Implement CodeGen for C++11 thread_local, following the Itanium ABI specification as discussed on cxx-abi-dev.

llvm-svn: 179858
This commit is contained in:
Richard Smith
2013-04-19 16:42:07 +00:00
parent 6ac9d13e61
commit 2fd1d7aee3
12 changed files with 425 additions and 13 deletions

View File

@@ -141,6 +141,16 @@ public:
raw_ostream &) {
llvm_unreachable("Target does not support mangling guard variables");
}
// FIXME: Revisit this once we know what we need to do for MSVC compatibility.
virtual void mangleItaniumThreadLocalInit(const VarDecl *D,
raw_ostream &) {
llvm_unreachable("Target does not support mangling thread_local variables");
}
virtual void mangleItaniumThreadLocalWrapper(const VarDecl *D,
raw_ostream &) {
llvm_unreachable("Target does not support mangling thread_local variables");
}
/// @}
};

View File

@@ -141,6 +141,8 @@ public:
raw_ostream &);
void mangleItaniumGuardVariable(const VarDecl *D, raw_ostream &);
void mangleItaniumThreadLocalInit(const VarDecl *D, raw_ostream &);
void mangleItaniumThreadLocalWrapper(const VarDecl *D, raw_ostream &);
void mangleInitDiscriminator() {
Discriminator = 0;
@@ -3531,6 +3533,22 @@ void ItaniumMangleContext::mangleItaniumGuardVariable(const VarDecl *D,
Mangler.mangleName(D);
}
void ItaniumMangleContext::mangleItaniumThreadLocalInit(const VarDecl *D,
raw_ostream &Out) {
// <special-name> ::= TH <object name>
CXXNameMangler Mangler(*this, Out);
Mangler.getStream() << "_ZTH";
Mangler.mangleName(D);
}
void ItaniumMangleContext::mangleItaniumThreadLocalWrapper(const VarDecl *D,
raw_ostream &Out) {
// <special-name> ::= TW <object name>
CXXNameMangler Mangler(*this, Out);
Mangler.getStream() << "_ZTW";
Mangler.mangleName(D);
}
void ItaniumMangleContext::mangleReferenceTemporary(const VarDecl *D,
raw_ostream &Out) {
// We match the GCC mangling here.

View File

@@ -259,3 +259,14 @@ llvm::BasicBlock *CGCXXABI::EmitCtorCompleteObjectHandler(
ErrorUnsupportedABI(CGF, "complete object detection in ctor");
return 0;
}
void CGCXXABI::EmitThreadLocalInitFuncs(
llvm::ArrayRef<std::pair<const VarDecl *, llvm::GlobalVariable *> > Decls,
llvm::Function *InitFunc) {
}
LValue CGCXXABI::EmitThreadLocalDeclRefExpr(CodeGenFunction &CGF,
const DeclRefExpr *DRE) {
ErrorUnsupportedABI(CGF, "odr-use of thread_local global");
return LValue();
}

View File

@@ -353,6 +353,25 @@ public:
/// \param addr - a pointer to pass to the destructor function.
virtual void registerGlobalDtor(CodeGenFunction &CGF, const VarDecl &D,
llvm::Constant *dtor, llvm::Constant *addr);
/*************************** thread_local initialization ********************/
/// Emits ABI-required functions necessary to initialize thread_local
/// variables in this translation unit.
///
/// \param Decls The thread_local declarations in this translation unit.
/// \param InitFunc If this translation unit contains any non-constant
/// initialization or non-trivial destruction for thread_local
/// variables, a function to perform the initialization. Otherwise, 0.
virtual void EmitThreadLocalInitFuncs(
llvm::ArrayRef<std::pair<const VarDecl *, llvm::GlobalVariable *> > Decls,
llvm::Function *InitFunc);
/// Emit a reference to a non-local thread_local variable (including
/// triggering the initialization of all thread_local variables in its
/// translation unit).
virtual LValue EmitThreadLocalDeclRefExpr(CodeGenFunction &CGF,
const DeclRefExpr *DRE);
};
// Create an instance of a C++ ABI class:

