Implement Attribute Target MultiVersioning (Improved edition!)

[erichkeane]

Summary

GCC's attribute 'target', in addition to being an optimization hint,
also allows function multiversioning. We currently have the former
implemented, this is the latter's implementation.

This works by enabling functions with the same name/signature to coexist,
so that they can all be emitted. Multiversion state is stored in the
FunctionDecl itself, and SemaDecl manages the definitions.
Note that it ends up having to permit redefinition of functions so
that they can all be emitted. Additionally, all versions of the function
must be emitted, so this also manages that.

Note that this includes some additional rules that GCC does not, since
defining something as a MultiVersion function after a usage has been made illegal.

The only 'history rewriting' that happens is if a function is emitted before
it has been converted to a multiversion'ed function, at which point its name
needs to be changed.

Function templates and virtual functions are NOT yet supported (not supported
in GCC either).

This SEMA design was discussed with @rsmith but additional opinions/preferences
here are greatly appreciated.

Options on how to split this patch up would also be particularly solicited,
since this IS a large patch.

This patch completely superceeds: https://reviews.llvm.org/D38596

[craig.topper]

include/clang/Basic/X86Target.def

148

Cannonlake can prioritize off of VBMI or IFMA.

236

Should we document that not all of these are prioritized by gcc? So the next person who tries to match gcc for new features knows this.

[rsmith]

I expect there are more places that need to be changed, but this is looking surprisingly clean.

You will need to teach the modules merging code that it needs to check this attribute in addition to checking that types match when considering merging declaration chains for functions. (See isSameEntity in Serialization/ASTReaderDecl.cpp).

I don't see any tests for non-call uses of multiversioned functions (eg, taking their address, binding function references to them). Does that work? (I'd expect you to need "ignore non-default versions" logic in more places for that.)

include/clang/AST/Decl.h

1753–1755

This flag needs serialization support for PCH / modules, and you'll need to do something if we find declarations with inconsistent values for the flag across modules (detect and diagnose, probably), and likewise in chained PCH (where the right thing to do is presumably to update all prior declarations to have the same value for the flag).

include/clang/Basic/Attr.td

1847

Our normal convention is to not put a space before == here.

include/clang/Basic/DiagnosticSemaKinds.td

9336–9337

There's some weird indentation hereabouts. Our normal convention is to put the Error< on the def line, and start the diagnostic text as a 2-space-indented string literal on the next line.

include/clang/Basic/TargetInfo.h

934

return 0;

lib/CodeGen/CodeGenModule.cpp

2143

You need to call getRedeclContext() on the value returned by getDeclContext() to skip "transparent" DeclContexts such as LinkageSpecDecl and C++ Modules TS export declarations.

2143

getIdentifier here is wrong; use getDeclName() instead, to correctly handle non-identifier names.

2146

You should skip functions you've already seen somewhere around here; we don't need to handle the same function multiple times.

lib/Sema/SemaDecl.cpp

9196

Maybe also check that the language linkage matches (extern "C" and extern "C++" could imply different calling conventions, even though they don't on any of our current targets).

9196–9198

Would it be possible to factor out the checks in MergeFunctionDecl that you're using here and reuse them directly?

9253–9254

This won't do the right thing for C++14 deduced return types. I'm not completely sure what the right thing is, though -- clearly we need to deduce the same return type in all versions of the function, but it's less clear whether we should also require the return-type-as-written to match across versions (as is required across redeclarations). Example:

VERSION_FOO auto f() { return 0; } // declared return type is `auto`, actual return type is `int`
VERSION_BAR auto f(); // will fail the current check, because the return type `auto` doesn't match the prior return type `int`
VERSION_BAR auto f() { return 0.0; } // should reject this because we deduced a different return type than on the VERSION_FOO version?

Perhaps the simplest thing would be to simply disallow deduced return types for multiversioned functions entirely for now?

9301–9302

Should we reject (eg) __attribute__((target("default"))) on main?

9305

Typo "than"

9329–9330

Can this happen if NewFD is not main?

9409–9410

We should be a little careful here: we don't want to allow multiversioning between declarations in different semantic DeclContexts:

VERSION_DEFAULT void f();
namespace N {
  using ::f;
  VERSION_FOO void f();
}

... because when we generate the dispatcher, we won't find the right set of versions. The easiest way to deal with this might be to check near the start of this function whether the semantic DeclContext of CurFD and OldFD are equal, and if not, just bail out and leave the redeclaration merging code to diagnose the conflict. (We already prevent a DeclContext from containing two non-overloadable functions where one of them is 'using'd from another DeclContext and the other is declared locally.)

[erichkeane]

Incoming patch!

include/clang/Basic/Attr.td

1847

Clang format seems to keep adding it interestingly enough. Perhaps because it is a .td file? Will remove.

include/clang/Basic/DiagnosticSemaKinds.td

9336–9337

Thanks! Clang-format really makes a mess of these too...

lib/CodeGen/CodeGenModule.cpp

2146

Will the lookup call give me duplicates? I need to check each at least once through this so that I can put them into "Options" so that I can emit them during the resolver.

Otherwise, I'm not terribly sure what you mean.

lib/Sema/SemaDecl.cpp

9196–9198

I'd thought about that, but I'm being WAAY more strict, so I didn't see a good way to factor them out without making things pretty complicated. Suggestions welcome :)

9253–9254

I think disallowing 'auto' return for now is a good idea. This next patch is using isUndeducedType, which I think is sufficient, but if there is a better function I'm missing, please let me know.

[erichkeane]

Fix all rsmith's and craig's fixes, AFAIK.

[erichkeane]

Ping! Anyone have time to take a look for me?

[aaron.ballman]

include/clang/Basic/Attr.td

1847

This function should be marked const.

include/clang/Basic/DiagnosticSemaKinds.td

9350–9352

Diagnostics are not complete sentences, so this should be reworded to be even less grammatically correct. ;-)

9356–9357

This worries me slightly. Is there a benefit to prohibiting this attribute with any other attribute? For instance, I'm thinking about a multiversioned noreturn function or a multiversioned function with a calling convention as plausible use cases.

9366

How about 'main' cannot be a multiversion function?

lib/CodeGen/CodeGenFunction.cpp

2311–2312

You can use llvm::for_each instead.

lib/CodeGen/CodeGenFunction.h

4000

Feat is already a StringRef?

[aaron.ballman]

lib/CodeGen/CodeGenModule.cpp

750

Might as well sink this into the call to std::sort().

lib/Sema/SemaDecl.cpp

9168

Don't use auto here as the type is not explicitly spelled out in the initialization.

9196

This function uses quite a few magic numbers that might be better expressed with named values instead.

9232–9234

Formatting is off here.

9233

Can use const auto * here.

9251

Can use const auto * here.

9318

So any other form of target attribute and feature string is fine to place on main()?

9348

Do not use auto here.

9364

Capitalization.

9432

Do not use auto here.

9447

Given how often the features need to be sorted, would it make sense to hoist this functionality into parse()?

[erichkeane]

Patch incoming, Thank you very much for the review @aaron.ballman

include/clang/Basic/DiagnosticSemaKinds.td

9350–9352

I tried again, hopefully this one is better? :) Let me know if my grammar is still too good...

9356–9357

This attribute can have negative effects if combined with certain others, though I don't have a complete list. GCC seems to 'ignore' a handful of others, but I don't have the complete list. It is my intent to disallow all others in this patch, and upon request/further work, permit them as an opt-in.

lib/Sema/SemaDecl.cpp

9318

It IS, it just doesn't cause multiversioning. The "return false" at the bottom of this condition returns to CheckFunctionDeclaration.

[erichkeane]

Fixed all of @aaron.ballman comments.

[erichkeane]

@aaron.ballman reminded me that this should only work on ELF targets, so this patch enforces that constraint.

[aaron.ballman]

include/clang/Basic/DiagnosticSemaKinds.td

9350

instead of duplicate, do you mean redeclarations? Or do you mean overloaded declarations? Both?

