Initial SEH IR generation implementation

[rnk]

Summary

Most of the complexity here is emitting filter expressions as standalone
LLVM functions. The filter functions are used in catch clauses in place
of type_info globals, and LLVM does the rest of the lowering.

Major aspects that still need work:

• __try body outlining with noinline and optnone.
• Cleanup support for __finally and destructors. LLVM won't be able to lower the current IR to anything useful.
• Local variable access in filter expressions.
• Don't use TLS to figure out which filter function fired. This requires work in LLVM.

[ABataev]

Reid,

__try body outlining with noinline and optnone.

You can use CapturedStmt for body outlining. It works just like Lambdas, but all captured vars are passed by reference

[rnk]

In D5607#5, @ABataev wrote:

Reid,

__try body outlining with noinline and optnone.

You can use CapturedStmt for body outlining. It works just like Lambdas, but all captured vars are passed by reference

Thanks, I actually tried that in a subsequent patch layered on this one. It worked great, but Richard isn't particularly happy with the direction. He feels that CapturedStmt is an implementation artifact that shouldn't be in the AST because it doesn't reflect anything user-written. On the other hand, IMO the AST already has explicit nodes for implicit things that the user didn't write, such as MaterializeTemporaryExpr. I will probably try doing outlining during IRGen to compare the approaches.

[ABataev]

He feels that CapturedStmt is an implementation artifact that shouldn't be in the AST because it doesn't reflect anything user-written.

Actually, initially this class was designed to be a basic class for other statements, which may require function outlining. But Doug Gregor recommended to make it a separate statement.

[rnk]

Ping. I'm mostly interested in coming to consensus on direction.

[rnk]

Ping.

[rnk]

ping

[rjmccall]

I think the best solution here is definitely to let the backend handle outlining handler functions: to get reasonable code, the outlining has to be done by something that understands the layout of the outer function's frame, and only the backend can do that. No high-level intrinsic can capture the relationship between the functions in any sort of reliable way without completely disabling inlining of the outer function. And we really want to still be able to do data-flow optimizations in the outer function; we just have to ensure that any live-in values have been spilled to the stack at the call site.

Unfortunately, I think the same principle is going to end up applying to filter functions. If they don't use values from the enclosing frame, then fine, you can outline them in the frontend; but if they do there's just no way that a frontend solution is ever going to work. This might mean that we need to change the invoke IR pattern again so that the filter functions are just basic blocks within the function (which end in some special intrinsic/instruction that guides outlining).

I think you need to have that conversation on the LLVM side before you can make any useful progress on the Clang side. Please CC me on that conversation when you start it.

lib/CodeGen/CGException.cpp

104

Seems like a reasonable thing for the CXXABI to return.

438

Don't do this. Make a flag somehow.

[rnk]

In D5607#13, @rjmccall wrote:

I think the best solution here is definitely to let the backend handle outlining handler functions: to get reasonable code, the outlining has to be done by something that understands the layout of the outer function's frame, and only the backend can do that. No high-level intrinsic can capture the relationship between the functions in any sort of reliable way without completely disabling inlining of the outer function. And we really want to still be able to do data-flow optimizations in the outer function; we just have to ensure that any live-in values have been spilled to the stack at the call site.

I agree, this is the right representation for catches and cleanups for the reasons you mention. I also don't want to introduce a new EH representation that is only used by SEH.

Unfortunately, I think the same principle is going to end up applying to filter functions. If they don't use values from the enclosing frame, then fine, you can outline them in the frontend; but if they do there's just no way that a frontend solution is ever going to work. This might mean that we need to change the invoke IR pattern again so that the filter functions are just basic blocks within the function (which end in some special intrinsic/instruction that guides outlining).

OK. I was thinking that cleanups are common (C++ destructors) while __try and __except filter expressions are rare. Therefore we can do something slow and dumb in the frontend for them. But, if we have to implement late outlining for EH handlers, we might as well extend it to filter expressions. __try bodies will still be outlined for simplicity.

I think you need to have that conversation on the LLVM side before you can make any useful progress on the Clang side. Please CC me on that conversation when you start it.

Sure.

[rnk]

• Rewrite without frontend outlining

[rjmccall]

lib/CodeGen/CGException.cpp

1702

Can you at least document what the point of this is?

1772

Doesn't the finally handler need to be active within the catch handler? You probably need to push it first.

1821

Shouldn't you be branching on the filter result or something? It feels wrong for the control flow here to be implicit.

1830

Yeah, you're popping it after you pop the catch. That can't possibly work if you pushed the catch first.

This also suggests that you aren't testing that adequately. :)

lib/Sema/SemaChecking.cpp

467

This checking can run in template instantiations, but getCurScope() isn't meaningful there.

[rnk]

lib/CodeGen/CGException.cpp

1702

Sure, check out the comment.

1702

Sure, see the new comment.

1772

A __try can only have one handler, __except or __finally, and not both. I'll straighten out the control flow to clarify that.

1821

The idea in this patch is that we enter the landing pad, and we sort of non-determinstically already know what the selector is. The @llvm.eh.seh.filter call acts as a barrier against code motion of side effects. It's kind of like a suspend / resume point in a coroutine. What happens in practice is that the code between the typeid check and the @llvm.eh.seh.filter call gets outlined.

Now that I'm thinking about this harder, I'll admit this is kind of broken for nested cleanups, as the cleanups will run before the filter, which is not what actually happens in practice (filters run during phase 1). =/

What do you think about going back to outlining filters in the frontend? I've had many weeks of coffee and whiteboard discussions trying to figure out the least disruptive way to fit filter functions into LLVM's EH scheme, and we haven't come up with anything good. We can still do backend outlining for cleanups and C++ EH. LLVM's EH model just doesn't have a place for code that gets executed during phase 1 EH unwinding.

1830

As mentioned, this is impossible, but it's a reasonable comment. :)

lib/Sema/SemaChecking.cpp

467

Right. :( This is always a syntactic nesting property, though. Instantiation cannot change it. Should I just check it during the parse (if so, where?), or just suppress this Sema check during instantiation, since it will always give the same result?

[rnk]

• straighten control flow

[rnk]

• Go back to outlined filter expressions with no captures

[rjmccall]

lib/CodeGen/CGException.cpp

1772

Okay, makes sense.

1821

I think outlining filters is an acceptable implementation model given the oddity of the control flow. Can we get away with just disallowing references to local variables in them for now, or is that an important use case?

lib/Sema/SemaChecking.cpp

467

In theory, you can construct builtin calls in template instantiations with e.g. non-type template arguments. It's probably safe to just not diagnose that well as long as you don't crash during IRGen if you miss a case.

You don't really have a choice about where you detect this; there *are* often separate Sema entrypoints for parsing vs. instantiation, but we don't want to burden every function call parse with an extra check just for this. I think Sema tracks that's currently instantiating a function body, but I don't remember where.

[rnk]

lib/CodeGen/CGException.cpp

1821

We can definitely get away with rejecting local variable capture for now. I can't find any instances of it in the code I have available.

However, eventually we will need to support it, at which point I would want to use the @llvm.frameallocate @llvm.framerecover intrinsics. Those inhibit inlining, which I think is OK for this particular corner case. We already support similarly heinous extensions like labels as values that inhibit inlining. :)

lib/Sema/SemaChecking.cpp

467

OK. We'll produce undef for these during codegen if they sneak in some how.

[rnk]

I think we're in reasonable agreement about the direction here, so I'm going to go ahead and commit this to unblock Andy who is working on the EH preparation changes.

With this commit, basic __try { } __except() { } will start to compile and behave correctly.