View File

@@ -156,7 +156,8 @@ void CodeGenFunction::EmitCXXGlobalVarDeclInit(const VarDecl &D,
static llvm::Function *
CreateGlobalInitOrDestructFunction(CodeGenModule &CGM,
llvm::FunctionType *ty,
const Twine &name);
const Twine &name,
bool TLS = false);
/// Create a stub function, suitable for being passed to atexit,
/// which passes the given address to the given destructor function.
@@ -225,11 +226,11 @@ void CodeGenFunction::EmitCXXGuardedInit(const VarDecl &D,
static llvm::Function *
CreateGlobalInitOrDestructFunction(CodeGenModule &CGM,
llvm::FunctionType *FTy,
const Twine &Name) {
const Twine &Name, bool TLS) {
llvm::Function *Fn =
llvm::Function::Create(FTy, llvm::GlobalValue::InternalLinkage,
Name, &CGM.getModule());
if (!CGM.getLangOpts().AppleKext) {
if (!CGM.getLangOpts().AppleKext && !TLS) {
// Set the section if needed.
if (const char *Section =
CGM.getTarget().getStaticInitSectionSpecifier())
@@ -255,9 +256,6 @@ void
CodeGenModule::EmitCXXGlobalVarDeclInitFunc(const VarDecl *D,
llvm::GlobalVariable *Addr,
bool PerformInit) {
if (D->getTLSKind())
ErrorUnsupported(D->getInit(), "dynamic TLS initialization");
llvm::FunctionType *FTy = llvm::FunctionType::get(VoidTy, false);
// Create a variable initialization function.
@@ -267,12 +265,20 @@ CodeGenModule::EmitCXXGlobalVarDeclInitFunc(const VarDecl *D,
CodeGenFunction(*this).GenerateCXXGlobalVarDeclInitFunc(Fn, D, Addr,
PerformInit);
if (D->hasAttr<InitPriorityAttr>()) {
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);
} else if (D->hasAttr<InitPriorityAttr>()) {
unsigned int order = D->getAttr<InitPriorityAttr>()->getPriority();
OrderGlobalInits Key(order, PrioritizedCXXGlobalInits.size());
PrioritizedCXXGlobalInits.push_back(std::make_pair(Key, Fn));
DelayedCXXInitPosition.erase(D);
} else {
} else {
llvm::DenseMap<const Decl *, unsigned>::iterator I =
DelayedCXXInitPosition.find(D);
if (I == DelayedCXXInitPosition.end()) {
@@ -285,6 +291,27 @@ CodeGenModule::EmitCXXGlobalVarDeclInitFunc(const VarDecl *D,
}
}
void CodeGenModule::EmitCXXThreadLocalInitFunc() {
llvm::Function *InitFn = 0;
if (!CXXThreadLocalInits.empty()) {
// Generate a guarded initialization function.
llvm::FunctionType *FTy = llvm::FunctionType::get(VoidTy, false);
InitFn = CreateGlobalInitOrDestructFunction(
*this, FTy, "__tls_init", /*TLS*/ true);
llvm::GlobalVariable *Guard = new llvm::GlobalVariable(
getModule(), Int8Ty, false, llvm::GlobalVariable::InternalLinkage,
llvm::ConstantInt::get(Int8Ty, 0), "__tls_guard");
Guard->setThreadLocal(true);
CodeGenFunction(*this).GenerateCXXGlobalInitFunc(
InitFn, CXXThreadLocalInits.data(), CXXThreadLocalInits.size(), Guard);
}
getCXXABI().EmitThreadLocalInitFuncs(CXXThreadLocals, InitFn);
CXXThreadLocalInits.clear();
CXXThreadLocals.clear();
}
void
CodeGenModule::EmitCXXGlobalInitFunc() {
while (!CXXGlobalInits.empty() && !CXXGlobalInits.back())
@@ -385,7 +412,8 @@ void CodeGenFunction::GenerateCXXGlobalVarDeclInitFunc(llvm::Function *Fn,
void CodeGenFunction::GenerateCXXGlobalInitFunc(llvm::Function *Fn,
llvm::Constant **Decls,
unsigned NumDecls) {
unsigned NumDecls,
llvm::GlobalVariable *Guard) {
// Initialize debug info if needed.
maybeInitializeDebugInfo();
@@ -393,6 +421,22 @@ void CodeGenFunction::GenerateCXXGlobalInitFunc(llvm::Function *Fn,
getTypes().arrangeNullaryFunction(),
FunctionArgList(), SourceLocation());
llvm::BasicBlock *ExitBlock = 0;
if (Guard) {
// If we have a guard variable, check whether we've already performed these
// initializations. This happens for TLS initialization functions.
llvm::Value *GuardVal = Builder.CreateLoad(Guard);
llvm::Value *Uninit = Builder.CreateIsNull(GuardVal, "guard.uninitialized");
// Mark as initialized before initializing anything else. If the
// initializers use previously-initialized thread_local vars, that's
// probably supposed to be OK, but the standard doesn't say.
Builder.CreateStore(llvm::ConstantInt::get(GuardVal->getType(), 1), Guard);
llvm::BasicBlock *InitBlock = createBasicBlock("init");
ExitBlock = createBasicBlock("exit");
Builder.CreateCondBr(Uninit, InitBlock, ExitBlock);
EmitBlock(InitBlock);
}
RunCleanupsScope Scope(*this);
// When building in Objective-C++ ARC mode, create an autorelease pool
@@ -407,7 +451,12 @@ void CodeGenFunction::GenerateCXXGlobalInitFunc(llvm::Function *Fn,
EmitRuntimeCall(Decls[i]);
Scope.ForceCleanup();
if (ExitBlock) {
Builder.CreateBr(ExitBlock);
EmitBlock(ExitBlock);
}
FinishFunction();
}