9350–9352

You've achieved sufficiently incorrect grammar. ;-)

[erichkeane]

include/clang/Basic/DiagnosticSemaKinds.td

9350

I do mean redeclarations, thanks! Change incoming.

[erichkeane]

Fix error message per @aaron.ballman s suggestion.

[erichkeane]

I'm sure most of you have taken off for the year, but if anyone had time, *ping* :D

[erichkeane]

Hi all-- I'm intending to miss the branch for 6.0, but I'd love to get this in soon after. Can anyone take another look?

Thanks,
Erich

[echristo]

Couple of inline comments, otherwise I'm pretty happy. I'd wait for an ack by Richard for this though.

-eric

lib/CodeGen/CGBuiltin.cpp

7673 ↗

(On Diff #127005)

Why do you need to pass in a Builder?

lib/CodeGen/CodeGenFunction.cpp

2328

Can you get here via trying to compile code for another cpu?

[rsmith]

Lots of comments, but no high-level design concerns. I think this is very close to being ready to go.

include/clang/AST/Decl.h

2164–2170

Instead of looping over redeclarations here, how about only storing the flag on the canonical declaration (and only looking there in isMultiVersion)?

include/clang/Basic/Attr.td

1847

Nit: space before {.

include/clang/Basic/DiagnosticSemaKinds.td

9336–9337

Is it "multiversion function" or "multiversioned function"? You're using both in these diagnostics; please pick one and use it consistently. I prefer the "-ed" form, but I'm happy with whichever you'd prefer.

9339

"multiversioning" maybe? (Again, this is used inconsistently; there's a "function multiversioning" in err_multiversion_not_supported).

9346–9348

Maybe we should just suppress the "candidate" note entirely for these cases?

lib/CodeGen/CodeGenModule.cpp

898–899

I'm not especially enamoured with getMangledName mutating the IR. Can we perform this rename as part of emitting the multiversion dispatcher or ifunc, rather than here? (Or do we really not have anywhere else that this can live?)

2061–2062

Please add a comment explaining this check. (I'm guessing the issue is that we can't form a cross-translation-unit reference to the right function from an IFUNC, so we can't rely on an available_externally definition actually being usable? But it's not clear to me that this is the right resolution to that, especially in light of the dllexport case below where it would not be correct to emit the definition.)

2143–2146

Consider moving the lookup, loop and type check here into an ASTContext or FunctionDecl utility to find or visit all the versions of a multiversioned function.

2145

if statements with nontrivial bodies should have their own braces.

2146

Yes, the lookup can in some cases find multiple declarations from the same redeclaration chain. (This happens particularly when the declarations are loaded from AST files.)

lib/Sema/SemaDecl.cpp

9396

Do not hardcode English text as a diagnostic parameter. And don't reuse a diagnostic to mean something else :) The %0 in this diagnostic is for the name of the declaration.

diag::note_previous_declaration seems to capture what you're actually looking for here. (It's a note for "previous declaration of the same entity", whereas note_previous_decl is a note for "here's some previous declaration of something else that's relevant").

9401–9402

This doesn't seem right: the OldFD isn't (necessarily) a declaration without a prototype.

9423

As above.

9477

As above.

9630

The parameter in CheckMultiVersionFunction corresponding to MergeTypeWithPrevious here is called MayNeedOverloadableChecks, which is also the name of a local variable in this function. Did you pass the wrong bool? Or is the name of the parameter to CheckMultiVersionFunction wrong?

lib/Sema/SemaOverload.cpp

5963

Please move this comparison of getFeaturesStr() against "default" into a dedicated function on TargetAttr (::isDefaultVersion()?). This magic string literal is appearing quite a lot in this patch...

lib/Serialization/ASTReaderDecl.cpp

2821

This comment is missing a

2825

This assertion will fire if X is multiversion but Y is not. It's probably not too crucial what happens in that case, but we shouldn't assert. An error would be nice, but we can't produce one from here, so how about we just say that if one is multiversion and the other is not, that they're considered to not be the same entity.

2827

Space after if, please.

test/CodeGen/attr-target-mv-func-ptrs.c

11–16

What about uses in contexts where there is no target function type? For example, +foo;

test/CodeGenCXX/attr-target-mv-member-funcs.cpp

4–9

OK, multiversioned member functions! Let's look at some nasty corner cases!

Do you allow multiversioning of special member functions (copy constructor, destructor, ...)? Some tests for that would be interesting. Note in particular that CXXRecordDecl::getDestructor assumes that there is only one destructor for a class, and I expect we make that assumption in a bunch of other places too. Might be best to disallow multiversioning destructors for now.

Do you allow a multiversioned function to have one defaulted version and one non-defaulted version? What does that mean for the properties of the class that depend on whether special member functions are trivial? Might be a good idea to disallow defaulting a multiversioned function. Oh, and we should probably not permit versions of a multiversioned function to be deleted either.

If you allow multiversioning of constructors, I'd like to see a test for multiversioning of inherited constructors. Likewise, if you allow multiversioning of conversion functions, I'd like to see a test for that. (I actually think there's a very good chance both of those will just work fine.)

You don't allow multiversioning of function templates right now, but what about multiversioning of member functions of a class template? Does that work? If so, does class template argument deduction using multiversioned constructors work?

Does befriending multiversioned functions work? Is the target attribute taken into account?

[erichkeane]

Patch incoming, sorry it took so long!

lib/CodeGen/CGBuiltin.cpp

7673 ↗

(On Diff #127005)

Because when I first wrote this, the EmitResolver function was a free function. I guess I missed this dependency along the way, thanks for the catch!

lib/CodeGen/CodeGenFunction.cpp

2328

At the moment, no. I'm asserting to make sure that is the case, and to be a hint for George/et-al who are implementing this in the future for other processors.

lib/CodeGen/CodeGenModule.cpp

898–899

Unfortunately it is too late at that point. The problem is you have:

TARGET_SSE MVFunc();
TARGET_DEF MVFunc(); // At this point, a mangling-conflict occurs.

void foo() {
MVFunc(); // Only at THIS point does the IFunc get created, too late to rewrite the SSE variant's name.
}

That said, I moved it to a more appropriate place.

2061–2062

The actual case I encountered was that when emitting the global in emitMultiVersionFunctions, the EmitGlobalDefinition was immediately skipping it here because it was an inline function.

I believe the correct response is to just call EmitGlobalFunctionDefinition from the emitMultiVersionFunctions handler. This will have to be modified when I support virtual functions or constructors, but this will make it work.

IMO, the cross-TU issue you come up with would only be an issue when the value isn't emitted due to it being inline/static/etc. In this case, the failed linking is an acceptable consequence to the user improperly providing a multiversion variant.

lib/Sema/SemaDecl.cpp

9630

Looks like I'd just grabbed the wrong name for the parameter in CheckMultiVersionFunction. Fixed!

test/CodeGen/attr-target-mv-func-ptrs.c

11–16

Added as a test in Sema/attr-target-mv.c.

test/CodeGenCXX/attr-target-mv-member-funcs.cpp

4–9

I gave CTORs a try, and found that there are a few subtle issues that need to be dealt with in code-gen. For the moment (and since GCC simply terminates if you try to use them), I'd like to disallow them.

Definitely going to disallow dtors/default/deleted functions.

[erichkeane]

Fixes for all @echristo and @rsmith s comments.

[erichkeane]

Forgot friend functions, Working on it now, sorry about that.

[erichkeane]

Added test for friend functions.