[rnk]

• Final version of the patch

include/clang/AST/Mangle.h

public:
 
   virtual void mangleStaticGuardVariable(const VarDecl *D, raw_ostream &) = 0;
 
   virtual void mangleDynamicInitializer(const VarDecl *D, raw_ostream &) = 0;
 
   virtual void mangleDynamicAtExitDestructor(const VarDecl *D,
                                              raw_ostream &) = 0;
 
+  virtual void mangleSEHFilterExpression(const NamedDecl *EnclosingDecl,
+                                         raw_ostream &Out) = 0;
+
   /// Generates a unique string for an externally visible type for use with TBAA
   /// or type uniquing.
   /// TODO: Extend this to internal types by generating names that are unique
   /// across translation units so it can be used with LTO.
   virtual void mangleTypeName(QualType T, raw_ostream &) = 0;
 
   /// @}
 };

include/clang/Basic/Builtins.def

 BUILTIN(__sync_fetch_and_umax, "UiUiD*Ui", "n")
 
 // Random libc builtins.
 BUILTIN(__builtin_abort, "v", "Fnr")
 BUILTIN(__builtin_index, "c*cC*i", "Fn")
 BUILTIN(__builtin_rindex, "c*cC*i", "Fn")
 
 // Microsoft builtins.  These are only active with -fms-extensions.
-LANGBUILTIN(_alloca,      "v*z", "n", ALL_MS_LANGUAGES)
-LANGBUILTIN(__assume,     "vb",  "n", ALL_MS_LANGUAGES)
-LANGBUILTIN(__noop,       "i.",  "n", ALL_MS_LANGUAGES)
-LANGBUILTIN(__debugbreak, "v",   "n", ALL_MS_LANGUAGES)
-LANGBUILTIN(__va_start,   "vc**.", "nt", ALL_MS_LANGUAGES)
+LANGBUILTIN(_alloca,          "v*z", "n", ALL_MS_LANGUAGES)
+LANGBUILTIN(__assume,         "vb",  "n", ALL_MS_LANGUAGES)
+LANGBUILTIN(_exception_code,  "ULi", "n", ALL_MS_LANGUAGES)
+LANGBUILTIN(__exception_code, "ULi", "n", ALL_MS_LANGUAGES)
+LANGBUILTIN(_exception_info,  "v*",  "n", ALL_MS_LANGUAGES)
+LANGBUILTIN(__exception_info, "v*",  "n", ALL_MS_LANGUAGES)
+LANGBUILTIN(__noop,           "i.",  "n", ALL_MS_LANGUAGES)
+LANGBUILTIN(__debugbreak,     "v",   "n", ALL_MS_LANGUAGES)
+LANGBUILTIN(__va_start,       "vc**.", "nt", ALL_MS_LANGUAGES)
 LANGBUILTIN(_InterlockedCompareExchange, "LiLiD*LiLi", "n", ALL_MS_LANGUAGES)
 LANGBUILTIN(_InterlockedCompareExchangePointer, "v*v*D*v*v*", "n", ALL_MS_LANGUAGES)
 LANGBUILTIN(_InterlockedIncrement, "LiLiD*", "n", ALL_MS_LANGUAGES)
 LANGBUILTIN(_InterlockedDecrement, "LiLiD*", "n", ALL_MS_LANGUAGES)
 LANGBUILTIN(_InterlockedExchangeAdd, "LiLiD*Li", "n", ALL_MS_LANGUAGES)
 LANGBUILTIN(_InterlockedExchangePointer, "v*v*D*v*", "n", ALL_MS_LANGUAGES)
 LANGBUILTIN(_InterlockedExchange, "LiLiD*Li", "n", ALL_MS_LANGUAGES)
 LANGBUILTIN(__readfsdword, "ULiULi", "n", ALL_MS_LANGUAGES)

include/clang/Basic/DiagnosticCommonKinds.td

 def err_integer_literal_too_large : Error<
   "integer literal is too large to be represented in any %select{signed |}0"
   "integer type">;
 def ext_integer_literal_too_large_for_signed : ExtWarn<
   "integer literal is too large to be represented in a signed integer type, "
   "interpreting as unsigned">,
   InGroup<DiagGroup<"implicitly-unsigned-literal">>;
 
+// SEH
+def err_seh_expected_handler : Error<
+  "expected '__except' or '__finally' block">;
+def err_seh___except_block : Error<
+  "%0 only allowed in __except block or filter expression">;
+def err_seh___except_filter : Error<
+  "%0 only allowed in __except filter expression">;
+def err_seh___finally_block : Error<
+  "%0 only allowed in __finally block">;
+
 // Sema && AST
 def note_invalid_subexpr_in_const_expr : Note<
   "subexpression not valid in a constant expression">;
 
 // Targets
 
 def err_target_unknown_triple : Error<
   "unknown target triple '%0', please use -triple or -arch">;

include/clang/Basic/DiagnosticParseKinds.td

 // OpenCL EXTENSION pragma (OpenCL 1.1 [9.1])
 def warn_pragma_expected_colon : Warning<
   "missing ':' after %0 - ignoring">, InGroup<IgnoredPragmas>;
 def warn_pragma_expected_enable_disable : Warning<
   "expected 'enable' or 'disable' - ignoring">, InGroup<IgnoredPragmas>;
 def warn_pragma_unknown_extension : Warning<
   "unknown OpenCL extension %0 - ignoring">, InGroup<IgnoredPragmas>;
 
-def err_seh_expected_handler : Error<
-  "expected '__except' or '__finally' block">;
-
-def err_seh___except_block : Error<
-  "%0 only allowed in __except block">;
-
-def err_seh___except_filter : Error<
-  "%0 only allowed in __except filter expression">;
-
-def err_seh___finally_block : Error<
-  "%0 only allowed in __finally block">;
-
 // OpenMP support.
 def warn_pragma_omp_ignored : Warning<
   "unexpected '#pragma omp ...' in program">, InGroup<SourceUsesOpenMP>, DefaultIgnore;
 def warn_omp_extra_tokens_at_eol : Warning<
   "extra tokens at the end of '#pragma omp %0' are ignored">,
   InGroup<ExtraTokens>;
 def err_omp_unknown_directive : Error<
   "expected an OpenMP directive">;

include/clang/Sema/Scope.h

enum ScopeFlags {
     /// \brief This is the scope of some OpenMP simd directive.
     /// For example, it is used for 'omp simd', 'omp for simd'.
     /// This flag is propagated to children scopes.
     OpenMPSimdDirectiveScope = 0x20000,
 
     /// This scope corresponds to an enum.
     EnumScope = 0x40000,
 
-    /// This scope corresponds to a SEH try.
+    /// This scope corresponds to an SEH try.
     SEHTryScope = 0x80000,
+
+    /// This scope corresponds to an SEH except.
+    SEHExceptScope = 0x100000,
+
+    /// We are currently in the filter expression of an SEH except block.
+    SEHFilterScope = 0x200000,
   };
 private:
   /// The parent scope for this scope.  This is null for the translation-unit
   /// scope.
   Scope *AnyParent;
 
   /// Flags - This contains a set of ScopeFlags, which indicates how the scope
   /// interrelates with other control flow statements.
public:
   }
 
   /// \brief Determine whether this scope is a C++ 'try' block.
   bool isTryScope() const { return getFlags() & Scope::TryScope; }
 
   /// \brief Determine whether this scope is a SEH '__try' block.
   bool isSEHTryScope() const { return getFlags() & Scope::SEHTryScope; }
 