View File

@@ -1824,8 +1824,12 @@ LValue CodeGenFunction::EmitDeclRefLValue(const DeclRefExpr *E) {
if (const VarDecl *VD = dyn_cast<VarDecl>(ND)) {
// Check if this is a global variable.
if (VD->hasLinkage() || VD->isStaticDataMember())
if (VD->hasLinkage() || VD->isStaticDataMember()) {
// If it's thread_local, emit a call to its wrapper function instead.
if (VD->getTLSKind() == VarDecl::TLS_Dynamic)
return CGM.getCXXABI().EmitThreadLocalDeclRefExpr(*this, E);
return EmitGlobalVarDeclLValue(*this, E, VD);
}
bool isBlockVariable = VD->hasAttr<BlocksAttr>();

View File

@@ -2628,7 +2628,8 @@ public:
/// variables.
void GenerateCXXGlobalInitFunc(llvm::Function *Fn,
llvm::Constant **Decls,
unsigned NumDecls);
unsigned NumDecls,
llvm::GlobalVariable *Guard = 0);
/// GenerateCXXGlobalDtorsFunc - Generates code for destroying global
/// variables.

View File

@@ -176,6 +176,7 @@ void CodeGenModule::Release() {
EmitDeferred();
EmitCXXGlobalInitFunc();
EmitCXXGlobalDtorFunc();
EmitCXXThreadLocalInitFunc();
if (ObjCRuntime)
if (llvm::Function *ObjCInitFunction = ObjCRuntime->ModuleInitFunction())
AddGlobalCtor(ObjCInitFunction);
@@ -1477,8 +1478,10 @@ CodeGenModule::GetOrCreateLLVMGlobal(StringRef MangledName,
GV->setVisibility(GetLLVMVisibility(LV.getVisibility()));
}
if (D->getTLSKind())
if (D->getTLSKind()) {
CXXThreadLocals.push_back(std::make_pair(D, GV));
setTLSMode(GV, *D);
}
}
if (AddrSpace != Ty->getAddressSpace())

View File

@@ -317,6 +317,14 @@ class CodeGenModule : public CodeGenTypeCache {
llvm::GlobalValue *> StaticExternCMap;
StaticExternCMap StaticExternCValues;
/// \brief thread_local variables defined or used in this TU.
std::vector<std::pair<const VarDecl *, llvm::GlobalVariable *> >
CXXThreadLocals;
/// \brief thread_local variables with initializers that need to run
/// before any thread_local variable in this TU is odr-used.
std::vector<llvm::Constant*> CXXThreadLocalInits;
/// CXXGlobalInits - Global variables with initializers that need to run
/// before main.
std::vector<llvm::Constant*> CXXGlobalInits;
@@ -1026,6 +1034,9 @@ private:
/// a C++ destructor Decl.
void EmitCXXDestructor(const CXXDestructorDecl *D, CXXDtorType Type);
/// \brief Emit the function that initializes C++ thread_local variables.
void EmitCXXThreadLocalInitFunc();
/// EmitCXXGlobalInitFunc - Emit the function that initializes C++ globals.
void EmitCXXGlobalInitFunc();

