diff --git a/clang/lib/CodeGen/CGDeclCXX.cpp b/clang/lib/CodeGen/CGDeclCXX.cpp
index 8d9d0b21bfe11..f56e182169311 100644
--- a/clang/lib/CodeGen/CGDeclCXX.cpp
+++ b/clang/lib/CodeGen/CGDeclCXX.cpp
@@ -353,9 +353,6 @@ CodeGenModule::EmitCXXGlobalVarDeclInitFunc(const VarDecl *D,
 
   if (D->getTLSKind()) {
     // FIXME: Should we support init_priority for thread_local?
-    // FIXME: Ideally, initialization of instantiated thread_local static data
-    // members of class templates should not trigger initialization of other
-    // entities in the TU.
     // FIXME: We only need to register one __cxa_thread_atexit function for the
     // entire TU.
     CXXThreadLocalInits.push_back(Fn);
diff --git a/clang/lib/CodeGen/ItaniumCXXABI.cpp b/clang/lib/CodeGen/ItaniumCXXABI.cpp
index b5d90ea59a498..f7a8dd66c527c 100644
--- a/clang/lib/CodeGen/ItaniumCXXABI.cpp
+++ b/clang/lib/CodeGen/ItaniumCXXABI.cpp
@@ -2272,7 +2272,21 @@ void ItaniumCXXABI::EmitThreadLocalInitFuncs(
     ArrayRef<llvm::Function *> CXXThreadLocalInits,
     ArrayRef<const VarDecl *> CXXThreadLocalInitVars) {
   llvm::Function *InitFunc = nullptr;
-  if (!CXXThreadLocalInits.empty()) {
+
+  // Separate initializers into those with ordered (or partially-ordered)
+  // initialization and those with unordered initialization.
+  llvm::SmallVector<llvm::Function *, 8> OrderedInits;
+  llvm::SmallDenseMap<const VarDecl *, llvm::Function *> UnorderedInits;
+  for (unsigned I = 0; I != CXXThreadLocalInits.size(); ++I) {
+    if (isTemplateInstantiation(
+            CXXThreadLocalInitVars[I]->getTemplateSpecializationKind()))
+      UnorderedInits[CXXThreadLocalInitVars[I]->getCanonicalDecl()] =
+          CXXThreadLocalInits[I];
+    else
+      OrderedInits.push_back(CXXThreadLocalInits[I]);
+  }
+
+  if (!OrderedInits.empty()) {
     // Generate a guarded initialization function.
     llvm::FunctionType *FTy =
         llvm::FunctionType::get(CGM.VoidTy, /*isVarArg=*/false);
@@ -2289,24 +2303,28 @@ void ItaniumCXXABI::EmitThreadLocalInitFuncs(
     CharUnits GuardAlign = CharUnits::One();
     Guard->setAlignment(GuardAlign.getQuantity());
 
-    CodeGenFunction(CGM)
-        .GenerateCXXGlobalInitFunc(InitFunc, CXXThreadLocalInits,
-                                   Address(Guard, GuardAlign));
+    CodeGenFunction(CGM).GenerateCXXGlobalInitFunc(InitFunc, OrderedInits,
+                                                   Address(Guard, GuardAlign));
     // On Darwin platforms, use CXX_FAST_TLS calling convention.
     if (CGM.getTarget().getTriple().isOSDarwin()) {
       InitFunc->setCallingConv(llvm::CallingConv::CXX_FAST_TLS);
       InitFunc->addFnAttr(llvm::Attribute::NoUnwind);
     }
   }
+
+  // Emit thread wrappers.
   for (const VarDecl *VD : CXXThreadLocals) {
     llvm::GlobalVariable *Var =
         cast<llvm::GlobalVariable>(CGM.GetGlobalValue(CGM.getMangledName(VD)));
+    llvm::Function *Wrapper = getOrCreateThreadLocalWrapper(VD, Var);
 
     // Some targets require that all access to thread local variables go through
     // the thread wrapper.  This means that we cannot attempt to create a thread
     // wrapper or a thread helper.
-    if (isThreadWrapperReplaceable(VD, CGM) && !VD->hasDefinition())
+    if (isThreadWrapperReplaceable(VD, CGM) && !VD->hasDefinition()) {
+      Wrapper->setLinkage(llvm::Function::ExternalLinkage);
       continue;
+    }
 
