[CodeGen][TLS] Set TLS Model for __tls_guard as well.

[junparser]

Summary

For now we do not set tls model for tls_guard with/without option -ftls-model which is suboptimal.
This patch set model from command line for tls_guard as well. This keeps same behavior with gcc.

TestPlan: check-clang

[junparser]

kindly ping~

[junparser]

@rnk @aaron.ballman any comments?

[aaron.ballman]

Changes generally LGTM, but this is not my area of expertise.

clang/lib/CodeGen/ItaniumCXXABI.cpp

2622

Do we need a similar change in MicrosoftCXXABI::EmitGuardedInit()? I notice it sets thread local to true but does not set the thread local mode.

[junparser]

address the comments.

[junparser]

@aaron.ballman thanks for the review. I have updated the patch.

clang/lib/CodeGen/ItaniumCXXABI.cpp

2622

Yes, It should be changed as well.

[aaron.ballman]

LGTM, thanks!

clang/lib/CodeGen/CodeGenModule.h

public:
   void setGVProperties(llvm::GlobalValue *GV, const NamedDecl *D) const;
 
   void setGVPropertiesAux(llvm::GlobalValue *GV, const NamedDecl *D) const;
 
   /// Set the TLS mode for the given LLVM GlobalValue for the thread-local
   /// variable declaration D.
   void setTLSMode(llvm::GlobalValue *GV, const VarDecl &D) const;
 
+  /// Get LLVM TLS mode from CodeGenOptions.
+  llvm::GlobalVariable::ThreadLocalMode GetDefaultLLVMTLSModel() const;
+
   static llvm::GlobalValue::VisibilityTypes GetLLVMVisibility(Visibility V) {
     switch (V) {
     case DefaultVisibility:   return llvm::GlobalValue::DefaultVisibility;
     case HiddenVisibility:    return llvm::GlobalValue::HiddenVisibility;
     case ProtectedVisibility: return llvm::GlobalValue::ProtectedVisibility;
     }
     llvm_unreachable("unknown visibility!");
   }

clang/lib/CodeGen/CodeGenModule.cpp

 static llvm::GlobalVariable::ThreadLocalMode GetLLVMTLSModel(StringRef S) {
   return llvm::StringSwitch<llvm::GlobalVariable::ThreadLocalMode>(S)
       .Case("global-dynamic", llvm::GlobalVariable::GeneralDynamicTLSModel)
       .Case("local-dynamic", llvm::GlobalVariable::LocalDynamicTLSModel)
       .Case("initial-exec", llvm::GlobalVariable::InitialExecTLSModel)
       .Case("local-exec", llvm::GlobalVariable::LocalExecTLSModel);
 }
 
-static llvm::GlobalVariable::ThreadLocalMode GetLLVMTLSModel(
-    CodeGenOptions::TLSModel M) {
-  switch (M) {
+llvm::GlobalVariable::ThreadLocalMode
+CodeGenModule::GetDefaultLLVMTLSModel() const {
+  switch (CodeGenOpts.getDefaultTLSModel()) {
   case CodeGenOptions::GeneralDynamicTLSModel:
     return llvm::GlobalVariable::GeneralDynamicTLSModel;
   case CodeGenOptions::LocalDynamicTLSModel:
     return llvm::GlobalVariable::LocalDynamicTLSModel;
   case CodeGenOptions::InitialExecTLSModel:
     return llvm::GlobalVariable::InitialExecTLSModel;
   case CodeGenOptions::LocalExecTLSModel:
     return llvm::GlobalVariable::LocalExecTLSModel;
   }
   llvm_unreachable("Invalid TLS model!");
 }
 
 void CodeGenModule::setTLSMode(llvm::GlobalValue *GV, const VarDecl &D) const {
   assert(D.getTLSKind() && "setting TLS mode on non-TLS var!");
 
   llvm::GlobalValue::ThreadLocalMode TLM;
-  TLM = GetLLVMTLSModel(CodeGenOpts.getDefaultTLSModel());
+  TLM = GetDefaultLLVMTLSModel();
 
   // Override the TLS model if it is explicitly specified.
   if (const TLSModelAttr *Attr = D.getAttr<TLSModelAttr>()) {
     TLM = GetLLVMTLSModel(Attr->getModel());
   }
 