+  /// \brief Determine whether this scope is a SEH '__except' block.
+  bool isSEHExceptScope() const { return getFlags() & Scope::SEHExceptScope; }
+
   /// containedInPrototypeScope - Return true if this or a parent scope
   /// is a FunctionPrototypeScope.
   bool containedInPrototypeScope() const;
 
   void PushUsingDirective(UsingDirectiveDecl *UDir) {
     UsingDirectives.push_back(UDir);
   }
 

lib/AST/ItaniumMangle.cpp

void mangleCXXDtor(const CXXDestructorDecl *D, CXXDtorType Type,
                      raw_ostream &) override;
 
   void mangleCXXCtorComdat(const CXXConstructorDecl *D, raw_ostream &) override;
   void mangleCXXDtorComdat(const CXXDestructorDecl *D, raw_ostream &) override;
   void mangleStaticGuardVariable(const VarDecl *D, raw_ostream &) override;
   void mangleDynamicInitializer(const VarDecl *D, raw_ostream &Out) override;
   void mangleDynamicAtExitDestructor(const VarDecl *D,
                                      raw_ostream &Out) override;
+  void mangleSEHFilterExpression(const NamedDecl *EnclosingDecl,
+                                 raw_ostream &Out) override;
   void mangleItaniumThreadLocalInit(const VarDecl *D, raw_ostream &) override;
   void mangleItaniumThreadLocalWrapper(const VarDecl *D,
                                        raw_ostream &) override;
 
   void mangleStringLiteral(const StringLiteral *, raw_ostream &) override;
 
   bool getNextDiscriminator(const NamedDecl *ND, unsigned &disc) {
     // Lambda closure types are already numbered.
void ItaniumMangleContextImpl::mangleDynamicAtExitDestructor(const VarDecl *D,
   CXXNameMangler Mangler(*this, Out);
   Mangler.getStream() << "__dtor_";
   if (shouldMangleDeclName(D))
     Mangler.mangle(D);
   else
     Mangler.getStream() << D->getName();
 }
 
+void ItaniumMangleContextImpl::mangleSEHFilterExpression(
+    const NamedDecl *EnclosingDecl, raw_ostream &Out) {
+  CXXNameMangler Mangler(*this, Out);
+  Mangler.getStream() << "__filt_";
+  if (shouldMangleDeclName(EnclosingDecl))
+    Mangler.mangle(EnclosingDecl);
+  else
+    Mangler.getStream() << EnclosingDecl->getName();
+}
+
 void ItaniumMangleContextImpl::mangleItaniumThreadLocalInit(const VarDecl *D,
                                                             raw_ostream &Out) {
   //  <special-name> ::= TH <object name>
   CXXNameMangler Mangler(*this, Out);
   Mangler.getStream() << "_ZTH";
   Mangler.mangleName(D);
 }
 

lib/AST/MicrosoftMangle.cpp

 
 /// MicrosoftMangleContextImpl - Overrides the default MangleContext for the
 /// Microsoft Visual C++ ABI.
 class MicrosoftMangleContextImpl : public MicrosoftMangleContext {
   typedef std::pair<const DeclContext *, IdentifierInfo *> DiscriminatorKeyTy;
   llvm::DenseMap<DiscriminatorKeyTy, unsigned> Discriminator;
   llvm::DenseMap<const NamedDecl *, unsigned> Uniquifier;
   llvm::DenseMap<const CXXRecordDecl *, unsigned> LambdaIds;
+  llvm::DenseMap<const NamedDecl *, unsigned> SEHFilterIds;
 
 public:
   MicrosoftMangleContextImpl(ASTContext &Context, DiagnosticsEngine &Diags)
       : MicrosoftMangleContext(Context, Diags) {}
   bool shouldMangleCXXName(const NamedDecl *D) override;
   bool shouldMangleStringLiteral(const StringLiteral *SL) override;
   void mangleCXXName(const NamedDecl *D, raw_ostream &Out) override;
   void mangleVirtualMemPtrThunk(const CXXMethodDecl *MD,
public:
   void mangleCXXDtor(const CXXDestructorDecl *D, CXXDtorType Type,
                      raw_ostream &) override;
   void mangleReferenceTemporary(const VarDecl *, unsigned ManglingNumber,
                                 raw_ostream &) override;
   void mangleStaticGuardVariable(const VarDecl *D, raw_ostream &Out) override;
   void mangleDynamicInitializer(const VarDecl *D, raw_ostream &Out) override;
   void mangleDynamicAtExitDestructor(const VarDecl *D,
                                      raw_ostream &Out) override;
+  void mangleSEHFilterExpression(const NamedDecl *EnclosingDecl,
+                                 raw_ostream &Out) override;
   void mangleStringLiteral(const StringLiteral *SL, raw_ostream &Out) override;
   bool getNextDiscriminator(const NamedDecl *ND, unsigned &disc) {
     // Lambda closure types are already numbered.
     if (isLambda(ND))
       return false;
 
     const DeclContext *DC = getEffectiveDeclContext(ND);
     if (!DC->isFunctionOrMethod())
void MicrosoftMangleContextImpl::mangleCXXRTTICompleteObjectLocator(
   Mangler.getStream() << "\01??_R4";
   Mangler.mangleName(Derived);
   Mangler.getStream() << "6B"; // '6' for vftable, 'B' for const.
   for (const CXXRecordDecl *RD : BasePath)
     Mangler.mangleName(RD);
   Mangler.getStream() << '@';
 }
 
+void MicrosoftMangleContextImpl::mangleSEHFilterExpression(
+    const NamedDecl *EnclosingDecl, raw_ostream &Out) {
+  MicrosoftCXXNameMangler Mangler(*this, Out);
+  // The function body is in the same comdat as the function with the handler,
+  // so the numbering here doesn't have to be the same across TUs.
+  //
+  // <mangled-name> ::= ?filt$ <filter-number> @0
+  Mangler.getStream() << "\01?filt$" << SEHFilterIds[EnclosingDecl]++ << "@0@";
+  Mangler.mangleName(EnclosingDecl);
+}
+
 void MicrosoftMangleContextImpl::mangleTypeName(QualType T, raw_ostream &Out) {
   // This is just a made up unique string for the purposes of tbaa.  undname
   // does *not* know how to demangle it.
   MicrosoftCXXNameMangler Mangler(*this, Out);
   Mangler.getStream() << '?';
   Mangler.mangleType(T, SourceRange());
 }
 

lib/CodeGen/CGBuiltin.cpp

   case Builtin::BI__readfsdword: {
     Value *IntToPtr =
       Builder.CreateIntToPtr(EmitScalarExpr(E->getArg(0)),
                              llvm::PointerType::get(CGM.Int32Ty, 257));
     LoadInst *Load =
         Builder.CreateAlignedLoad(IntToPtr, /*Align=*/4, /*isVolatile=*/true);
     return RValue::get(Load);
   }
+
+  case Builtin::BI__exception_code:
+  case Builtin::BI_exception_code:
+    return RValue::get(EmitSEHExceptionCode());
+  case Builtin::BI__exception_info:
+  case Builtin::BI_exception_info:
+    return RValue::get(EmitSEHExceptionInfo());
   }
 
   // If this is an alias for a lib function (e.g. __builtin_sin), emit
   // the call using the normal call path, but using the unmangled
   // version of the function name.
   if (getContext().BuiltinInfo.isLibFunction(BuiltinID))
     return emitLibraryCall(*this, FD, E,
                            CGM.getBuiltinLibFunction(FD, BuiltinID));

lib/CodeGen/CGException.cpp

 //
 //===----------------------------------------------------------------------===//
 