View File

@@ -146,6 +146,14 @@ public:
llvm::GlobalVariable *DeclPtr, bool PerformInit);
void registerGlobalDtor(CodeGenFunction &CGF, const VarDecl &D,
llvm::Constant *dtor, llvm::Constant *addr);
llvm::Function *getOrCreateThreadLocalWrapper(const VarDecl *VD,
llvm::GlobalVariable *Var);
void EmitThreadLocalInitFuncs(
llvm::ArrayRef<std::pair<const VarDecl *, llvm::GlobalVariable *> > Decls,
llvm::Function *InitFunc);
LValue EmitThreadLocalDeclRefExpr(CodeGenFunction &CGF,
const DeclRefExpr *DRE);
};
class ARMCXXABI : public ItaniumCXXABI {
@@ -1249,3 +1257,138 @@ void ItaniumCXXABI::registerGlobalDtor(CodeGenFunction &CGF,
CGF.registerGlobalDtorWithAtExit(dtor, addr);
}
/// Get the appropriate linkage for the wrapper function. This is essentially
/// the weak form of the variable's linkage; every translation unit which wneeds
/// the wrapper emits a copy, and we want the linker to merge them.
static llvm::GlobalValue::LinkageTypes getThreadLocalWrapperLinkage(
llvm::GlobalValue::LinkageTypes VarLinkage) {
if (llvm::GlobalValue::isLinkerPrivateLinkage(VarLinkage))
return llvm::GlobalValue::LinkerPrivateWeakLinkage;
// For internal linkage variables, we don't need an external or weak wrapper.
if (llvm::GlobalValue::isLocalLinkage(VarLinkage))
return VarLinkage;
return llvm::GlobalValue::WeakODRLinkage;
}
llvm::Function *
ItaniumCXXABI::getOrCreateThreadLocalWrapper(const VarDecl *VD,
llvm::GlobalVariable *Var) {
// Mangle the name for the thread_local wrapper function.
SmallString<256> WrapperName;
{
llvm::raw_svector_ostream Out(WrapperName);
getMangleContext().mangleItaniumThreadLocalWrapper(VD, Out);
Out.flush();
}
if (llvm::Value *V = Var->getParent()->getNamedValue(WrapperName))
return cast<llvm::Function>(V);
llvm::Type *RetTy = Var->getType();
if (VD->getType()->isReferenceType())
RetTy = RetTy->getPointerElementType();
llvm::FunctionType *FnTy = llvm::FunctionType::get(RetTy, false);
llvm::Function *Wrapper = llvm::Function::Create(
FnTy, getThreadLocalWrapperLinkage(Var->getLinkage()), WrapperName.str(),
&CGM.getModule());
// Always resolve references to the wrapper at link time.
Wrapper->setVisibility(llvm::GlobalValue::HiddenVisibility);
return Wrapper;
}
void ItaniumCXXABI::EmitThreadLocalInitFuncs(
llvm::ArrayRef<std::pair<const VarDecl *, llvm::GlobalVariable *> > Decls,
llvm::Function *InitFunc) {
for (unsigned I = 0, N = Decls.size(); I != N; ++I) {
const VarDecl *VD = Decls[I].first;
llvm::GlobalVariable *Var = Decls[I].second;
// Mangle the name for the thread_local initialization function.
SmallString<256> InitFnName;
{
llvm::raw_svector_ostream Out(InitFnName);
getMangleContext().mangleItaniumThreadLocalInit(VD, Out);
Out.flush();
}
// If we have a definition for the variable, emit the initialization
// function as an alias to the global Init function (if any). Otherwise,
// produce a declaration of the initialization function.
llvm::GlobalValue *Init = 0;
bool InitIsInitFunc = false;
if (VD->hasDefinition()) {
InitIsInitFunc = true;
if (InitFunc)
Init =
new llvm::GlobalAlias(InitFunc->getType(), Var->getLinkage(),
InitFnName.str(), InitFunc, &CGM.getModule());
} 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());
}
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(Entry);
if (InitIsInitFunc) {
if (Init)
Builder.CreateCall(Init);
} else {
// Don't know whether we have an init function. Call it if it exists.
llvm::Value *Have = Builder.CreateIsNotNull(Init);
llvm::BasicBlock *InitBB = llvm::BasicBlock::Create(Context, "", Wrapper);
llvm::BasicBlock *ExitBB = llvm::BasicBlock::Create(Context, "", Wrapper);
Builder.CreateCondBr(Have, InitBB, ExitBB);
Builder.SetInsertPoint(InitBB);
Builder.CreateCall(Init);
Builder.CreateBr(ExitBB);
Builder.SetInsertPoint(ExitBB);
}
// For a reference, the result of the wrapper function is a pointer to
// the referenced object.
llvm::Value *Val = Var;
if (VD->getType()->isReferenceType()) {
llvm::LoadInst *LI = Builder.CreateLoad(Val);
LI->setAlignment(CGM.getContext().getDeclAlign(VD).getQuantity());
Val = LI;
}
Builder.CreateRet(Val);
}
}
LValue ItaniumCXXABI::EmitThreadLocalDeclRefExpr(CodeGenFunction &CGF,
const DeclRefExpr *DRE) {
const VarDecl *VD = cast<VarDecl>(DRE->getDecl());
QualType T = VD->getType();
llvm::Type *Ty = CGF.getTypes().ConvertTypeForMem(T);
llvm::Value *Val = CGF.CGM.GetAddrOfGlobalVar(VD, Ty);
llvm::Function *Wrapper =
getOrCreateThreadLocalWrapper(VD, cast<llvm::GlobalVariable>(Val));
Val = CGF.Builder.CreateCall(Wrapper);
LValue LV;
if (VD->getType()->isReferenceType())
LV = CGF.MakeNaturalAlignAddrLValue(Val, T);
else
LV = CGF.MakeAddrLValue(Val, DRE->getType(),
CGF.getContext().getDeclAlign(VD));
// FIXME: need setObjCGCLValueClass?
return LV;
}