     // Mangle the name for the thread_local initialization function.
     SmallString<256> InitFnName;
@@ -2322,18 +2340,21 @@ void ItaniumCXXABI::EmitThreadLocalInitFuncs(
     bool InitIsInitFunc = false;
     if (VD->hasDefinition()) {
       InitIsInitFunc = true;
-      if (InitFunc)
+      llvm::Function *InitFuncToUse = InitFunc;
+      if (isTemplateInstantiation(VD->getTemplateSpecializationKind()))
+        InitFuncToUse = UnorderedInits.lookup(VD->getCanonicalDecl());
+      if (InitFuncToUse)
         Init = llvm::GlobalAlias::create(Var->getLinkage(), InitFnName.str(),
-                                         InitFunc);
+                                         InitFuncToUse);
     } else {
       // Emit a weak global function referring to the initialization function.
       // This function will not exist if the TU defining the thread_local
       // variable in question does not need any dynamic initialization for
       // its thread_local variables.
       llvm::FunctionType *FnTy = llvm::FunctionType::get(CGM.VoidTy, false);
-      Init = llvm::Function::Create(
-          FnTy, llvm::GlobalVariable::ExternalWeakLinkage, InitFnName.str(),
-          &CGM.getModule());
+      Init = llvm::Function::Create(FnTy,
+                                    llvm::GlobalVariable::ExternalWeakLinkage,
+                                    InitFnName.str(), &CGM.getModule());
       const CGFunctionInfo &FI = CGM.getTypes().arrangeNullaryFunction();
       CGM.SetLLVMFunctionAttributes(nullptr, FI, cast<llvm::Function>(Init));
     }
@@ -2341,7 +2362,6 @@ void ItaniumCXXABI::EmitThreadLocalInitFuncs(
     if (Init)
       Init->setVisibility(Var->getVisibility());
 
-    llvm::Function *Wrapper = getOrCreateThreadLocalWrapper(VD, Var);
     llvm::LLVMContext &Context = CGM.getModule().getContext();
     llvm::BasicBlock *Entry = llvm::BasicBlock::Create(Context, "", Wrapper);
     CGBuilderTy Builder(CGM, Entry);
diff --git a/clang/test/CodeGenCXX/cxx11-thread-local.cpp b/clang/test/CodeGenCXX/cxx11-thread-local.cpp
index f465cbdeea847..3231a76ba9208 100644
--- a/clang/test/CodeGenCXX/cxx11-thread-local.cpp
+++ b/clang/test/CodeGenCXX/cxx11-thread-local.cpp
@@ -6,18 +6,18 @@
 int f();
 int g();
 
-// LINUX: @a = thread_local global i32 0
-// DARWIN: @a = internal thread_local global i32 0
+// LINUX-DAG: @a = thread_local global i32 0
+// DARWIN-DAG: @a = internal thread_local global i32 0
 thread_local int a = f();
 extern thread_local int b;
-// CHECK: @c = global i32 0
+// CHECK-DAG: @c = global i32 0
 int c = b;
-// CHECK: @_ZL1d = internal thread_local global i32 0
+// CHECK-DAG: @_ZL1d = internal thread_local global i32 0
 static thread_local int d = g();
 
 struct U { static thread_local int m; };
-// LINUX: @_ZN1U1mE = thread_local global i32 0
-// DARWIN: @_ZN1U1mE = internal thread_local global i32 0
+// LINUX-DAG: @_ZN1U1mE = thread_local global i32 0
+// DARWIN-DAG: @_ZN1U1mE = internal thread_local global i32 0
 thread_local int U::m = f();
 
 namespace MismatchedInitType {
@@ -35,37 +35,64 @@ namespace MismatchedInitType {
 template<typename T> struct V { static thread_local int m; };
 template<typename T> thread_local int V<T>::m = g();
 
-// CHECK: @e = global i32 0
-int e = V<int>::m;
+template<typename T> struct W { static thread_local int m; };
+template<typename T> thread_local int W<T>::m = 123;
 
-// CHECK: @_ZN1VIiE1mE = linkonce_odr thread_local global i32 0
+struct Dtor { ~Dtor(); };
+template<typename T> struct X { static thread_local Dtor m; };
+template<typename T> thread_local Dtor X<T>::m;
 
-// CHECK: @_ZZ1fvE1n = internal thread_local global i32 0
+// CHECK-DAG: @e = global
+void *e = V<int>::m + W<int>::m + &X<int>::m;
 
-// CHECK: @_ZGVZ1fvE1n = internal thread_local global i8 0
+template thread_local int V<float>::m;
+template thread_local int W<float>::m;
+template thread_local Dtor X<float>::m;
 