 #include "CodeGenFunction.h"
 #include "CGCXXABI.h"
 #include "CGCleanup.h"
 #include "CGObjCRuntime.h"
 #include "TargetInfo.h"
+#include "clang/AST/Mangle.h"
 #include "clang/AST/StmtCXX.h"
 #include "clang/AST/StmtObjC.h"
 #include "llvm/IR/CallSite.h"
 #include "llvm/IR/Intrinsics.h"
 
 using namespace clang;
 using namespace CodeGen;
 
static llvm::Constant *getTerminateFn(CodeGenModule &CGM) {
   // void __terminate();
 
   llvm::FunctionType *FTy =
     llvm::FunctionType::get(CGM.VoidTy, /*IsVarArgs=*/false);
 
   StringRef name;
 
   // In C++, use std::terminate().
-  if (CGM.getLangOpts().CPlusPlus)
-    name = "_ZSt9terminatev"; // FIXME: mangling!
-  else if (CGM.getLangOpts().ObjC1 &&
+  if (CGM.getLangOpts().CPlusPlus &&
+      CGM.getTarget().getCXXABI().isItaniumFamily()) {
+    name = "_ZSt9terminatev";
+  } else if (CGM.getLangOpts().ObjC1 &&
            CGM.getLangOpts().ObjCRuntime.hasTerminate())
     name = "objc_terminate";
   else
     name = "abort";
   return CGM.CreateRuntimeFunction(FTy, name);
 }
 
 static llvm::Constant *getCatchallRethrowFn(CodeGenModule &CGM,
struct EHPersonality {
     static const EHPersonality GNU_C_SEH;
     static const EHPersonality GNU_ObjC;
     static const EHPersonality GNUstep_ObjC;
     static const EHPersonality GNU_ObjCXX;
     static const EHPersonality NeXT_ObjC;
     static const EHPersonality GNU_CPlusPlus;
     static const EHPersonality GNU_CPlusPlus_SJLJ;
     static const EHPersonality GNU_CPlusPlus_SEH;
+    static const EHPersonality MSVC_except_handler;
+    static const EHPersonality MSVC_C_specific_handler;
   };
 }
 
 const EHPersonality EHPersonality::GNU_C = { "__gcc_personality_v0", nullptr };
 const EHPersonality
 EHPersonality::GNU_C_SJLJ = { "__gcc_personality_sj0", nullptr };
 const EHPersonality
 EHPersonality::GNU_C_SEH = { "__gcc_personality_seh0", nullptr };
 const EHPersonality
 EHPersonality::NeXT_ObjC = { "__objc_personality_v0", nullptr };
 const EHPersonality
 EHPersonality::GNU_CPlusPlus = { "__gxx_personality_v0", nullptr };
 const EHPersonality
 EHPersonality::GNU_CPlusPlus_SJLJ = { "__gxx_personality_sj0", nullptr };
 const EHPersonality
 EHPersonality::GNU_CPlusPlus_SEH = { "__gxx_personality_seh0", nullptr };
 const EHPersonality
 EHPersonality::GNU_ObjC = {"__gnu_objc_personality_v0", "objc_exception_throw"};
 const EHPersonality
 EHPersonality::GNU_ObjCXX = { "__gnustep_objcxx_personality_v0", nullptr };
 const EHPersonality
 EHPersonality::GNUstep_ObjC = { "__gnustep_objc_personality_v0", nullptr };
+const EHPersonality
+EHPersonality::MSVC_except_handler = { "_except_handler4", nullptr };
+const EHPersonality
+EHPersonality::MSVC_C_specific_handler = { "__C_specific_handler", nullptr };
 
 /// On Win64, use libgcc's SEH personality function. We fall back to dwarf on
 /// other platforms, unless the user asked for SjLj exceptions.
 static bool useLibGCCSEHPersonality(const llvm::Triple &T) {
   return T.isOSWindows() && T.getArch() == llvm::Triple::x86_64;
 }
 
 static const EHPersonality &getCPersonality(const llvm::Triple &T,
static const EHPersonality &getObjCXXPersonality(const llvm::Triple &T,
   case ObjCRuntime::ObjFW: // XXX: this will change soon
     return EHPersonality::GNU_ObjC;
   case ObjCRuntime::GNUstep:
     return EHPersonality::GNU_ObjCXX;
   }
   llvm_unreachable("bad runtime kind");
 }
 
+static const EHPersonality &getCPersonalityMSVC(const llvm::Triple &T,
+                                                const LangOptions &L) {
+  if (L.SjLjExceptions)
+    return EHPersonality::GNU_C_SJLJ;
+
+  if (T.getArch() == llvm::Triple::x86)
+    return EHPersonality::MSVC_except_handler;
+  return EHPersonality::MSVC_C_specific_handler;
+}
+
+static const EHPersonality &getCXXPersonalityMSVC(const llvm::Triple &T,
+                                                  const LangOptions &L) {
+  if (L.SjLjExceptions)
+    return EHPersonality::GNU_CPlusPlus_SJLJ;
+  // FIXME: Implement C++ exceptions.
+  return getCPersonalityMSVC(T, L);
+}
+
 const EHPersonality &EHPersonality::get(CodeGenModule &CGM) {
   const llvm::Triple &T = CGM.getTarget().getTriple();
   const LangOptions &L = CGM.getLangOpts();
+  // Try to pick a personality function that is compatible with MSVC if we're
+  // not compiling Obj-C. Obj-C users better have an Obj-C runtime that supports
+  // the GCC-style personality function.
+  if (T.isWindowsMSVCEnvironment() && !L.ObjC1) {
+    if (L.CPlusPlus)
+      return getCXXPersonalityMSVC(T, L);
+    else
+      return getCPersonalityMSVC(T, L);
+  }
+
   if (L.CPlusPlus && L.ObjC1)
     return getObjCXXPersonality(T, L);
   else if (L.CPlusPlus)
     return getCXXPersonality(T, L);
   else if (L.ObjC1)
     return getObjCPersonality(T, L);
   else
     return getCPersonality(T, L);
static void EmitAnyExprToExn(CodeGenFunction &CGF, const Expr *e,
   // differently if it can't be elided.
   CGF.EmitAnyExprToMem(e, typedAddr, e->getType().getQualifiers(), 
                        /*IsInit*/ true);
 
   // Deactivate the cleanup block.
   CGF.DeactivateCleanupBlock(cleanup, cast<llvm::Instruction>(typedAddr));
 }
 
+/// Returns true if this landing pad is using filter functions to select catches
+/// instead of RTTI.
+static bool EHUsesFilterFunctions(CodeGenModule &CGM) {
+  const EHPersonality &P = EHPersonality::get(CGM);
+  return StringRef("__C_specific_handler") == P.PersonalityFn;
+}
+
 llvm::Value *CodeGenFunction::getExceptionSlot() {
   if (!ExceptionSlot)
     ExceptionSlot = CreateTempAlloca(Int8PtrTy, "exn.slot");
   return ExceptionSlot;
 }
 
 llvm::Value *CodeGenFunction::getEHSelectorSlot() {
   if (!EHSelectorSlot)
llvm::BasicBlock *CodeGenFunction::getEHResumeBlock(bool isCleanup) {
   LPadVal = Builder.CreateInsertValue(LPadVal, Sel, 1, "lpad.val");
 