[erichkeane]

Sorry for the thrash, thought of a better way to do the test just as I clicked 'save' last time :)

[rsmith]

One other test I'd like to see is what happens if you declare a version before the first use, and define it afterwards, particularly if the version is an inline function:

inline __attribute__((target("default"))) void f();
inline __attribute__((target("foo"))) void f();
void g() { f(); }
__attribute__((target("default"))) void f() {}
__attribute__((target("foo"))) void f() {}

include/clang/AST/ASTContext.h

2643–2648

This should deal with the case where lookup finds multiple declarations of the same version (which can happen in particular when serializing to AST files; we keep around a handle to the version from each AST file). Also, consider moving this to the .cpp file (use llvm::function_ref<void(const FunctionDecl*)> to pass in the callback).

include/clang/Basic/Attr.td

1847

Space before { seems to have been removed again. Clang-format really doesn't like tablegen files :)

include/clang/Basic/DiagnosticSemaKinds.td

9354–9356

"multiversioned function declaration has a different destructors" doesn't sound right to me :)

lib/Sema/SemaDecl.cpp

9228–9229

It seems strange to have a "caused here" note pointing at the same location that you just produced an error on. Is this note really adding value? (Likewise on all later diagnostics that add this note at the same place as the error.)

lib/Sema/SemaOverload.cpp

10220

Couple of typos here.

lib/Serialization/ASTReaderDecl.cpp

2825

This is still not quite right for the case of a multiversion / non-multiversion mismatch: you don't return 'false' when Y is multiversion and X is not.

test/CodeGenCXX/attr-target-mv-constexpr.cpp

1–16

Can you move this to test/SemaCXX and change it to only test the constant evaluation side of things (eg, static_assert(foo() == 2))?

[erichkeane]

patch incoming!

include/clang/AST/ASTContext.h

2643–2648

Ah, right, this part got dropped when I was moving it around...

There is only 2 ways I could think of to do this properly, first is to actually check the 'target' strings, second is to check the address of the canonical decl. Checking target strings seems expensive, but I don't know if the pointer values are sufficient?

I'll implement the 2nd for this patch, but if you could give me a hint as the proper way (if this isn't sufficient), I'd be grateful.

lib/Sema/SemaDecl.cpp

9228–9229

I think that makes sense. I'll remove all of those diagnostics where the error and the 'caused by' are different lines.

[erichkeane]

All of @rsmith 's comments on the last patch. 1 "open" item on the duplicate discover in the resolver generation, but hopefully this otherwise acceptable.

[rsmith]

Looks good to me with just a few more tweaks (assuming these comments don't uncover any new issues). Thank you!

lib/AST/ASTContext.cpp

9490 ↗

(On Diff #128950)

Consider using SmallDenseSet here. (In the modules case, we could potentially have hundreds or even thousands of lookup results, and the better asymptotics in the bad cases seems on balance worth the higher constant factor.)

Also, you'll need to add declarations to this set :)

A testcase for this would look something like:

#pragma clang module build A
module A {}
#pragma clang module contents
#pragma clang module begin A
__attribute__((target("default"))) void f();
__attribute__((target("foo"))) void f();
#pragma clang module end
#pragma clang module endbuild

#pragma clang module build B
module B {}
#pragma clang module contents
#pragma clang module begin B
__attribute__((target("default"))) void f();
__attribute__((target("foo"))) void f();
#pragma clang module end
#pragma clang module endbuild

#pragma clang module import A
#pragma clang module import B
void g() { f(); }

... with a check that the resolver ifunc only includes each case once.

lib/Serialization/ASTReaderDecl.cpp

2832

The !TAY check here is now redundant and can be moved into the assertion.

test/CodeGen/attr-target-mv.c

21–27

This will either need to be a C++ test or will need inline on the definitions to have the intended effect of creating a discardable function definition.

test/CodeGenCXX/attr-target-mv-member-funcs.cpp

4–9

Can you also test that out-of-line definitions of multiversioned member functions are properly matched with the in-class declaration, and that you can't add new versions not present in the class definition that way? (Again, I'm not expecting any problems here, but it seems like something we should have coverage for.)

[erichkeane]

Closed with revision 322028

include/clang/AST/ASTContext.h

public:
   /// \brief Determines if the decl can be CodeGen'ed or deserialized from PCH
   /// lazily, only when used; this is only relevant for function or file scoped
   /// var definitions.
   ///
   /// \returns true if the function/var must be CodeGen'ed/deserialized even if
   /// it is not used.
   bool DeclMustBeEmitted(const Decl *D);
 
+  /// \brief Visits all versions of a multiversioned function with the passed
+  /// predicate.
+  template <typename Pred>
+  void forEachMultiversionedFunctionVersion(const FunctionDecl *FD,
+                                            Pred P) const {
+    assert(FD->isMultiVersion() && "Only valid for multiversioned functions");
+    FD = FD->getCanonicalDecl();
+    for (auto *CurDecl :
+         FD->getDeclContext()->getRedeclContext()->lookup(FD->getDeclName())) {
+      FunctionDecl *CurFD = CurDecl->getAsFunction()->getCanonicalDecl();
+      if (CurFD && hasSameType(CurFD->getType(), FD->getType()))
+        P(CurFD);
+    }
+  }
+
   const CXXConstructorDecl *
   getCopyConstructorForExceptionObject(CXXRecordDecl *RD);
 
   void addCopyConstructorForExceptionObject(CXXRecordDecl *RD,
                                             CXXConstructorDecl *CD);
 
   void addTypedefNameForUnnamedTagDecl(TagDecl *TD, TypedefNameDecl *TND);
 

include/clang/AST/Decl.h

private:
   /// \brief Indicates if the function was a definition but its body was
   /// skipped.
   unsigned HasSkippedBody : 1;
 
   /// Indicates if the function declaration will have a body, once we're done
   /// parsing it.
   unsigned WillHaveBody : 1;
 
+  /// Indicates that this function is a multiversioned function using attribute
+  /// 'target'.
+  unsigned IsMultiVersion : 1;
+
 protected:
   /// [C++17] Only used by CXXDeductionGuideDecl. Declared here to avoid
   /// increasing the size of CXXDeductionGuideDecl by the size of an unsigned
   /// int as opposed to adding a single bit to FunctionDecl.
   /// Indicates that the Deduction Guide is the implicitly generated 'copy
   /// deduction candidate' (is used during overload resolution).
   unsigned IsCopyDeductionCandidate : 1;
 
FunctionDecl(Kind DK, ASTContext &C, DeclContext *DC, SourceLocation StartLoc,
         DeclContext(DK), redeclarable_base(C), SClass(S),
         IsInline(isInlineSpecified), IsInlineSpecified(isInlineSpecified),
         IsExplicitSpecified(false), IsVirtualAsWritten(false), IsPure(false),
         HasInheritedPrototype(false), HasWrittenPrototype(true),
         IsDeleted(false), IsTrivial(false), IsDefaulted(false),
         IsExplicitlyDefaulted(false), HasImplicitReturnZero(false),
         IsLateTemplateParsed(false), IsConstexpr(isConstexprSpecified),
         InstantiationIsPending(false), UsesSEHTry(false), HasSkippedBody(false),
-        WillHaveBody(false), IsCopyDeductionCandidate(false), HasODRHash(false),
-        ODRHash(0), EndRangeLoc(NameInfo.getEndLoc()),
-        DNLoc(NameInfo.getInfo()) {}
+        WillHaveBody(false), IsMultiVersion(false),
+        IsCopyDeductionCandidate(false), HasODRHash(false), ODRHash(0),
+        EndRangeLoc(NameInfo.getEndLoc()), DNLoc(NameInfo.getInfo()) {}
 
   using redeclarable_base = Redeclarable<FunctionDecl>;
 
   FunctionDecl *getNextRedeclarationImpl() override {
     return getNextRedeclaration();
   }
 
   FunctionDecl *getPreviousDeclImpl() override {
public:
   /// \brief True if the function was a definition but its body was skipped.
   bool hasSkippedBody() const { return HasSkippedBody; }
   void setHasSkippedBody(bool Skipped = true) { HasSkippedBody = Skipped; }
 
   /// True if this function will eventually have a body, once it's fully parsed.
   bool willHaveBody() const { return WillHaveBody; }
   void setWillHaveBody(bool V = true) { WillHaveBody = V; }
 
+  /// True if this function is considered a multiversioned function.
+  bool isMultiVersion() const { return getCanonicalDecl()->IsMultiVersion; }
+
+  /// Sets the multiversion state for this declaration and all of its
+  /// redeclarations.
+  void setIsMultiVersion(bool V = true) {
+    getCanonicalDecl()->IsMultiVersion = V;
+  }
+
   void setPreviousDeclaration(FunctionDecl * PrevDecl);
 