-// CHECK: @_ZZ8tls_dtorvE1s = internal thread_local global
-// CHECK: @_ZGVZ8tls_dtorvE1s = internal thread_local global i8 0
+extern template thread_local int V<char>::m;
+extern template thread_local int W<char>::m;
+extern template thread_local Dtor X<char>::m;
 
-// CHECK: @_ZZ8tls_dtorvE1t = internal thread_local global
-// CHECK: @_ZGVZ8tls_dtorvE1t = internal thread_local global i8 0
+void *e2 = V<char>::m + W<char>::m + &X<char>::m;
 
-// CHECK: @_ZZ8tls_dtorvE1u = internal thread_local global
-// CHECK: @_ZGVZ8tls_dtorvE1u = internal thread_local global i8 0
-// CHECK: @_ZGRZ8tls_dtorvE1u_ = internal thread_local global
+// CHECK-DAG: @_ZN1VIiE1mE = linkonce_odr thread_local global i32 0
+// CHECK-DAG: @_ZN1WIiE1mE = linkonce_odr thread_local global i32 123
+// CHECK-DAG: @_ZN1XIiE1mE = linkonce_odr thread_local global {{.*}}
+// CHECK-DAG: @_ZN1VIfE1mE = weak_odr thread_local global i32 0
+// CHECK-DAG: @_ZN1WIfE1mE = weak_odr thread_local global i32 123
+// CHECK-DAG: @_ZN1XIfE1mE = weak_odr thread_local global {{.*}}
 
-// CHECK: @_ZGVN1VIiE1mE = linkonce_odr thread_local global i64 0
+// CHECK-DAG: @_ZZ1fvE1n = internal thread_local global i32 0
 
-// CHECK: @__tls_guard = internal thread_local global i8 0
+// CHECK-DAG: @_ZGVZ1fvE1n = internal thread_local global i8 0
 
-// CHECK: @llvm.global_ctors = appending global {{.*}} @[[GLOBAL_INIT:[^ ]*]]
+// CHECK-DAG: @_ZZ8tls_dtorvE1s = internal thread_local global
+// CHECK-DAG: @_ZGVZ8tls_dtorvE1s = internal thread_local global i8 0
 
-// LINUX: @_ZTH1a = alias void (), void ()* @__tls_init
-// DARWIN: @_ZTH1a = internal alias void (), void ()* @__tls_init
-// CHECK: @_ZTHL1d = internal alias void (), void ()* @__tls_init
-// LINUX: @_ZTHN1U1mE = alias void (), void ()* @__tls_init
-// DARWIN: @_ZTHN1U1mE = internal alias void (), void ()* @__tls_init
-// CHECK: @_ZTHN1VIiE1mE = linkonce_odr alias void (), void ()* @__tls_init
+// CHECK-DAG: @_ZZ8tls_dtorvE1t = internal thread_local global
+// CHECK-DAG: @_ZGVZ8tls_dtorvE1t = internal thread_local global i8 0
+
+// CHECK-DAG: @_ZZ8tls_dtorvE1u = internal thread_local global
+// CHECK-DAG: @_ZGVZ8tls_dtorvE1u = internal thread_local global i8 0
+// CHECK-DAG: @_ZGRZ8tls_dtorvE1u_ = internal thread_local global
+
+// CHECK-DAG: @_ZGVN1VIiE1mE = linkonce_odr thread_local global i64 0
+
+// CHECK-DAG: @__tls_guard = internal thread_local global i8 0
+
+// CHECK-DAG: @llvm.global_ctors = appending global {{.*}} @[[GLOBAL_INIT:[^ ]*]]
+
+// LINUX-DAG: @_ZTH1a = alias void (), void ()* @__tls_init
+// DARWIN-DAG: @_ZTH1a = internal alias void (), void ()* @__tls_init
+// CHECK-DAG: @_ZTHL1d = internal alias void (), void ()* @__tls_init
+// LINUX-DAG: @_ZTHN1U1mE = alias void (), void ()* @__tls_init
+// DARWIN-DAG: @_ZTHN1U1mE = internal alias void (), void ()* @__tls_init
+// CHECK-DAG: @_ZTHN1VIiE1mE = linkonce_odr alias void (), void ()* @[[V_M_INIT:[^, ]*]]
+// CHECK-NOT: @_ZTHN1WIiE1mE =
+// CHECK-DAG: @_ZTHN1XIiE1mE = linkonce_odr alias void (), void ()* @[[X_M_INIT:[^, ]*]]
+// CHECK-DAG: @_ZTHN1VIfE1mE = weak_odr alias void (), void ()* @[[VF_M_INIT:[^, ]*]]
+// CHECK-NOT: @_ZTHN1WIfE1mE =
+// CHECK-DAG: @_ZTHN1XIfE1mE = weak_odr alias void (), void ()* @[[XF_M_INIT:[^, ]*]]
 