   Builder.CreateResume(LPadVal);
   Builder.restoreIP(SavedIP);
   return EHResumeBlock;
 }
 
 void CodeGenFunction::EmitSEHTryStmt(const SEHTryStmt &S) {
-  CGM.ErrorUnsupported(&S, "SEH __try");
+  EnterSEHTryStmt(S);
+  EmitStmt(S.getTryBlock());
+  ExitSEHTryStmt(S);
+}
+
+namespace {
+struct PerformSEHFinally : EHScopeStack::Cleanup  {
+  Stmt *Block;
+  PerformSEHFinally(Stmt *Block) : Block(Block) {}
+  void Emit(CodeGenFunction &CGF, Flags F) override { CGF.EmitStmt(Block); }
+};
+}
+
+/// Create a stub filter function that will ultimately hold the code of the
+/// filter expression. The EH preparation passes in LLVM will outline the code
+/// from the main function body into this stub.
+llvm::Constant *
+CodeGenFunction::GenerateSEHFilterFunction() {
+  // Get the mangled function name.
+  SmallString<128> Name;
+  {
+    llvm::raw_svector_ostream OS(Name);
+    const NamedDecl *Parent = dyn_cast_or_null<NamedDecl>(CurCodeDecl);
+    assert(Parent && "FIXME: handle unnamed decls (lambdas, blocks) with SEH");
+    CGM.getCXXABI().getMangleContext().mangleSEHFilterExpression(Parent, OS);
+  }
+
+  // Arrange a function with the declaration:
+  // int filt(EXCEPTION_POINTERS *exception_pointers, void *frame_pointer)
+  QualType ArgTys[2] = { getContext().VoidPtrTy, getContext().VoidPtrTy };
+  FunctionProtoType::ExtProtoInfo EPI;
+  QualType FTP =
+      getContext().getFunctionType(getContext().IntTy, ArgTys, EPI);
+  CanQual<FunctionProtoType> CanFTP =
+      FTP->getCanonicalTypeUnqualified().getAs<FunctionProtoType>();
+  const CGFunctionInfo &FnInfo = CGM.getTypes().arrangeFreeFunctionType(CanFTP);
+  llvm::FunctionType *FnTy = CGM.getTypes().GetFunctionType(FnInfo);
+  llvm::Function *Fn = llvm::Function::Create(FnTy, CurFn->getLinkage(),
+                                              Name.str(), &CGM.getModule());
+
+  // The filter is either in the same comdat as the function, or it's internal.
+  if (llvm::Comdat *C = CurFn->getComdat()) {
+    Fn->setComdat(C);
+  } else if (CurFn->hasWeakLinkage() || CurFn->hasLinkOnceLinkage()) {
+    llvm::Comdat *C = CGM.getModule().getOrInsertComdat(CurFn->getName());
+    CurFn->setComdat(C);
+    Fn->setComdat(C);
+  } else {
+    Fn->setLinkage(llvm::GlobalValue::InternalLinkage);
+  }
+
+  // Insert an unreachable entry block.
+  llvm::BasicBlock *Entry = llvm::BasicBlock::Create(getLLVMContext(), "", Fn);
+  CGBuilderTy(Entry).CreateUnreachable();
+
+  return llvm::ConstantExpr::getBitCast(Fn, Int8PtrTy);
+}
+
+llvm::Value *CodeGenFunction::EmitSEHExceptionInfo() {
+  return getExceptionFromSlot();
+}
+
+llvm::Value *CodeGenFunction::EmitSEHExceptionCode() {
+  // FIXME: Implement outside of filter expressions.
+  //
+  // struct EXCEPTION_POINTERS {
+  //   EXCEPTION_RECORD *ExceptionRecord;
+  //   CONTEXT *ContextRecord;
+  // };
+  // ((EXCEPTION_POINTERS*)ehptr)->ExceptionRecord->ExceptionCode;
+  llvm::Type *RecordTy = CGM.Int32Ty->getPointerTo();
+  llvm::Type *PtrsTy = llvm::StructType::get(RecordTy, CGM.VoidPtrTy, nullptr);
+  llvm::Value *EHPtr =
+      Builder.CreateBitCast(getExceptionFromSlot(), PtrsTy->getPointerTo());
+  llvm::Value *Rec = Builder.CreateStructGEP(EHPtr, 0);
+  Rec = Builder.CreateLoad(Rec);
+  return Builder.CreateLoad(Rec);
+}
+
+void CodeGenFunction::EnterSEHTryStmt(const SEHTryStmt &S) {
+  if (SEHFinallyStmt *Finally = S.getFinallyHandler()) {
+    // Push a cleanup for __finally blocks.
+    EHStack.pushCleanup<PerformSEHFinally>(NormalAndEHCleanup,
+                                           Finally->getBlock());
+    return;
+  }
+
+  // Otherwise, we must have an __except block.
+  assert(S.getExceptHandler());
+  EHCatchScope *CatchScope = EHStack.pushCatch(1);
+  llvm::Constant *FilterFuncStub = GenerateSEHFilterFunction();
+  llvm::BasicBlock *ExceptBB = createBasicBlock("__except");
+  CatchScope->setHandler(0, FilterFuncStub, ExceptBB);
+}
+
+void CodeGenFunction::ExitSEHTryStmt(const SEHTryStmt &S) {
+  // Just pop the cleanup if it's a __finally block.
+  if (SEHFinallyStmt *Finally = S.getFinallyHandler()) {
+    PopCleanupBlock();
+    return;
+  }
+
+  // Otherwise, we must have an __except block.
+  SEHExceptStmt *Except = S.getExceptHandler();
+  assert(Except && "__try must have __finally xor __except");
+  EHCatchScope &CatchScope = cast<EHCatchScope>(*EHStack.begin());
+
+  // Don't emit the __except block if the __try block lacked invokes.
+  // TODO: Model unwind edges from instructions, either with iload / istore or
+  // a try body function.
+  if (!CatchScope.hasEHBranches()) {
+    CatchScope.clearHandlerBlocks();
+    EHStack.popCatch();
+    return;
+  }
+
+  // The fall-through block.
+  llvm::BasicBlock *ContBB = createBasicBlock("__try.cont");
+
+  // We just emitted the body of the __try; jump to the continue block.
+  if (HaveInsertPoint())
+    Builder.CreateBr(ContBB);
+
+  // Check if our filter function returned true.
+  emitCatchDispatchBlock(*this, CatchScope);
+
+  // Grab the block before we pop the handler.
+  llvm::BasicBlock *ExceptBB = CatchScope.getHandler(0).Block;
+  EHStack.popCatch();
+
+  EmitBlockAfterUses(ExceptBB);
+
+  // Emit the filter expression, and pass the result to @llvm.eh.seh.filter().
+  llvm::Value *R = EmitScalarExpr(Except->getFilterExpr());
+  assert(R->getType() == Int32Ty && "filter expr isn't i32");
+  // FIXME: Use getIntrinsic when this is a real intrinsic.
+  llvm::Function *FilterIntrin = cast_or_null<llvm::Function>(
+      CGM.getModule().getNamedValue("llvm.eh.seh.filter"));
+  if (!FilterIntrin) {
+    llvm::FunctionType *FnTy =
+        llvm::FunctionType::get(VoidTy, Int32Ty, /*isVarArg=*/false);
+    FilterIntrin =
+        llvm::Function::Create(FnTy, llvm::GlobalValue::ExternalLinkage,
+                               "llvm.eh.seh.filter", &CGM.getModule());
+  }
+  Builder.CreateCall(FilterIntrin, R);
+
+  // Emit the __except body.
+  EmitStmt(Except->getBlock());
+
+  Builder.CreateBr(ContBB);
+
+  EmitBlock(ContBB);
 }
 
 void CodeGenFunction::EmitSEHLeaveStmt(const SEHLeaveStmt &S) {
   CGM.ErrorUnsupported(&S, "SEH __leave");
 }

lib/CodeGen/CodeGenFunction.h

#endif
   void EmitObjCAutoreleasePoolStmt(const ObjCAutoreleasePoolStmt &S);
 
   void EnterCXXTryStmt(const CXXTryStmt &S, bool IsFnTryBlock = false);
   void ExitCXXTryStmt(const CXXTryStmt &S, bool IsFnTryBlock = false);
 
   void EmitCXXTryStmt(const CXXTryStmt &S);
   void EmitSEHTryStmt(const SEHTryStmt &S);
   void EmitSEHLeaveStmt(const SEHLeaveStmt &S);
+  void EnterSEHTryStmt(const SEHTryStmt &S);
+  void ExitSEHTryStmt(const SEHTryStmt &S);
+
+  llvm::Constant *GenerateSEHFilterFunction();
+
+  llvm::Value *EmitSEHExceptionCode();
+  llvm::Value *EmitSEHExceptionInfo();
+
   void EmitCXXForRangeStmt(const CXXForRangeStmt &S,
                            ArrayRef<const Attr *> Attrs = None);
 