   FunctionDecl *getCanonicalDecl() override;
   const FunctionDecl *getCanonicalDecl() const {
     return const_cast<FunctionDecl*>(this)->getCanonicalDecl();
   }
 
   unsigned getBuiltinID() const;
 

include/clang/Basic/Attr.td

def Target : InheritableAttr {
   let Args = [StringArgument<"featuresStr">];
   let Subjects = SubjectList<[Function], ErrorDiag>;
   let Documentation = [TargetDocs];
   let AdditionalMembers = [{
     struct ParsedTargetAttr {
       std::vector<std::string> Features;
       StringRef Architecture;
       bool DuplicateArchitecture = false;
+      bool operator==(const ParsedTargetAttr &Other) const{
+        return DuplicateArchitecture == Other.DuplicateArchitecture &&
+               Architecture == Other.Architecture && Features == Other.Features;
+      }
     };
     ParsedTargetAttr parse() const {
       return parse(getFeaturesStr());
     }
+
+    template<class Compare>
+    ParsedTargetAttr parse(Compare cmp) const {
+      ParsedTargetAttr Attrs = parse();
+      std::sort(std::begin(Attrs.Features), std::end(Attrs.Features), cmp);
+      return Attrs;
+    }
+
+    bool isDefaultVersion() const { return getFeaturesStr() == "default"; }
+
     static ParsedTargetAttr parse(StringRef Features) {
       ParsedTargetAttr Ret;
+      if (Features == "default") return Ret;
       SmallVector<StringRef, 1> AttrFeatures;
       Features.split(AttrFeatures, ",");
 
       // Grab the various features and prepend a "+" to turn on the feature to
       // the backend and add them to our existing set of features.
       for (auto &Feature : AttrFeatures) {
         // Go ahead and trim whitespace rather than either erroring or
         // accepting it weirdly.

include/clang/Basic/DiagnosticSemaKinds.td

def ext_warn_gnu_final : ExtWarn<
   "__final is a GNU extension, consider using C++11 final">,
   InGroup<GccCompat>;
 
 def warn_shadow_field :
   Warning<"non-static data member '%0' of '%1' shadows member inherited from type '%2'">,
   InGroup<ShadowField>, DefaultIgnore;
 def note_shadow_field : Note<"declared here">;
 
+def err_target_required_in_redecl : Error<
+  "function declaration is missing 'target' attribute in a multiversioned "
+  "function">;
+def note_multiversioning_caused_here : Note<
+  "function multiversioning caused by this declaration">;
+def err_multiversion_after_used : Error<
+  "function declaration cannot become a multiversioned function after first "
+  "usage">;
+def err_bad_multiversion_option : Error<
+  "function multiversioning doesn't support %select{feature|architecture}0 "
+  "'%1'">;
+def err_multiversion_duplicate : Error<
+  "multiversioned function redeclarations require identical target attributes">;
+def err_multiversion_noproto : Error<
+  "multiversioned function must have a prototype">;
+def err_multiversion_no_other_attrs : Error<
+  "attribute 'target' multiversioning cannot be combined with other "
+  "attributes">;
+def err_multiversion_diff : Error<
+  "multiversioned function declaration has a different %select{calling convention"
+  "|return type|constexpr specification|inline specification|storage class|"
+  "linkage|destructors}0">;
+def err_multiversion_doesnt_support : Error<
+  "multiversioned functions do not yet support %select{function templates|"
+  "virtual functions|deduced return types|constructors|destructors|"
+  "deleted functions|defaulted functions}0">;
+def err_multiversion_not_allowed_on_main : Error<
+  "'main' cannot be a multiversioned function">;
+def err_multiversion_not_supported : Error<
+ "function multiversioning is not supported on the current target">;
+
 } // end of sema component.

include/clang/Basic/TargetInfo.h

virtual bool handleTargetFeatures(std::vector<std::string> &Features,
     return true;
   }
 
   /// \brief Determine whether the given target has the given feature.
   virtual bool hasFeature(StringRef Feature) const {
     return false;
   }
 
+  /// \brief Identify whether this taret supports multiversioning of functions,
+  /// which requires support for cpu_supports and cpu_is functionality.
+  virtual bool supportsMultiVersioning() const { return false; }
+
   // \brief Validate the contents of the __builtin_cpu_supports(const char*)
   // argument.
   virtual bool validateCpuSupports(StringRef Name) const { return false; }
 
+  // \brief Return the target-specific priority for features/cpus/vendors so
+  // that they can be properly sorted for checking.
+  virtual unsigned multiVersionSortPriority(StringRef Name) const {
+    return 0;
+  }
+
   // \brief Validate the contents of the __builtin_cpu_is(const char*)
   // argument.
   virtual bool validateCpuIs(StringRef Name) const { return false; }
 
   // \brief Returns maximal number of args passed in registers.
   unsigned getRegParmMax() const {
     assert(RegParmMax < 7 && "RegParmMax value is larger than AST can handle");
     return RegParmMax;

include/clang/Basic/X86Target.def

 //===--- X86Target.def - X86 Feature/Processor Database ---------*- C++ -*-===//
 //
 //                     The LLVM Compiler Infrastructure
 //
 // This file is distributed under the University of Illinois Open Source
 // License. See LICENSE.TXT for details.
 //
 //===----------------------------------------------------------------------===//
 //
 // This file defines the X86-specific Features and Processors, as used by
 // the X86 Targets.
 //
 //===----------------------------------------------------------------------===//
 
+#ifndef PROC_WITH_FEAT
+#define PROC_WITH_FEAT(ENUM, STRING, IS64BIT, KEYFEATURE)                      \
+  PROC(ENUM, STRING, IS64BIT)
+#endif
 
 #ifndef PROC
 #define PROC(ENUM, STRING, IS64BIT)
 #endif
 
 #ifndef PROC_ALIAS
 #define PROC_ALIAS(ENUM, ALIAS)
 #endif
 
+#ifndef FEATURE
+#define FEATURE(ENUM)
+#endif
+
 #define PROC_64_BIT true
 #define PROC_32_BIT false
 
 /// \name i386
 /// i386-generation processors.
 //@{
 PROC(i386, "i386", PROC_32_BIT)
 //@}

 
 PROC(Prescott, "prescott", PROC_32_BIT)
 PROC(Nocona, "nocona", PROC_64_BIT)
 //@}
 
 /// \name Core
 /// Core microarchitecture based processors.
 //@{
-PROC(Core2, "core2", PROC_64_BIT)
+PROC_WITH_FEAT(Core2, "core2", PROC_64_BIT, FEATURE_SSSE3)
 
 /// This enumerator, like Yonah, is a bit odd. It is another
 /// codename which GCC no longer accepts as an option to -march, but Clang
 /// has some logic for recognizing it.
 // FIXME: Warn, deprecate, and potentially remove this.
 PROC(Penryn, "penryn", PROC_64_BIT)
 //@}
 
 /// \name Atom
 /// Atom processors
 //@{
-PROC(Bonnell, "bonnell", PROC_64_BIT)
+PROC_WITH_FEAT(Bonnell, "bonnell", PROC_64_BIT, FEATURE_SSSE3)
 PROC_ALIAS(Bonnell, "atom")
 
-PROC(Silvermont, "silvermont", PROC_64_BIT)
+PROC_WITH_FEAT(Silvermont, "silvermont", PROC_64_BIT, FEATURE_SSE4_2)
 PROC_ALIAS(Silvermont, "slm")
 
 PROC(Goldmont, "goldmont", PROC_64_BIT)
 //@}
 
 /// \name Nehalem
 /// Nehalem microarchitecture based processors.
-PROC(Nehalem, "nehalem", PROC_64_BIT)
+PROC_WITH_FEAT(Nehalem, "nehalem", PROC_64_BIT, FEATURE_SSE4_2)
 PROC_ALIAS(Nehalem, "corei7")
 
 /// \name Westmere
 /// Westmere microarchitecture based processors.
-PROC(Westmere, "westmere", PROC_64_BIT)
+PROC_WITH_FEAT(Westmere, "westmere", PROC_64_BIT, FEATURE_PCLMUL)
 