 
 // Individual variable initialization functions:
@@ -118,7 +145,9 @@ int f() {
 // LINUX: call i32* @_ZTWN1VIiE1mE()
 // DARWIN: call cxx_fast_tlscc i32* @_ZTWN1VIiE1mE()
 // CHECK-NEXT: load i32, i32* %{{.*}}, align 4
-// CHECK-NEXT: store i32 %{{.*}}, i32* @e, align 4
+// LINUX: call {{.*}}* @_ZTWN1XIiE1mE()
+// DARWIN: call cxx_fast_tlscc {{.*}}* @_ZTWN1XIiE1mE()
+// CHECK: store {{.*}} @e
 
 // LINUX-LABEL: define weak_odr hidden i32* @_ZTWN1VIiE1mE()
 // DARWIN-LABEL: define weak_odr hidden cxx_fast_tlscc i32* @_ZTWN1VIiE1mE()
@@ -126,6 +155,64 @@ int f() {
 // DARWIN: call cxx_fast_tlscc void @_ZTHN1VIiE1mE()
 // CHECK: ret i32* @_ZN1VIiE1mE
 
+// LINUX-LABEL: define weak_odr hidden i32* @_ZTWN1WIiE1mE()
+// DARWIN-LABEL: define weak_odr hidden cxx_fast_tlscc i32* @_ZTWN1WIiE1mE()
+// CHECK-NOT: call
+// CHECK: ret i32* @_ZN1WIiE1mE
+
+// LINUX-LABEL: define weak_odr hidden {{.*}}* @_ZTWN1XIiE1mE()
+// DARWIN-LABEL: define weak_odr hidden cxx_fast_tlscc {{.*}}* @_ZTWN1XIiE1mE()
+// LINUX: call void @_ZTHN1XIiE1mE()
+// DARWIN: call cxx_fast_tlscc void @_ZTHN1XIiE1mE()
+// CHECK: ret {{.*}}* @_ZN1XIiE1mE
+
+// CHECK: define internal {{.*}} @[[VF_M_INIT]]()
+// LINUX-SAME: comdat($_ZN1VIfE1mE)
+// DARWIN-NOT: comdat
+// CHECK: load i8, i8* bitcast (i64* @_ZGVN1VIfE1mE to i8*)
+// CHECK: %[[VF_M_INITIALIZED:.*]] = icmp eq i8 %{{.*}}, 0
+// CHECK: br i1 %[[VF_M_INITIALIZED]],
+// need init:
+// CHECK: call i32 @_Z1gv()
+// CHECK: store i32 %{{.*}}, i32* @_ZN1VIfE1mE, align 4
+// CHECK: store i64 1, i64* @_ZGVN1VIfE1mE
+// CHECK: br label
+
+// CHECK: define internal {{.*}} @[[XF_M_INIT]]()
+// LINUX-SAME: comdat($_ZN1XIfE1mE)
+// DARWIN-NOT: comdat
+// CHECK: load i8, i8* bitcast (i64* @_ZGVN1XIfE1mE to i8*)
+// CHECK: %[[XF_M_INITIALIZED:.*]] = icmp eq i8 %{{.*}}, 0
+// CHECK: br i1 %[[XF_M_INITIALIZED]],
+// need init:
+// LINUX: call {{.*}}__cxa_thread_atexit
+// DARWIN: call {{.*}}_tlv_atexit
+// CHECK: store i64 1, i64* @_ZGVN1XIfE1mE
+// CHECK: br label
+
+// LINUX: declare i32 @__cxa_thread_atexit(void (i8*)*, i8*, i8*)
+// DARWIN: declare i32 @_tlv_atexit(void (i8*)*, i8*, i8*)
+
+// DARWIN: declare cxx_fast_tlscc i32* @_ZTWN1VIcE1mE()
+// LINUX: define weak_odr hidden i32* @_ZTWN1VIcE1mE()
+// LINUX-NOT: comdat
+// LINUX: br i1 icmp ne (void ()* @_ZTHN1VIcE1mE,
+// LINUX: call void @_ZTHN1VIcE1mE()
+// LINUX: ret i32* @_ZN1VIcE1mE
+
+// DARWIN: declare cxx_fast_tlscc i32* @_ZTWN1WIcE1mE()
+// LINUX: define weak_odr hidden i32* @_ZTWN1WIcE1mE()
+// LINUX-NOT: comdat
+// LINUX: br i1 icmp ne (void ()* @_ZTHN1WIcE1mE,
+// LINUX: call void @_ZTHN1WIcE1mE()
+// LINUX: ret i32* @_ZN1WIcE1mE
+
+// DARWIN: declare cxx_fast_tlscc {{.*}}* @_ZTWN1XIcE1mE()
+// LINUX: define weak_odr hidden {{.*}}* @_ZTWN1XIcE1mE()
+// LINUX-NOT: comdat
+// LINUX: br i1 icmp ne (void ()* @_ZTHN1XIcE1mE,
+// LINUX: call void @_ZTHN1XIcE1mE()
+// LINUX: ret {{.*}}* @_ZN1XIcE1mE
 