   LValue InitCapturedStruct(const CapturedStmt &S);
   llvm::Function *EmitCapturedStmt(const CapturedStmt &S, CapturedRegionKind K);
   void GenerateCapturedStmtFunctionProlog(const CapturedStmt &S);
   llvm::Function *GenerateCapturedStmtFunctionEpilog(const CapturedStmt &S);
   llvm::Function *GenerateCapturedStmtFunction(const CapturedStmt &S);

lib/Parse/ParseStmt.cpp

 StmtResult Parser::ParseSEHExceptBlock(SourceLocation ExceptLoc) {
   PoisonIdentifierRAIIObject raii(Ident__exception_code, false),
     raii2(Ident___exception_code, false),
     raii3(Ident_GetExceptionCode, false);
 
   if (ExpectAndConsume(tok::l_paren))
     return StmtError();
 
-  ParseScope ExpectScope(this, Scope::DeclScope | Scope::ControlScope);
+  ParseScope ExpectScope(this, Scope::DeclScope | Scope::ControlScope |
+                                   Scope::SEHExceptScope);
 
   if (getLangOpts().Borland) {
     Ident__exception_info->setIsPoisoned(false);
     Ident___exception_info->setIsPoisoned(false);
     Ident_GetExceptionInfo->setIsPoisoned(false);
   }
-  ExprResult FilterExpr(ParseExpression());
+
+  ExprResult FilterExpr;
+  {
+    ParseScopeFlags FilterScope(this, getCurScope()->getFlags() |
+                                          Scope::SEHFilterScope);
+    FilterExpr = ParseExpression();
+  }
 
   if (getLangOpts().Borland) {
     Ident__exception_info->setIsPoisoned(true);
     Ident___exception_info->setIsPoisoned(true);
     Ident_GetExceptionInfo->setIsPoisoned(true);
   }
 
   if(FilterExpr.isInvalid())

lib/Sema/JumpDiagnostics.cpp

for (unsigned I = 0, E = TS->getNumHandlers(); I != E; ++I) {
                                  diag::note_exits_cxx_catch,
                                  CS->getSourceRange().getBegin()));
       BuildScopeInformation(CS->getHandlerBlock(), 
                             (newParentScope = Scopes.size()-1));
     }
     return;
   }
 
+  case Stmt::SEHTryStmtClass:
+    // FIXME: Implement jump diagnostics for bad SEH jumps.
+    break;
+
   default:
     break;
   }
 
   for (Stmt::child_range CI = S->children(); CI; ++CI) {
     if (SkipFirstSubStmt) {
       SkipFirstSubStmt = false;
       continue;

lib/Sema/Scope.cpp

if (Flags & FnScope) {
       OS << "TryScope";
       Flags &= ~TryScope;
     } else if (Flags & FnTryCatchScope) {
       OS << "FnTryCatchScope";
       Flags &= ~FnTryCatchScope;
     } else if (Flags & SEHTryScope) {
       OS << "SEHTryScope";
       Flags &= ~SEHTryScope;
+    } else if (Flags & SEHExceptScope) {
+      OS << "SEHExceptScope";
+      Flags &= ~SEHExceptScope;
     } else if (Flags & OpenMPDirectiveScope) {
       OS << "OpenMPDirectiveScope";
       Flags &= ~OpenMPDirectiveScope;
     } else if (Flags & OpenMPLoopDirectiveScope) {
       OS << "OpenMPLoopDirectiveScope";
       Flags &= ~OpenMPLoopDirectiveScope;
     } else if (Flags & OpenMPSimdDirectiveScope) {
       OS << "OpenMPSimdDirectiveScope";

lib/Sema/SemaChecking.cpp

#include "clang/Basic/Builtins.def"
   case Builtin::BI__builtin___vsnprintf_chk:
     SemaBuiltinMemChkCall(*this, FDecl, TheCall, 1, 3);
     break;
 
   case Builtin::BI__builtin_call_with_static_chain:
     if (SemaBuiltinCallWithStaticChain(*this, TheCall))
       return ExprError();
     break;
+
+  case Builtin::BI__exception_code:
+  case Builtin::BI_exception_code: {
+    Scope *S = getCurScope();
+    while (S && !S->isSEHExceptScope())
+      S = S->getParent();
+    if (!S || !S->isSEHExceptScope()) {
+      auto *DRE = cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts());
+      Diag(TheCall->getExprLoc(), diag::err_seh___except_block)
+          << DRE->getDecl()->getIdentifier();
+      return ExprError();
+    }
+    break;
+  }
+
+  case Builtin::BI__exception_info:
+  case Builtin::BI_exception_info: {
+    // FIXME: Diagnose if this is used in the __except block.
+    Scope *S = getCurScope();
+    while (S && !S->isSEHExceptScope())
+      S = S->getParent();
+    if (!S || !(S->getFlags() & Scope::SEHFilterScope)) {
+      auto *DRE = cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts());
+      Diag(TheCall->getExprLoc(), diag::err_seh___except_filter)
+          << DRE->getDecl()->getIdentifier();
+      return ExprError();
+    }
+    break;
+  }
+
   }
 
   // Since the target specific builtins for each arch overlap, only check those
   // of the arch we are compiling for.
   if (BuiltinID >= Builtin::FirstTSBuiltin) {
     switch (Context.getTargetInfo().getTriple().getArch()) {
       case llvm::Triple::arm:
       case llvm::Triple::armeb:

test/CodeGen/exceptions-seh-leave.c

__try {
     __leave;
   } __except (1) {
     return 0;
   }
   *res = myres;
   return 1;
 }
 // CHECK-NOT: error:
-// CHECK: error: cannot compile this SEH __try yet
+// CHECK: error: cannot compile this SEH __leave yet
 // CHECK-NOT: error:

test/CodeGen/exceptions-seh.c

-// RUN: not %clang_cc1 -triple i686-pc-win32 -fexceptions -fms-extensions -emit-llvm -o - %s 2>&1 | FileCheck %s
+// RUN: %clang_cc1 %s -triple x86_64-pc-win32 -fexceptions -fms-extensions -emit-llvm -o - | FileCheck %s
 
-// This is a codegen test because we only emit the diagnostic when we start
-// generating code.
+// FIXME: Perform this outlining automatically CodeGen.
+void try_body(int numerator, int denominator, int *myres) {
+  *myres = numerator / denominator;
+}
+// CHECK-LABEL: define void @try_body(i32 %numerator, i32 %denominator, i32* %myres)
+// CHECK: sdiv i32
+// CHECK: store i32 %{{.*}}, i32*
+// CHECK: ret void
 