 /// \name Sandy Bridge
 /// Sandy Bridge microarchitecture based processors.
-PROC(SandyBridge, "sandybridge", PROC_64_BIT)
+PROC_WITH_FEAT(SandyBridge, "sandybridge", PROC_64_BIT, FEATURE_AVX)
 PROC_ALIAS(SandyBridge, "corei7-avx")
 
 /// \name Ivy Bridge
 /// Ivy Bridge microarchitecture based processors.
-PROC(IvyBridge, "ivybridge", PROC_64_BIT)
+PROC_WITH_FEAT(IvyBridge, "ivybridge", PROC_64_BIT, FEATURE_AVX)
 PROC_ALIAS(IvyBridge, "core-avx-i")
 
 /// \name Haswell
 /// Haswell microarchitecture based processors.
-PROC(Haswell, "haswell", PROC_64_BIT)
+PROC_WITH_FEAT(Haswell, "haswell", PROC_64_BIT, FEATURE_AVX2)
 PROC_ALIAS(Haswell, "core-avx2")
 
 /// \name Broadwell
 /// Broadwell microarchitecture based processors.
-PROC(Broadwell, "broadwell", PROC_64_BIT)
+PROC_WITH_FEAT(Broadwell, "broadwell", PROC_64_BIT, FEATURE_AVX2)
 
 /// \name Skylake Client
 /// Skylake client microarchitecture based processors.
-PROC(SkylakeClient, "skylake", PROC_64_BIT)
+PROC_WITH_FEAT(SkylakeClient, "skylake", PROC_64_BIT, FEATURE_AVX2)
 
 /// \name Skylake Server
 /// Skylake server microarchitecture based processors.
-PROC(SkylakeServer, "skylake-avx512", PROC_64_BIT)
+PROC_WITH_FEAT(SkylakeServer, "skylake-avx512", PROC_64_BIT, FEATURE_AVX512F)
 PROC_ALIAS(SkylakeServer, "skx")
 
 /// \name Cannonlake Client
 /// Cannonlake client microarchitecture based processors.
-PROC(Cannonlake, "cannonlake", PROC_64_BIT)
+PROC_WITH_FEAT(Cannonlake, "cannonlake", PROC_64_BIT, FEATURE_AVX512VBMI)
 
 /// \name Icelake Client
 /// Icelake client microarchitecture based processors.
 PROC(Icelake, "icelake", PROC_64_BIT)
 
 /// \name Knights Landing
 /// Knights Landing processor.
-PROC(KNL, "knl", PROC_64_BIT)
+PROC_WITH_FEAT(KNL, "knl", PROC_64_BIT, FEATURE_AVX512F)
 
 /// \name Knights Mill
 /// Knights Mill processor.
-PROC(KNM, "knm", PROC_64_BIT)
+PROC_WITH_FEAT(KNM, "knm", PROC_64_BIT, FEATURE_AVX5124FMAPS)
 
 /// \name Lakemont
 /// Lakemont microarchitecture based processors.
 PROC(Lakemont, "lakemont", PROC_32_BIT)
 
 /// \name K6
 /// K6 architecture processors.
 //@{

 PROC_ALIAS(K8, "athlon64")
 PROC_ALIAS(K8, "athlon-fx")
 PROC_ALIAS(K8, "opteron")
 
 PROC(K8SSE3, "k8-sse3", PROC_64_BIT)
 PROC_ALIAS(K8SSE3, "athlon64-sse3")
 PROC_ALIAS(K8SSE3, "opteron-sse3")
 
-PROC(AMDFAM10, "amdfam10", PROC_64_BIT)
+PROC_WITH_FEAT(AMDFAM10, "amdfam10", PROC_64_BIT, FEATURE_SSE4_A)
 PROC_ALIAS(AMDFAM10, "barcelona")
 //@}
 
 /// \name Bobcat
 /// Bobcat architecture processors.
 //@{
-PROC(BTVER1, "btver1", PROC_64_BIT)
-PROC(BTVER2, "btver2", PROC_64_BIT)
+PROC_WITH_FEAT(BTVER1, "btver1", PROC_64_BIT, FEATURE_SSE4_A)
+PROC_WITH_FEAT(BTVER2, "btver2", PROC_64_BIT, FEATURE_BMI)
 //@}
 
 /// \name Bulldozer
 /// Bulldozer architecture processors.
 //@{
-PROC(BDVER1, "bdver1", PROC_64_BIT)
-PROC(BDVER2, "bdver2", PROC_64_BIT)
-PROC(BDVER3, "bdver3", PROC_64_BIT)
-PROC(BDVER4, "bdver4", PROC_64_BIT)
+PROC_WITH_FEAT(BDVER1, "bdver1", PROC_64_BIT, FEATURE_XOP)
+PROC_WITH_FEAT(BDVER2, "bdver2", PROC_64_BIT, FEATURE_FMA)
+PROC_WITH_FEAT(BDVER3, "bdver3", PROC_64_BIT, FEATURE_FMA)
+PROC_WITH_FEAT(BDVER4, "bdver4", PROC_64_BIT, FEATURE_AVX2)
 //@}
 
 /// \name zen
 /// Zen architecture processors.
 //@{
-PROC(ZNVER1, "znver1", PROC_64_BIT)
+PROC_WITH_FEAT(ZNVER1, "znver1", PROC_64_BIT, FEATURE_AVX2)
 //@}
 
 /// This specification is deprecated and will be removed in the future.
 /// Users should prefer K8.
 // FIXME: Warn on this when the CPU is set to it.
 //@{
 PROC(x86_64, "x86-64", PROC_64_BIT)
 //@}
 
 /// \name Geode
 /// Geode processors.
 //@{
 PROC(Geode, "geode", PROC_32_BIT)
 //@}
 
+// List of CPU Supports features in order.  These need to remain in the order
+// required by attribute 'target' checking.  Note that not all are supported/
+// prioritized by GCC, so synchronization with GCC's implementation may require
+// changing some existing values.
+FEATURE(FEATURE_CMOV)
+FEATURE(FEATURE_MMX)
+FEATURE(FEATURE_SSE)
+FEATURE(FEATURE_SSE2)
+FEATURE(FEATURE_SSE3)
+FEATURE(FEATURE_SSSE3)
+FEATURE(FEATURE_SSE4_A)
+FEATURE(FEATURE_SSE4_1)
+FEATURE(FEATURE_SSE4_2)
+FEATURE(FEATURE_POPCNT)
+FEATURE(FEATURE_AES)
+FEATURE(FEATURE_PCLMUL)
+FEATURE(FEATURE_AVX)
+FEATURE(FEATURE_BMI)
+FEATURE(FEATURE_FMA4)
+FEATURE(FEATURE_XOP)
+FEATURE(FEATURE_FMA)
+FEATURE(FEATURE_BMI2)
+FEATURE(FEATURE_AVX2)
+FEATURE(FEATURE_AVX512F)
+FEATURE(FEATURE_AVX512VL)
+FEATURE(FEATURE_AVX512BW)
+FEATURE(FEATURE_AVX512DQ)
+FEATURE(FEATURE_AVX512CD)
+FEATURE(FEATURE_AVX512ER)
+FEATURE(FEATURE_AVX512PF)
+FEATURE(FEATURE_AVX512VBMI)
+FEATURE(FEATURE_AVX512IFMA)
+FEATURE(FEATURE_AVX5124VNNIW)
+FEATURE(FEATURE_AVX5124FMAPS)
+FEATURE(FEATURE_AVX512VPOPCNTDQ)
 
 #undef PROC_64_BIT
 #undef PROC_32_BIT
+#undef FEATURE
 #undef PROC
 #undef PROC_ALIAS
+#undef PROC_WITH_FEAT

include/clang/Sema/Overload.h

enum OverloadFailureKind {
     ovl_fail_addr_not_available,
 
     /// This candidate was not viable because its OpenCL extension is disabled.
     ovl_fail_ext_disabled,
 