   GV->setThreadLocalMode(TLM);
 }

clang/lib/CodeGen/ItaniumCXXABI.cpp

if (!OrderedInits.empty()) {
     InitFunc = CGM.CreateGlobalInitOrDestructFunction(FTy, "__tls_init", FI,
                                                       SourceLocation(),
                                                       /*TLS=*/true);
     llvm::GlobalVariable *Guard = new llvm::GlobalVariable(
         CGM.getModule(), CGM.Int8Ty, /*isConstant=*/false,
         llvm::GlobalVariable::InternalLinkage,
         llvm::ConstantInt::get(CGM.Int8Ty, 0), "__tls_guard");
     Guard->setThreadLocal(true);
+    Guard->setThreadLocalMode(CGM.GetDefaultLLVMTLSModel());
 
     CharUnits GuardAlign = CharUnits::One();
     Guard->setAlignment(GuardAlign.getAsAlign());
 
     CodeGenFunction(CGM).GenerateCXXGlobalInitFunc(
         InitFunc, OrderedInits, ConstantAddress(Guard, GuardAlign));
     // On Darwin platforms, use CXX_FAST_TLS calling convention.
     if (CGM.getTarget().getTriple().isOSDarwin()) {

clang/test/CodeGen/tls-model.cpp

 int z2;
 int __thread x;
 int f() {
   static int __thread y;
   return y++;
 }
 int __thread __attribute__((tls_model("initial-exec"))) z;
 
+struct S {
+  S();
+  ~S();
+};
+struct T {
+  ~T();
+};
+
+struct S thread_local s1;
+struct T thread_local t1;
+
 // Note that unlike normal C uninitialized global variables,
 // uninitialized TLS variables do NOT have COMMON linkage.
 
 // CHECK-GD: @z1 = global i32 0
-// CHECK-GD: @f.y = internal thread_local global i32 0
 // CHECK-GD: @z2 = global i32 0
 // CHECK-GD: @x = thread_local global i32 0
+// CHECK-GD: @_ZZ1fvE1y = internal thread_local global i32 0
 // CHECK-GD: @z = thread_local(initialexec) global i32 0
+// CHECK-GD: @s1 = thread_local global %struct.S zeroinitializer
+// CHECK-GD: @t1 = thread_local global %struct.T zeroinitializer
+// CHECK-GD: @__tls_guard = internal thread_local global i8 0
 
 // CHECK-LD: @z1 = global i32 0
-// CHECK-LD: @f.y = internal thread_local(localdynamic) global i32 0
 // CHECK-LD: @z2 = global i32 0
 // CHECK-LD: @x = thread_local(localdynamic) global i32 0
+// CHECK-LD: @_ZZ1fvE1y = internal thread_local(localdynamic) global i32 0
 // CHECK-LD: @z = thread_local(initialexec) global i32 0
+// CHECK-LD: @s1 = thread_local(localdynamic) global %struct.S zeroinitializer
+// CHECK-LD: @t1 = thread_local(localdynamic) global %struct.T zeroinitializer
+// CHECK-LD: @__tls_guard = internal thread_local(localdynamic) global i8 0
 
 // CHECK-IE: @z1 = global i32 0
-// CHECK-IE: @f.y = internal thread_local(initialexec) global i32 0
 // CHECK-IE: @z2 = global i32 0
 // CHECK-IE: @x = thread_local(initialexec) global i32 0
+// CHECK-IE: @_ZZ1fvE1y = internal thread_local(initialexec) global i32 0
 // CHECK-IE: @z = thread_local(initialexec) global i32 0
+// CHECK-IE: @s1 = thread_local(initialexec) global %struct.S zeroinitializer
+// CHECK-IE: @t1 = thread_local(initialexec) global %struct.T zeroinitializer
+// CHECK-IE: @__tls_guard = internal thread_local(initialexec) global i8 0
 
 // CHECK-LE: @z1 = global i32 0
-// CHECK-LE: @f.y = internal thread_local(localexec) global i32 0
 // CHECK-LE: @z2 = global i32 0
 // CHECK-LE: @x = thread_local(localexec) global i32 0
+// CHECK-LE: @_ZZ1fvE1y = internal thread_local(localexec) global i32 0
 // CHECK-LE: @z = thread_local(initialexec) global i32 0
+// CHECK-LE: @s1 = thread_local(localexec) global %struct.S zeroinitializer
+// CHECK-LE: @t1 = thread_local(localexec) global %struct.T zeroinitializer
+// CHECK-LE: @__tls_guard = internal thread_local(localexec) global i8 0