-int SaveDiv(int numerator, int denominator, int *res) {
+int safe_div(int numerator, int denominator, int *res) {
   int myres = 0;
+  int success = 1;
   __try {
-    myres = numerator / denominator;
-    __leave;
+    try_body(numerator, denominator, &myres);
   } __except (1) {
-    return 0;
+    success = -42;
   }
   *res = myres;
-  return 1;
+  return success;
+}
+// CHECK-LABEL: define i32 @safe_div(i32 %numerator, i32 %denominator, i32* %res)
+// CHECK: invoke void @try_body(i32 %{{.*}}, i32 %{{.*}}, i32* %{{.*}})
+// CHECK:       to label %{{.*}} unwind label %[[lpad:[^ ]*]]
+//
+// CHECK: [[lpad]]
+// CHECK: landingpad { i8*, i32 } personality i8* bitcast (i32 (...)* @__C_specific_handler to i8*)
+// CHECK: %[[sel:[^ ]*]] = load i32*
+// CHECK: %[[filt_id:[^ ]*]] = call i32 @llvm.eh.typeid.for(i8* bitcast (i32 (i8*, i8*)* @"\01?filt$0@0@safe_div@@" to i8*))
+// CHECK: %[[matches:[^ ]*]] = icmp eq i32 %[[sel]], %[[filt_id]]
+// CHECK: br i1 %[[matches]], label %[[except_bb:[^ ]*]], label %{{.*}}
+//
+// CHECK: [[except_bb]]
+// CHECK: call void @llvm.eh.seh.filter(i32 1)
+// CHECK: store i32 -42, i32* %[[success:[^ ]*]]
+//
+// CHECK: %[[res:[^ ]*]] = load i32* %[[success]]
+// CHECK: ret i32 %[[res]]
+
+void j(void);
+
+int nested_try() {
+  int r = 42;
+  __try {
+    __try {
+      j();
+      r = 0;
+    } __except(_exception_code() == 123) {
+      r = 123;
+    }
+  } __except(_exception_code() == 456) {
+    r = 456;
+  }
+  return r;
+}
+// CHECK-LABEL: define i32 @nested_try()
+// CHECK: store i32 42, i32* %[[r:[^ ,]*]]
+// CHECK: invoke void @j()
+// CHECK:       to label %[[cont:[^ ]*]] unwind label %[[lpad:[^ ]*]]
+//
+// CHECK: [[cont]]
+// CHECK: store i32 0, i32* %[[r]]
+// CHECK: br label %[[inner_try_cont:[^ ]*]]
+//
+// CHECK: [[lpad]]
+// CHECK: landingpad { i8*, i32 } personality i8* bitcast (i32 (...)* @__C_specific_handler to i8*)
+// CHECK: catch i8* bitcast (i32 (i8*, i8*)* @"\01?filt$1@0@nested_try@@" to i8*)
+// CHECK: catch i8* bitcast (i32 (i8*, i8*)* @"\01?filt$0@0@nested_try@@" to i8*)
+// CHECK: store i8* %{{.*}}, i8** %[[ehptr_slot:[^ ]*]]
+// CHECK: store i32 %{{.*}}, i32* %[[sel_slot:[^ ]*]]
+//
+// CHECK: load i32* %[[sel_slot]]
+// CHECK: call i32 @llvm.eh.typeid.for(i8* bitcast (i32 (i8*, i8*)* @"\01?filt$1@0@nested_try@@" to i8*))
+// CHECK: icmp eq i32
+// CHECK: br i1
+//
+// CHECK: load i32* %[[sel_slot]]
+// CHECK: call i32 @llvm.eh.typeid.for(i8* bitcast (i32 (i8*, i8*)* @"\01?filt$0@0@nested_try@@" to i8*))
+// CHECK: icmp eq i32
+// CHECK: br i1
+//
+// CHECK: load i8** %[[ehptr_slot]]
+// CHECK: call void @llvm.eh.seh.filter(i32 %{{.*}})
+// CHECK: store i32 456, i32* %[[r]]
+// CHECK: br label %[[outer_try_cont:[^ ]*]]
+//
+// CHECK: [[outer_try_cont]]
+// CHECK: %[[r_load:[^ ]*]] = load i32* %[[r]]
+// CHECK: ret i32 %[[r_load]]
+//
+// CHECK: load i8** %[[ehptr_slot]]
+// CHECK: call void @llvm.eh.seh.filter(i32 %{{.*}})
+// CHECK: store i32 123, i32* %[[r]]
+// CHECK: br label %[[inner_try_cont]]
+//
+// CHECK: [[inner_try_cont]]
+// CHECK: br label %[[outer_try_cont]]
+
+// FIXME: This lowering of __finally can't actually work, it will have to
+// change.
+static unsigned g = 0;
+void basic_finally() {
+  ++g;
+  __try {
+    j();
+  } __finally {
+    --g;
+  }
 }
-// CHECK-NOT: error:
-// CHECK: error: cannot compile this SEH __try yet
-// CHECK-NOT: error:
+// CHECK-LABEL: define void @basic_finally()
+// CHECK: load i32* @g
+// CHECK: add i32 %{{.*}}, 1
+// CHECK: store i32 %{{.*}}, i32* @g
+//
+// CHECK: invoke void @j()
+// CHECK:       to label %[[cont:[^ ]*]] unwind label %[[lpad:[^ ]*]]
+//
+// CHECK: [[cont]]
+// CHECK: load i32* @g
+// CHECK: add i32 %{{.*}}, -1
+// CHECK: store i32 %{{.*}}, i32* @g
+// CHECK: ret void
+//
+// CHECK: [[lpad]]
+// CHECK: landingpad { i8*, i32 } personality i8* bitcast (i32 (...)* @__C_specific_handler to i8*)
+// CHECK-NEXT: cleanup
+// CHECK: load i32* @g
+// CHECK: add i32 %{{.*}}, -1
+// CHECK: store i32 %{{.*}}, i32* @g
+// CHECK: resume

test/OpenMP/parallel_codegen.cpp

 // CHECK:       [[CONTEXT_ADDR:%.+]] = alloca %struct.anon*
 // CHECK:       store %struct.anon* %__context, %struct.anon** [[CONTEXT_ADDR]]
 // CHECK:       [[CONTEXT_PTR:%.+]] = load %struct.anon** [[CONTEXT_ADDR]]
 // CHECK-NEXT:  [[ARGC_PTR_REF:%.+]] = getelementptr inbounds %struct.anon* [[CONTEXT_PTR]], i32 0, i32 0
 // CHECK-NEXT:  [[ARGC_REF:%.+]] = load i32** [[ARGC_PTR_REF]]
 // CHECK-NEXT:  [[ARGC:%.+]] = load i32* [[ARGC_REF]]
 // CHECK-NEXT:  invoke void [[FOO:@.+foo.+]](i32{{[ ]?[a-z]*}} [[ARGC]])
 // CHECK:       ret void
-// CHECK:       call void @{{.+terminate.*}}(
+// CHECK:       call void @{{.+terminate.*|abort}}(
 // CHECK-NEXT:  unreachable
 // CHECK-NEXT:  }
 // CHECK-DEBUG-LABEL: define internal void @.omp_outlined.(i32* %.global_tid., i32* %.bound_tid., %struct.anon* %__context)
 // CHECK-DEBUG:       [[CONTEXT_ADDR:%.+]] = alloca %struct.anon*
 // CHECK-DEBUG:       store %struct.anon* %__context, %struct.anon** [[CONTEXT_ADDR]]
 // CHECK-DEBUG:       [[CONTEXT_PTR:%.+]] = load %struct.anon** [[CONTEXT_ADDR]]
 // CHECK-DEBUG-NEXT:  [[ARGC_PTR_REF:%.+]] = getelementptr inbounds %struct.anon* [[CONTEXT_PTR]], i32 0, i32 0
 // CHECK-DEBUG-NEXT:  [[ARGC_REF:%.+]] = load i32** [[ARGC_PTR_REF]]
 // CHECK-DEBUG-NEXT:  [[ARGC:%.+]] = load i32* [[ARGC_REF]]
 // CHECK-DEBUG-NEXT:  invoke void [[FOO:@.+foo.+]](i32 [[ARGC]])
 // CHECK-DEBUG:       ret void
-// CHECK-DEBUG:       call void @{{.+terminate.*}}(
+// CHECK-DEBUG:       call void @{{.+terminate.*|abort}}(
 // CHECK-DEBUG-NEXT:  unreachable
 // CHECK-DEBUG-NEXT:  }
 