     /// This inherited constructor is not viable because it would slice the
     /// argument.
     ovl_fail_inhctor_slice,
+
+    /// This candidate was not viable because it is a non-default multiversioned
+    /// function.
+    ovl_non_default_multiversion_function,
   };
 
   /// A list of implicit conversion sequences for the arguments of an
   /// OverloadCandidate.
   typedef llvm::MutableArrayRef<ImplicitConversionSequence>
       ConversionSequenceList;
 
   /// OverloadCandidate - A single candidate in an overload set (C++ 13.3).

lib/Basic/Targets/X86.h

 #define PROC(ENUM, STRING, IS64BIT) CK_##ENUM,
 #include "clang/Basic/X86Target.def"
   } CPU = CK_Generic;
 
   bool checkCPUKind(CPUKind Kind) const;
 
   CPUKind getCPUKind(StringRef CPU) const;
 
+  std::string getCPUKindCanonicalName(CPUKind Kind) const;
+
   enum FPMathKind { FP_Default, FP_SSE, FP_387 } FPMath = FP_Default;
 
 public:
   X86TargetInfo(const llvm::Triple &Triple, const TargetOptions &)
       : TargetInfo(Triple) {
     LongDoubleFormat = &llvm::APFloat::x87DoubleExtended();
   }
   
public:
   bool isValidCPUName(StringRef Name) const override {
     return checkCPUKind(getCPUKind(Name));
   }
 
   bool setCPU(const std::string &Name) override {
     return checkCPUKind(CPU = getCPUKind(Name));
   }
 
+  bool supportsMultiVersioning() const override {
+    return getTriple().isOSBinFormatELF();
+  }
+  unsigned multiVersionSortPriority(StringRef Name) const override;
+
   bool setFPMath(StringRef Name) override;
 
   CallingConvCheckResult checkCallingConvention(CallingConv CC) const override {
     // Most of the non-ARM calling conventions are i386 conventions.
     switch (CC) {
     case CC_X86ThisCall:
     case CC_X86FastCall:
     case CC_X86StdCall:

lib/Basic/Targets/X86.cpp

 //===----------------------------------------------------------------------===//
 
 #include "X86.h"
 #include "clang/Basic/Builtins.h"
 #include "clang/Basic/Diagnostic.h"
 #include "clang/Basic/TargetBuiltins.h"
 #include "llvm/ADT/StringRef.h"
 #include "llvm/ADT/StringSwitch.h"
+#include "llvm/Support/TargetParser.h"
 
 namespace clang {
 namespace targets {
 
 const Builtin::Info BuiltinInfoX86[] = {
 #define BUILTIN(ID, TYPE, ATTRS)                                               \
   {#ID, TYPE, ATTRS, nullptr, ALL_LANGUAGES, nullptr},
 #define TARGET_BUILTIN(ID, TYPE, ATTRS, FEATURE)                               \

 // X86TargetInfo::hasFeature for a somewhat comprehensive list).
 bool X86TargetInfo::validateCpuSupports(StringRef FeatureStr) const {
   return llvm::StringSwitch<bool>(FeatureStr)
 #define X86_FEATURE_COMPAT(VAL, ENUM, STR) .Case(STR, true)
 #include "llvm/Support/X86TargetParser.def"
       .Default(false);
 }
 
+static llvm::X86::ProcessorFeatures getFeature(StringRef Name) {
+  return llvm::StringSwitch<llvm::X86::ProcessorFeatures>(Name)
+#define X86_FEATURE_COMPAT(VAL, ENUM, STR) .Case(STR, llvm::X86::ENUM)
+#include "llvm/Support/X86TargetParser.def"
+      ;
+  // Note, this function should only be used after ensuring the value is
+  // correct, so it asserts if the value is out of range.
+}
+
+static unsigned getFeaturePriority(llvm::X86::ProcessorFeatures Feat) {
+  enum class FeatPriority {
+#define FEATURE(FEAT) FEAT,
+#include "clang/Basic/X86Target.def"
+  };
+  switch (Feat) {
+#define FEATURE(FEAT)                                                          \
+  case llvm::X86::FEAT:                                                        \
+    return static_cast<unsigned>(FeatPriority::FEAT);
+#include "clang/Basic/X86Target.def"
+  default:
+    llvm_unreachable("No Feature Priority for non-CPUSupports Features");
+  }
+}
+
+unsigned X86TargetInfo::multiVersionSortPriority(StringRef Name) const {
+  // Valid CPUs have a 'key feature' that compares just better than its key
+  // feature.
+  CPUKind Kind = getCPUKind(Name);
+  if (Kind != CK_Generic) {
+    switch (Kind) {
+    default:
+      llvm_unreachable(
+          "CPU Type without a key feature used in 'target' attribute");
+#define PROC_WITH_FEAT(ENUM, STR, IS64, KEY_FEAT)                              \
+  case CK_##ENUM:                                                              \
+    return (getFeaturePriority(llvm::X86::KEY_FEAT) << 1) + 1;
+#include "clang/Basic/X86Target.def"
+    }
+  }
+
+  // Now we know we have a feature, so get its priority and shift it a few so
+  // that we have sufficient room for the CPUs (above).
+  return getFeaturePriority(getFeature(Name)) << 1;
+}
+
+std::string X86TargetInfo::getCPUKindCanonicalName(CPUKind Kind) const {
+  switch (Kind) {
+  case CK_Generic:
+    return "";
+#define PROC(ENUM, STRING, IS64BIT)                                            \
+  case CK_##ENUM:                                                              \
+    return STRING;
+#include "clang/Basic/X86Target.def"
+  }
+  llvm_unreachable("Invalid CPUKind");
+}
+
 // We can't use a generic validation scheme for the cpus accepted here
 // versus subtarget cpus accepted in the target attribute because the
 // variables intitialized by the runtime only support the below currently
 // rather than the full range of cpus.
 bool X86TargetInfo::validateCpuIs(StringRef FeatureStr) const {
   return llvm::StringSwitch<bool>(FeatureStr)
 #define X86_VENDOR(ENUM, STRING) .Case(STRING, true)
 #define X86_CPU_TYPE_COMPAT_WITH_ALIAS(ARCHNAME, ENUM, STR, ALIAS)             \

lib/CodeGen/CodeGenFunction.h

#endif
 
   /// If \p E references a parameter with pass_object_size info or a constant
   /// array size modifier, emit the object size divided by the size of \p EltTy.
   /// Otherwise return null.
   llvm::Value *LoadPassedObjectSize(const Expr *E, QualType EltTy);
 
   void EmitSanitizerStatReport(llvm::SanitizerStatKind SSK);
 
+  struct MultiVersionResolverOption {
+    llvm::Function *Function;
+    TargetAttr::ParsedTargetAttr ParsedAttribute;
+    unsigned Priority;
+    MultiVersionResolverOption(const TargetInfo &TargInfo, llvm::Function *F,
+                               const clang::TargetAttr::ParsedTargetAttr &PT)
+        : Function(F), ParsedAttribute(PT), Priority(0u) {
+      for (StringRef Feat : PT.Features)
+        Priority = std::max(Priority,
+                            TargInfo.multiVersionSortPriority(Feat.substr(1)));
+
+      if (!PT.Architecture.empty())
+        Priority = std::max(Priority,
+                            TargInfo.multiVersionSortPriority(PT.Architecture));
+    }
+
+    bool operator>(const MultiVersionResolverOption &Other) const {
+      return Priority > Other.Priority;
+    }
+  };
+  void EmitMultiVersionResolver(llvm::Function *Resolver,
+                                ArrayRef<MultiVersionResolverOption> Options);
+
 private:
   QualType getVarArgType(const Expr *Arg);
 
   void EmitDeclMetadata();
 
   BlockByrefHelpers *buildByrefHelpers(llvm::StructType &byrefType,
                                   const AutoVarEmission &emission);
 
   void AddObjCARCExceptionMetadata(llvm::Instruction *Inst);
 
   llvm::Value *GetValueForARMHint(unsigned BuiltinID);
   llvm::Value *EmitX86CpuIs(const CallExpr *E);
   llvm::Value *EmitX86CpuIs(StringRef CPUStr);
   llvm::Value *EmitX86CpuSupports(const CallExpr *E);
   llvm::Value *EmitX86CpuSupports(ArrayRef<StringRef> FeatureStrs);
   llvm::Value *EmitX86CpuInit();
+  llvm::Value *FormResolverCondition(const MultiVersionResolverOption &RO);
 };
 