 struct S { S(); ~S(); };
 struct T { ~T(); };
@@ -154,9 +241,6 @@ void tls_dtor() {
   static thread_local const S &u = S();
 }
 
-// LINUX: declare i32 @__cxa_thread_atexit(void (i8*)*, i8*, i8*)
-// DARWIN: declare i32 @_tlv_atexit(void (i8*)*, i8*, i8*)
-
 // CHECK: define {{.*}} @_Z7PR15991v(
 int PR15991() {
   thread_local int n;
@@ -184,7 +268,9 @@ void set_anon_i() {
 // LINUX-LABEL: define internal i32* @_ZTWN12_GLOBAL__N_16anon_iE()
 // DARWIN-LABEL: define internal cxx_fast_tlscc i32* @_ZTWN12_GLOBAL__N_16anon_iE()
 
-// CHECK: define {{.*}} @[[V_M_INIT:.*]]()
+// CHECK: define internal {{.*}} @[[V_M_INIT]]()
+// LINUX-SAME: comdat($_ZN1VIiE1mE)
+// DARWIN-NOT: comdat
 // CHECK: load i8, i8* bitcast (i64* @_ZGVN1VIiE1mE to i8*)
 // CHECK: %[[V_M_INITIALIZED:.*]] = icmp eq i8 %{{.*}}, 0
 // CHECK: br i1 %[[V_M_INITIALIZED]],
@@ -194,6 +280,18 @@ void set_anon_i() {
 // CHECK: store i64 1, i64* @_ZGVN1VIiE1mE
 // CHECK: br label
 
+// CHECK: define internal {{.*}} @[[X_M_INIT]]()
+// LINUX-SAME: comdat($_ZN1XIiE1mE)
+// DARWIN-NOT: comdat
+// CHECK: load i8, i8* bitcast (i64* @_ZGVN1XIiE1mE to i8*)
+// CHECK: %[[X_M_INITIALIZED:.*]] = icmp eq i8 %{{.*}}, 0
+// CHECK: br i1 %[[X_M_INITIALIZED]],
+// need init:
+// LINUX: call {{.*}}__cxa_thread_atexit
+// DARWIN: call {{.*}}_tlv_atexit
+// CHECK: store i64 1, i64* @_ZGVN1XIiE1mE
+// CHECK: br label
+
 // CHECK: define {{.*}}@[[GLOBAL_INIT:.*]]()
 // CHECK: call void @[[C_INIT]]()
 // CHECK: call void @[[E_INIT]]()
@@ -205,10 +303,13 @@ void set_anon_i() {
 // CHECK: br i1 %[[NEED_TLS_INIT]],
 // init:
 // CHECK: store i8 1, i8* @__tls_guard
+// CHECK-NOT: call void @[[V_M_INIT]]()
 // CHECK: call void @[[A_INIT]]()
+// CHECK-NOT: call void @[[V_M_INIT]]()
 // CHECK: call void @[[D_INIT]]()
+// CHECK-NOT: call void @[[V_M_INIT]]()
 // CHECK: call void @[[U_M_INIT]]()
-// CHECK: call void @[[V_M_INIT]]()
+// CHECK-NOT: call void @[[V_M_INIT]]()
 