View File

@@ -0,0 +1,26 @@
// RUN: %clang_cc1 -std=c++11 -emit-llvm %s -o - -triple x86_64-linux-gnu | FileCheck %s
int &f();
// CHECK: @r = thread_local global i32* null
thread_local int &r = f();
// CHECK: @_ZTH1r = alias void ()* @__tls_init
int &g() { return r; }
// CHECK: define {{.*}} @[[R_INIT:.*]]()
// CHECK: call i32* @_Z1fv()
// CHECK: store i32* %{{.*}}, i32** @r, align 8
// CHECK: define i32* @_Z1gv()
// CHECK: call i32* @_ZTW1r()
// CHECK: ret i32* %{{.*}}
// CHECK: define weak_odr hidden i32* @_ZTW1r() {
// CHECK: call void @_ZTH1r()
// CHECK: load i32** @r, align 8
// CHECK: ret i32* %{{.*}}
// CHECK: define internal void @__tls_init()
// CHECK: call void @[[R_INIT]]()

View File

@@ -1,8 +1,30 @@
// RUN: %clang_cc1 -std=c++11 -emit-llvm %s -o - -triple x86_64-linux-gnu | FileCheck %s
int f();
int g();
// CHECK: @a = thread_local global i32 0
thread_local int a = f();
extern thread_local int b;
// CHECK: @c = global i32 0
int c = b;
// CHECK: @_ZL1d = internal thread_local global i32 0
static thread_local int d = g();
struct U { static thread_local int m; };
// CHECK: @_ZN1U1mE = thread_local global i32 0
thread_local int U::m = f();
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;
// CHECK: @_ZN1VIiE1mE = weak_odr thread_local global i32 0
// CHECK: @_ZZ1fvE1n = internal thread_local global i32 0
// CHECK: @_ZGVZ1fvE1n = internal thread_local global i8 0
// CHECK: @_ZZ8tls_dtorvE1s = internal thread_local global
@@ -15,6 +37,24 @@ int g();
// CHECK: @_ZGVZ8tls_dtorvE1u = internal thread_local global i8 0
// CHECK: @_ZGRZ8tls_dtorvE1u = internal thread_local global
// CHECK: @_ZGVN1VIiE1mE = weak_odr thread_local global i64 0
// CHECK: @__tls_guard = internal thread_local global i8 0
// CHECK: @llvm.global_ctors = appending global {{.*}} @[[GLOBAL_INIT:[^ ]*]]
// CHECK: @_ZTH1a = alias void ()* @__tls_init
// CHECK: @_ZTHL1d = alias internal void ()* @__tls_init
// CHECK: @_ZTHN1U1mE = alias void ()* @__tls_init
// CHECK: @_ZTHN1VIiE1mE = alias weak_odr void ()* @__tls_init
// Individual variable initialization functions:
// CHECK: define {{.*}} @[[A_INIT:.*]]()
// CHECK: call i32 @_Z1fv()
// CHECK-NEXT: store i32 {{.*}}, i32* @a, align 4
// CHECK: define i32 @_Z1fv()
int f() {
// CHECK: %[[GUARD:.*]] = load i8* @_ZGVZ1fvE1n, align 1
@@ -31,6 +71,37 @@ int f() {
return n;
}