 /// Helper class with most of the code for saving a value for a
 /// conditional expression cleanup.
 struct DominatingLLVMValue {
   typedef llvm::PointerIntPair<llvm::Value*, 1, bool> saved_type;
 
   /// Answer whether the given value needs extra work to be saved.

lib/CodeGen/CodeGenFunction.cpp

void CodeGenFunction::EmitSanitizerStatReport(llvm::SanitizerStatKind SSK) {
   if (!CGM.getCodeGenOpts().SanitizeStats)
     return;
 
   llvm::IRBuilder<> IRB(Builder.GetInsertBlock(), Builder.GetInsertPoint());
   IRB.SetCurrentDebugLocation(Builder.getCurrentDebugLocation());
   CGM.getSanStats().create(IRB, SSK);
 }
 
+llvm::Value *
+CodeGenFunction::FormResolverCondition(const MultiVersionResolverOption &RO) {
+  llvm::Value *TrueCondition = nullptr;
+  if (!RO.ParsedAttribute.Architecture.empty())
+    TrueCondition = EmitX86CpuIs(RO.ParsedAttribute.Architecture);
+
+  if (!RO.ParsedAttribute.Features.empty()) {
+    SmallVector<StringRef, 8> FeatureList;
+    llvm::for_each(RO.ParsedAttribute.Features,
+                   [&FeatureList](const std::string &Feature) {
+                     FeatureList.push_back(StringRef{Feature}.substr(1));
+                   });
+    llvm::Value *FeatureCmp = EmitX86CpuSupports(FeatureList);
+    TrueCondition = TrueCondition ? Builder.CreateAnd(TrueCondition, FeatureCmp)
+                                  : FeatureCmp;
+  }
+  return TrueCondition;
+}
+
+void CodeGenFunction::EmitMultiVersionResolver(
+    llvm::Function *Resolver, ArrayRef<MultiVersionResolverOption> Options) {
+  assert((getContext().getTargetInfo().getTriple().getArch() ==
+              llvm::Triple::x86 ||
+          getContext().getTargetInfo().getTriple().getArch() ==
+              llvm::Triple::x86_64) &&
+         "Only implemented for x86 targets");
+
+  // Main function's basic block.
+  llvm::BasicBlock *CurBlock = createBasicBlock("entry", Resolver);
+  Builder.SetInsertPoint(CurBlock);
+  EmitX86CpuInit();
+
+  llvm::Function *DefaultFunc = nullptr;
+  for (const MultiVersionResolverOption &RO : Options) {
+    Builder.SetInsertPoint(CurBlock);
+    llvm::Value *TrueCondition = FormResolverCondition(RO);
+
+    if (!TrueCondition) {
+      DefaultFunc = RO.Function;
+    } else {
+      llvm::BasicBlock *RetBlock = createBasicBlock("ro_ret", Resolver);
+      llvm::IRBuilder<> RetBuilder(RetBlock);
+      RetBuilder.CreateRet(RO.Function);
+      CurBlock = createBasicBlock("ro_else", Resolver);
+      Builder.CreateCondBr(TrueCondition, RetBlock, CurBlock);
+    }
+  }
+
+  assert(DefaultFunc && "No default version?");
+  // Emit return from the 'else-ist' block.
+  Builder.SetInsertPoint(CurBlock);
+  Builder.CreateRet(DefaultFunc);
+}
+
 llvm::DebugLoc CodeGenFunction::SourceLocToDebugLoc(SourceLocation Location) {
   if (CGDebugInfo *DI = getDebugInfo())
     return DI->SourceLocToDebugLoc(Location);
 
   return llvm::DebugLoc();
 }

lib/CodeGen/CodeGenModule.h

private:
   void addDeferredDeclToEmit(GlobalDecl GD) {
     DeferredDeclsToEmit.emplace_back(GD);
   }
 
   /// List of alias we have emitted. Used to make sure that what they point to
   /// is defined once we get to the end of the of the translation unit.
   std::vector<GlobalDecl> Aliases;
 
+  /// List of multiversion functions that have to be emitted.  Used to make sure
+  /// we properly emit the iFunc.
+  std::vector<GlobalDecl> MultiVersionFuncs;
+
   typedef llvm::StringMap<llvm::TrackingVH<llvm::Constant> > ReplacementsTy;
   ReplacementsTy Replacements;
 
   /// List of global values to be replaced with something else. Used when we
   /// want to replace a GlobalValue but can't identify it by its mangled name
   /// anymore (because the name is already taken).
   llvm::SmallVector<std::pair<llvm::GlobalValue *, llvm::Constant *>, 8>
     GlobalValReplacements;

 
 private:
   llvm::Constant *GetOrCreateLLVMFunction(
       StringRef MangledName, llvm::Type *Ty, GlobalDecl D, bool ForVTable,
       bool DontDefer = false, bool IsThunk = false,
       llvm::AttributeList ExtraAttrs = llvm::AttributeList(),
       ForDefinition_t IsForDefinition = NotForDefinition);
 
+  llvm::Constant *GetOrCreateMultiVersionIFunc(GlobalDecl GD,
+                                               llvm::Type *DeclTy,
+                                               StringRef MangledName,
+                                               const FunctionDecl *FD);
+  void UpdateMultiVersionNames(GlobalDecl GD, const FunctionDecl *FD);
+
   llvm::Constant *GetOrCreateLLVMGlobal(StringRef MangledName,
                                         llvm::PointerType *PTy,
                                         const VarDecl *D,
                                         ForDefinition_t IsForDefinition
                                           = NotForDefinition);
 
   void setNonAliasAttributes(const Decl *D, llvm::GlobalObject *GO);
 
private:
   /// Call replaceAllUsesWith on all pairs in Replacements.
   void applyReplacements();
 
   /// Call replaceAllUsesWith on all pairs in GlobalValReplacements.
   void applyGlobalValReplacements();
 
   void checkAliases();
 
+  void emitMultiVersionFunctions();
+
   /// Emit any vtables which we deferred and still have a use for.
   void EmitDeferredVTables();
 
   /// Emit a dummy function that reference a CoreFoundation symbol when
   /// @available is used on Darwin.
   void emitAtAvailableLinkGuard();
 
   /// Emit the llvm.used and llvm.compiler.used metadata.

lib/CodeGen/CodeGenModule.cpp

 }
 
 void CodeGenModule::Release() {
   EmitDeferred();
   EmitVTablesOpportunistically();
   applyGlobalValReplacements();
   applyReplacements();
   checkAliases();
+  emitMultiVersionFunctions();
   EmitCXXGlobalInitFunc();
   EmitCXXGlobalDtorFunc();
   EmitCXXThreadLocalInitFunc();
   if (ObjCRuntime)
     if (llvm::Function *ObjCInitFunction = ObjCRuntime->ModuleInitFunction())
       AddGlobalCtor(ObjCInitFunction);
   if (Context.getLangOpts().CUDA && !Context.getLangOpts().CUDAIsDevice &&
       CUDARuntime) {
void CodeGenModule::setTLSMode(llvm::GlobalValue *GV, const VarDecl &D) const {
   // Override the TLS model if it is explicitly specified.
   if (const TLSModelAttr *Attr = D.getAttr<TLSModelAttr>()) {
     TLM = GetLLVMTLSModel(Attr->getModel());
   }
 
   GV->setThreadLocalMode(TLM);
 }
 
-StringRef CodeGenModule::getMangledName(GlobalDecl GD) {
-  GlobalDecl CanonicalGD = GD.getCanonicalDecl();
+static void AppendTargetMangling(const CodeGenModule &CGM,
+                                 const TargetAttr *Attr, raw_ostream &Out) {
+  if (Attr->isDefaultVersion())
+    return;
 