 
 // LIUNX: define weak_odr hidden i32* @_ZTW1a() {
diff --git a/clang/test/OpenMP/threadprivate_codegen.cpp b/clang/test/OpenMP/threadprivate_codegen.cpp
index 09f5ed5060ba3..318415761ac77 100644
--- a/clang/test/OpenMP/threadprivate_codegen.cpp
+++ b/clang/test/OpenMP/threadprivate_codegen.cpp
@@ -179,9 +179,9 @@ struct S5 {
 // CHECK-TLS-DAG:  @__dso_handle = external global i8
 // CHECK-TLS-DAG:  [[GS1_TLS_INIT:@_ZTHL3gs1]] = internal alias void (), void ()* @__tls_init
 // CHECK-TLS-DAG:  [[ARR_X_TLS_INIT:@_ZTH5arr_x]] = alias void (), void ()* @__tls_init
-// CHECK-TLS-DAG:  [[ST_INT_ST_TLS_INIT:@_ZTHN2STIiE2stE]] = linkonce_odr alias void (), void ()* @__tls_init
-// CHECK-TLS-DAG:  [[ST_FLOAT_ST_TLS_INIT:@_ZTHN2STIfE2stE]] = linkonce_odr alias void (), void ()* @__tls_init
-// CHECK-TLS-DAG:  [[ST_S4_ST_TLS_INIT:@_ZTHN2STI2S4E2stE]] = linkonce_odr alias void (), void ()* @__tls_init
+
+
+// CHECK-TLS-DAG:  [[ST_S4_ST_TLS_INIT:@_ZTHN2STI2S4E2stE]] = linkonce_odr alias void (), void ()* [[ST_S4_ST_CXX_INIT:@[^, ]*]]
 
 struct Static {
   static S3 s;
@@ -640,11 +640,11 @@ int main() {
 // CHECK-TLS:   ret [2 x [3 x [[S1]]]]* [[ARR_X]]
 // CHECK-TLS: }
 // CHECK-TLS: define {{.*}} i32* [[ST_INT_ST_TLS_INITD]] {{#[0-9]+}} {
-// CHECK-TLS:   call void [[ST_INT_ST_TLS_INIT]]
+// CHECK-TLS-NOT:   call
 // CHECK-TLS:   ret i32* [[ST_INT_ST]]
 // CHECK-TLS: }
 // CHECK-TLS: define {{.*}} float* [[ST_FLOAT_ST_TLS_INITD]] {{#[0-9]+}} {
-// CHECK-TLS:   call void [[ST_FLOAT_ST_TLS_INIT]]
+// CHECK-TLS-NOT:   call
 // CHECK-TLS:   ret float* [[ST_FLOAT_ST]]
 // CHECK-TLS: }
 // CHECK-TLS: define {{.*}} [[S4]]* [[ST_S4_ST_TLS_INITD]] {{#[0-9]+}} {
@@ -923,7 +923,7 @@ int foobar() {
 // CHECK-TLS: define {{.*}}void [[SM_CTOR2]]([[SMAIN]]* {{.*}}, i32 {{.*}})
 // CHECK-TLS: define {{.*}}void [[SM_DTOR2]]([[SMAIN]]* {{.*}})
 
-// CHECK-TLS: define internal void [[ST_S4_ST_CXX_INIT:@.*]]()
+// CHECK-TLS: define internal void [[ST_S4_ST_CXX_INIT]]()
 // CHECK-TLS: call void [[ST_S4_ST_CTOR1:@.*]]([[S4]]* [[ST_S4_ST]], i32 23)
 // CHECK-TLS: call i32 @__cxa_thread_atexit(void (i8*)* bitcast (void ([[S4]]*)* [[ST_S4_ST_DTOR1:.*]] to void (i8*)*), i8* bitcast ([[S4]]* [[ST_S4_ST]] to i8*)
 // CHECK-TLS: }
@@ -945,7 +945,7 @@ int foobar() {
 // CHECK-TLS:      call void [[GS1_CXX_INIT]]
 // CHECK-TLS-NOT:  call void [[GS2_CXX_INIT]]
 // CHECK-TLS:      call void [[ARR_X_CXX_INIT]]
-// CHECK-TLS:      call void [[ST_S4_ST_CXX_INIT]]
+// CHECK-TLS-NOT:  call void [[ST_S4_ST_CXX_INIT]]
 // CHECK-TLS:      [[DONE_LABEL]]
 
 // CHECK-TLS-DAG:      declare {{.*}} void [[GS3_TLS_INIT]]