ -2310,8 +2324,10 @@ int main() {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 4
+// CHECK3-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 4
+// CHECK3-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 4
 // CHECK3-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S_0:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 4
diff --git a/clang/test/OpenMP/teams_distribute_simd_firstprivate_codegen.cpp b/clang/test/OpenMP/teams_distribute_simd_firstprivate_codegen.cpp
index 38609ccb..164a8d9 100644
--- a/clang/test/OpenMP/teams_distribute_simd_firstprivate_codegen.cpp
+++ b/clang/test/OpenMP/teams_distribute_simd_firstprivate_codegen.cpp
@@ -517,8 +517,10 @@ int main() {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
 // CHECK1-NEXT:    call void @_ZN1SIfEC2ERKS0_2St(ptr noundef nonnull align 4 dereferenceable(4) [[THIS1]], ptr noundef nonnull align 4 dereferenceable(4) [[TMP0]], ptr noundef [[T]])
@@ -660,8 +662,10 @@ int main() {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
@@ -869,8 +873,10 @@ int main() {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
 // CHECK1-NEXT:    call void @_ZN1SIiEC2ERKS0_2St(ptr noundef nonnull align 4 dereferenceable(4) [[THIS1]], ptr noundef nonnull align 4 dereferenceable(4) [[TMP0]], ptr noundef [[T]])
@@ -920,8 +926,10 @@ int main() {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S_0:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
@@ -1332,8 +1340,10 @@ int main() {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 4
+// CHECK3-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 4
+// CHECK3-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 4
 // CHECK3-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 4
 // CHECK3-NEXT:    call void @_ZN1SIfEC2ERKS0_2St(ptr noundef nonnull align 4 dereferenceable(4) [[THIS1]], ptr noundef nonnull align 4 dereferenceable(4) [[TMP0]], ptr noundef [[T]])
@@ -1475,8 +1485,10 @@ int main() {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 4
+// CHECK3-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 4
+// CHECK3-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 4
 // CHECK3-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 4
@@ -1682,8 +1694,10 @@ int main() {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 4
+// CHECK3-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 4
+// CHECK3-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 4
 // CHECK3-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 4
 // CHECK3-NEXT:    call void @_ZN1SIiEC2ERKS0_2St(ptr noundef nonnull align 4 dereferenceable(4) [[THIS1]], ptr noundef nonnull align 4 dereferenceable(4) [[TMP0]], ptr noundef [[T]])
@@ -1733,8 +1747,10 @@ int main() {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 4
+// CHECK3-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 4
+// CHECK3-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 4
 // CHECK3-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
 // CHECK3-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S_0:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 4
diff --git a/clang/test/OpenMP/teams_firstprivate_codegen.cpp b/clang/test/OpenMP/teams_firstprivate_codegen.cpp
index 4dc09e0..e08d498 100644
--- a/clang/test/OpenMP/teams_firstprivate_codegen.cpp
+++ b/clang/test/OpenMP/teams_firstprivate_codegen.cpp
@@ -573,8 +573,10 @@ void array_func(float a[3], St s[2], int n, long double vla1[n]) {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK9-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK9-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK9-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
 // CHECK9-NEXT:    call void @_ZN1SIfEC2ERKS0_2St(ptr nonnull align 4 dereferenceable(4) [[THIS1]], ptr nonnull align 4 dereferenceable(4) [[TMP0]], ptr [[T]])
@@ -816,8 +818,10 @@ void array_func(float a[3], St s[2], int n, long double vla1[n]) {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK9-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK9-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK9-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK9-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK9-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
@@ -964,8 +968,10 @@ void array_func(float a[3], St s[2], int n, long double vla1[n]) {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK9-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK9-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK9-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
 // CHECK9-NEXT:    call void @_ZN1SIiEC2ERKS0_2St(ptr nonnull align 4 dereferenceable(4) [[THIS1]], ptr nonnull align 4 dereferenceable(4) [[TMP0]], ptr [[T]])
@@ -1044,8 +1050,10 @@ void array_func(float a[3], St s[2], int n, long double vla1[n]) {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 8
+// CHECK9-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 8
+// CHECK9-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 8
 // CHECK9-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK9-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S_0:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK9-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 8
@@ -1366,8 +1374,10 @@ void array_func(float a[3], St s[2], int n, long double vla1[n]) {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 4
+// CHECK11-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 4
+// CHECK11-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 4
 // CHECK11-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 4
 // CHECK11-NEXT:    call void @_ZN1SIfEC2ERKS0_2St(ptr nonnull align 4 dereferenceable(4) [[THIS1]], ptr nonnull align 4 dereferenceable(4) [[TMP0]], ptr [[T]])
@@ -1609,8 +1619,10 @@ void array_func(float a[3], St s[2], int n, long double vla1[n]) {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 4
+// CHECK11-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 4
+// CHECK11-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 4
 // CHECK11-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
 // CHECK11-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK11-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 4
@@ -1757,8 +1769,10 @@ void array_func(float a[3], St s[2], int n, long double vla1[n]) {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 4
+// CHECK11-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 4
+// CHECK11-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 4
 // CHECK11-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 4
 // CHECK11-NEXT:    call void @_ZN1SIiEC2ERKS0_2St(ptr nonnull align 4 dereferenceable(4) [[THIS1]], ptr nonnull align 4 dereferenceable(4) [[TMP0]], ptr [[T]])
@@ -1837,8 +1851,10 @@ void array_func(float a[3], St s[2], int n, long double vla1[n]) {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[S_ADDR:%.*]] = alloca ptr, align 4
+// CHECK11-NEXT:    [[T_INDIRECT_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[S]], ptr [[S_ADDR]], align 4
+// CHECK11-NEXT:    store ptr [[T]], ptr [[T_INDIRECT_ADDR]], align 4
 // CHECK11-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
 // CHECK11-NEXT:    [[F:%.*]] = getelementptr inbounds [[STRUCT_S_0:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK11-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ADDR]], align 4