// CHECK: define {{.*}} @[[C_INIT:.*]]()
// CHECK: call i32* @_ZTW1b()
// CHECK-NEXT: load i32* %{{.*}}, align 4
// CHECK-NEXT: store i32 %{{.*}}, i32* @c, align 4
// CHECK: define weak_odr hidden i32* @_ZTW1b()
// CHECK: br i1 icmp ne (void ()* @_ZTH1b, void ()* null),
// not null:
// CHECK: call void @_ZTH1b()
// CHECK: br label
// finally:
// CHECK: ret i32* @b
// CHECK: define {{.*}} @[[D_INIT:.*]]()
// CHECK: call i32 @_Z1gv()
// CHECK-NEXT: store i32 %{{.*}}, i32* @_ZL1d, align 4
// CHECK: define {{.*}} @[[U_M_INIT:.*]]()
// CHECK: call i32 @_Z1fv()
// CHECK-NEXT: store i32 %{{.*}}, i32* @_ZN1U1mE, align 4
// CHECK: define {{.*}} @[[E_INIT:.*]]()
// CHECK: call i32* @_ZTWN1VIiE1mE()
// CHECK-NEXT: load i32* %{{.*}}, align 4
// CHECK-NEXT: store i32 %{{.*}}, i32* @e, align 4
// CHECK: define weak_odr hidden i32* @_ZTWN1VIiE1mE()
// CHECK: call void @_ZTHN1VIiE1mE()
// CHECK: ret i32* @_ZN1VIiE1mE
struct S { S(); ~S(); };
struct T { ~T(); };
@@ -54,3 +125,49 @@ void tls_dtor() {
// CHECK: store i8 1, i8* @_ZGVZ8tls_dtorvE1u
static thread_local const S &u = S();
}
// CHECK: declare i32 @__cxa_thread_atexit(void (i8*)*, i8*, i8*)
// CHECK: define {{.*}} @[[V_M_INIT:.*]]()
// CHECK: load i8* bitcast (i64* @_ZGVN1VIiE1mE to i8*)
// CHECK: %[[V_M_INITIALIZED:.*]] = icmp eq i8 %{{.*}}, 0
// CHECK: br i1 %[[V_M_INITIALIZED]],
// need init:
// CHECK: call i32 @_Z1gv()
// CHECK: store i32 %{{.*}}, i32* @_ZN1VIiE1mE, align 4
// CHECK: store i64 1, i64* @_ZGVN1VIiE1mE
// CHECK: br label
// CHECK: define {{.*}}@[[GLOBAL_INIT:.*]]()
// CHECK: call void @[[C_INIT]]()
// CHECK: call void @[[E_INIT]]()
// CHECK: define {{.*}}@__tls_init()
// CHECK: load i8* @__tls_guard
// CHECK: %[[NEED_TLS_INIT:.*]] = icmp eq i8 %{{.*}}, 0
// CHECK: store i8 1, i8* @__tls_guard
// CHECK: br i1 %[[NEED_TLS_INIT]],
// init:
// CHECK: call void @[[A_INIT]]()
// CHECK: call void @[[D_INIT]]()
// CHECK: call void @[[U_M_INIT]]()
// CHECK: call void @[[V_M_INIT]]()
// CHECK: define weak_odr hidden i32* @_ZTW1a() {
// CHECK: call void @_ZTH1a()
// CHECK: ret i32* @a
// CHECK: }
// CHECK: declare extern_weak void @_ZTH1b()
// CHECK: define internal hidden i32* @_ZTWL1d()
// CHECK: call void @_ZTHL1d()
// CHECK: ret i32* @_ZL1d
// CHECK: define weak_odr hidden i32* @_ZTWN1U1mE()
// CHECK: call void @_ZTHN1U1mE()
// CHECK: ret i32* @_ZN1U1mE