 // CHECK-DAG: define linkonce_odr void [[FOO]]({{i32[ ]?[a-z]*}} %argc)
 // CHECK-DAG: declare void @__kmpc_fork_call(%ident_t*, i32, void (i32*, i32*, ...)*, ...)
 // CHECK-DEBUG-DAG: define linkonce_odr void [[FOO]](i32 %argc)
 // CHECK-DEBUG-DAG: declare void @__kmpc_fork_call(%ident_t*, i32, void (i32*, i32*, ...)*, ...)
 

 // CHECK:       [[CONTEXT_ADDR:%.+]] = alloca %struct.anon.0*
 // CHECK:       store %struct.anon.0* %__context, %struct.anon.0** [[CONTEXT_ADDR]]
 // CHECK:       [[CONTEXT_PTR:%.+]] = load %struct.anon.0** [[CONTEXT_ADDR]]
 // CHECK-NEXT:  [[ARGC_PTR_REF:%.+]] = getelementptr inbounds %struct.anon.0* [[CONTEXT_PTR]], i32 0, i32 0
 // CHECK-NEXT:  [[ARGC_REF:%.+]] = load i8**** [[ARGC_PTR_REF]]
 // CHECK-NEXT:  [[ARGC:%.+]] = load i8*** [[ARGC_REF]]
 // CHECK-NEXT:  invoke void [[FOO1:@.+foo.+]](i8** [[ARGC]])
 // CHECK:       ret void
-// CHECK:       call void @{{.+terminate.*}}(
+// CHECK:       call void @{{.+terminate.*|abort}}(
 // CHECK-NEXT:  unreachable
 // CHECK-NEXT:  }
 // CHECK-DEBUG-LABEL: define internal void @.omp_outlined.1(i32* %.global_tid., i32* %.bound_tid., %struct.anon.0* %__context)
 // CHECK-DEBUG:       [[CONTEXT_ADDR:%.+]] = alloca %struct.anon.0*
 // CHECK-DEBUG:       store %struct.anon.0* %__context, %struct.anon.0** [[CONTEXT_ADDR]]
 // CHECK-DEBUG:       [[CONTEXT_PTR:%.+]] = load %struct.anon.0** [[CONTEXT_ADDR]]
 // CHECK-DEBUG-NEXT:  [[ARGC_PTR_REF:%.+]] = getelementptr inbounds %struct.anon.0* [[CONTEXT_PTR]], i32 0, i32 0
 // CHECK-DEBUG-NEXT:  [[ARGC_REF:%.+]] = load i8**** [[ARGC_PTR_REF]]
 // CHECK-DEBUG-NEXT:  [[ARGC:%.+]] = load i8*** [[ARGC_REF]]
 // CHECK-DEBUG-NEXT:  invoke void [[FOO1:@.+foo.+]](i8** [[ARGC]])
 // CHECK-DEBUG:       ret void
-// CHECK-DEBUG:       call void @{{.+terminate.*}}(
+// CHECK-DEBUG:       call void @{{.+terminate.*|abort}}(
 // CHECK-DEBUG-NEXT:  unreachable
 // CHECK-DEBUG-NEXT:  }
 
 // CHECK: define linkonce_odr void [[FOO1]](i8** %argc)
 // CHECK-DEBUG: define linkonce_odr void [[FOO1]](i8** %argc)
 
 #endif

test/Sema/__try.c

-// RUN: %clang_cc1 -fborland-extensions -fsyntax-only -verify %s
+// RUN: %clang_cc1 -fborland-extensions -DBORLAND -fsyntax-only -verify %s
+// RUN: %clang_cc1 -fms-extensions -fsyntax-only -verify %s
 
 #define JOIN2(x,y) x ## y
 #define JOIN(x,y) JOIN2(x,y)
 #define TEST2(name) JOIN(name,__LINE__)
 #define TEST TEST2(test)
 typedef int DWORD;
 
 #pragma sysheader begin
 
 struct EXCEPTION_INFO{};
 
-int __exception_code();
+unsigned long __exception_code();
+#ifdef BORLAND
 struct EXCEPTION_INFO* __exception_info();
+#endif
 void __abnormal_termination();
 
 #define GetExceptionCode __exception_code
 #define GetExceptionInformation __exception_info
 #define AbnormalTermination __abnormal_termination
 
 #pragma sysheader end
 

 void TEST() {
   int function_scope = 0;
   __try {
 
   }
   __except( function_scope ? 1 : -1 ) {}
 }
 
+#ifdef BORLAND
 void TEST() {
+  (void)__abnormal_termination(); // expected-error{{only allowed in __finally block}}
+  (void)AbnormalTermination();  // expected-error{{only allowed in __finally block}}
+
   __try {
     (void)AbnormalTermination;  // expected-error{{only allowed in __finally block}}
     (void)__abnormal_termination; // expected-error{{only allowed in __finally block}}
   }
   __except( 1 ) {
     (void)AbnormalTermination;  // expected-error{{only allowed in __finally block}}
     (void)__abnormal_termination; // expected-error{{only allowed in __finally block}}
   }
 
   __try {
   }
   __finally {
     AbnormalTermination();
     __abnormal_termination();
   }
 }
+#endif
 
 void TEST() {
-  (void)__exception_code;       // expected-error{{only allowed in __except block}}
-  (void)__exception_info;       // expected-error{{only allowed in __except filter expression}}
-  (void)__abnormal_termination; // expected-error{{only allowed in __finally block}}
-
-  (void)GetExceptionCode();     // expected-error{{only allowed in __except block}}
+  (void)__exception_info();       // expected-error{{only allowed in __except filter expression}}
   (void)GetExceptionInformation(); // expected-error{{only allowed in __except filter expression}}
-  (void)AbnormalTermination();  // expected-error{{only allowed in __finally block}}
+}
+
+void TEST() {
+#ifndef BORLAND
+  (void)__exception_code;     // expected-error{{builtin functions must be directly called}}
+#endif
+  (void)__exception_code();     // expected-error{{only allowed in __except block or filter expression}}
+  (void)GetExceptionCode();     // expected-error{{only allowed in __except block or filter expression}}
+}
+
+void TEST() {
+  __try {
+  } __except(1) {
+    GetExceptionCode(); // valid
+    GetExceptionInformation(); // expected-error{{only allowed in __except filter expression}}
+  }
 }
 
 void test_seh_leave_stmt() {
   __leave; // expected-error{{'__leave' statement not in __try block}}
 
   __try {
     __leave;
     __leave 4; // expected-error{{expected ';' after __leave statement}}
   } __except(1) {
     __leave; // expected-error{{'__leave' statement not in __try block}}
   }
 
   __try {
     __leave;
   } __finally {
     __leave; // expected-error{{'__leave' statement not in __try block}}
   }
   __leave; // expected-error{{'__leave' statement not in __try block}}
 }
-