-  // Some ABIs don't have constructor variants.  Make sure that base and
-  // complete constructors get mangled the same.
-  if (const auto *CD = dyn_cast<CXXConstructorDecl>(CanonicalGD.getDecl())) {
-    if (!getTarget().getCXXABI().hasConstructorVariants()) {
-      CXXCtorType OrigCtorType = GD.getCtorType();
-      assert(OrigCtorType == Ctor_Base || OrigCtorType == Ctor_Complete);
-      if (OrigCtorType == Ctor_Base)
-        CanonicalGD = GlobalDecl(CD, Ctor_Complete);
-    }
+  Out << '.';
+  const auto &Target = CGM.getTarget();
+  TargetAttr::ParsedTargetAttr Info =
+      Attr->parse([&Target](StringRef LHS, StringRef RHS) {
+                    return Target.multiVersionSortPriority(LHS) >
+                           Target.multiVersionSortPriority(RHS);
+                  });
+
+  bool IsFirst = true;
+
+  if (!Info.Architecture.empty()) {
+    IsFirst = false;
+    Out << "arch_" << Info.Architecture;
   }
 
-  auto FoundName = MangledDeclNames.find(CanonicalGD);
-  if (FoundName != MangledDeclNames.end())
-    return FoundName->second;
+  for (StringRef Feat : Info.Features) {
+    if (!IsFirst)
+      Out << '_';
+    IsFirst = false;
+    Out << Feat.substr(1);
+  }
+}
 
-  const auto *ND = cast<NamedDecl>(GD.getDecl());
+static std::string getMangledNameImpl(const CodeGenModule &CGM, GlobalDecl GD,
+                                      const NamedDecl *ND,
+                                      bool OmitTargetMangling = false) {
   SmallString<256> Buffer;
-  StringRef Str;
-  if (getCXXABI().getMangleContext().shouldMangleDeclName(ND)) {
+  llvm::raw_svector_ostream Out(Buffer);
+  MangleContext &MC = CGM.getCXXABI().getMangleContext();
+  if (MC.shouldMangleDeclName(ND)) {
     llvm::raw_svector_ostream Out(Buffer);
     if (const auto *D = dyn_cast<CXXConstructorDecl>(ND))
-      getCXXABI().getMangleContext().mangleCXXCtor(D, GD.getCtorType(), Out);
+      MC.mangleCXXCtor(D, GD.getCtorType(), Out);
     else if (const auto *D = dyn_cast<CXXDestructorDecl>(ND))
-      getCXXABI().getMangleContext().mangleCXXDtor(D, GD.getDtorType(), Out);
+      MC.mangleCXXDtor(D, GD.getDtorType(), Out);
     else
-      getCXXABI().getMangleContext().mangleName(ND, Out);
-    Str = Out.str();
+      MC.mangleName(ND, Out);
   } else {
     IdentifierInfo *II = ND->getIdentifier();
     assert(II && "Attempt to mangle unnamed decl.");
     const auto *FD = dyn_cast<FunctionDecl>(ND);
 
     if (FD &&
         FD->getType()->castAs<FunctionType>()->getCallConv() == CC_X86RegCall) {
       llvm::raw_svector_ostream Out(Buffer);
       Out << "__regcall3__" << II->getName();
-      Str = Out.str();
     } else {
-      Str = II->getName();
-    }
-  }
+      Out << II->getName();
+    }
+  }
+
+  if (const auto *FD = dyn_cast<FunctionDecl>(ND))
+    if (FD->isMultiVersion() && !OmitTargetMangling)
+      AppendTargetMangling(CGM, FD->getAttr<TargetAttr>(), Out);
+  return Out.str();
+}
+
+void CodeGenModule::UpdateMultiVersionNames(GlobalDecl GD,
+                                            const FunctionDecl *FD) {
+  if (!FD->isMultiVersion())
+    return;
+
+  // Get the name of what this would be without the 'target' attribute.  This
+  // allows us to lookup the version that was emitted when this wasn't a
+  // multiversion function.
+  std::string NonTargetName =
+      getMangledNameImpl(*this, GD, FD, /*OmitTargetMangling=*/true);
+  GlobalDecl OtherGD;
+  if (lookupRepresentativeDecl(NonTargetName, OtherGD)) {
+    assert(OtherGD.getCanonicalDecl()
+               .getDecl()
+               ->getAsFunction()
+               ->isMultiVersion() &&
+           "Other GD should now be a multiversioned function");
+    // OtherFD is the version of this function that was mangled BEFORE
+    // becoming a MultiVersion function.  It potentially needs to be updated.
+    const FunctionDecl *OtherFD =
+        OtherGD.getCanonicalDecl().getDecl()->getAsFunction();
+    std::string OtherName = getMangledNameImpl(*this, OtherGD, OtherFD);
+    // This is so that if the initial version was already the 'default'
+    // version, we don't try to update it.
+    if (OtherName != NonTargetName) {
+      Manglings.erase(NonTargetName);
+      auto Result = Manglings.insert(std::make_pair(OtherName, OtherGD));
+      MangledDeclNames[OtherGD.getCanonicalDecl()] = Result.first->first();
+      if (llvm::GlobalValue *Entry = GetGlobalValue(NonTargetName))
+        Entry->setName(OtherName);
+    }
 
+StringRef CodeGenModule::getMangledName(GlobalDecl GD) {
+  GlobalDecl CanonicalGD = GD.getCanonicalDecl();
+
+  // Some ABIs don't have constructor variants.  Make sure that base and
+  // complete constructors get mangled the same.
+  if (const auto *CD = dyn_cast<CXXConstructorDecl>(CanonicalGD.getDecl())) {
+    if (!getTarget().getCXXABI().hasConstructorVariants()) {
+      CXXCtorType OrigCtorType = GD.getCtorType();
+      assert(OrigCtorType == Ctor_Base || OrigCtorType == Ctor_Complete);
+      if (OrigCtorType == Ctor_Base)
+        CanonicalGD = GlobalDecl(CD, Ctor_Complete);
+    }
+  }
+
+  auto FoundName = MangledDeclNames.find(CanonicalGD);
+  if (FoundName != MangledDeclNames.end())
+    return FoundName->second;
+
+
   // Keep the first result in the case of a mangling collision.
-  auto Result = Manglings.insert(std::make_pair(Str, GD));
+  const auto *ND = cast<NamedDecl>(GD.getDecl());
+  auto Result =
+      Manglings.insert(std::make_pair(getMangledNameImpl(*this, GD, ND), GD));
   return MangledDeclNames[CanonicalGD] = Result.first->first();
 }
 
 StringRef CodeGenModule::getBlockMangledName(GlobalDecl GD,
                                              const BlockDecl *BD) {
   MangleContext &MangleCtx = getCXXABI().getMangleContext();
   const Decl *D = GD.getDecl();
 

 
 void CodeGenModule::EmitCtorList(CtorList &Fns, const char *GlobalName) {
   if (Fns.empty()) return;
 
   // Ctor function type is void()*.
   llvm::FunctionType* CtorFTy = llvm::FunctionType::get(VoidTy, false);
   llvm::Type *CtorPFTy = llvm::PointerType::getUnqual(CtorFTy);
 
   // Get the type of a ctor entry, { i32, void ()*, i8* }.
   llvm::StructType *CtorStructTy = llvm::StructType::get(
       Int32Ty, llvm::PointerType::getUnqual(CtorFTy), VoidPtrTy);
 
   // Construct the constructor and destructor arrays.
   ConstantInitBuilder builder(*this);
   auto ctors = builder.beginArray(CtorStructTy);
   for (const auto &I : Fns) {
     auto ctor = ctors.beginStruct(CtorStructTy);
     ctor.addInt(Int32Ty, I.Priority);
CodeGenModule::isTriviallyRecursive(const FunctionDecl *FD) {
   Walker.TraverseFunctionDecl(const_cast<FunctionDecl*>(FD));
   return Walker.Result;
 }
 
 bool CodeGenModule::shouldEmitFunction(GlobalDecl GD) {
   if (getFunctionLinkage(GD) != llvm::Function::AvailableExternallyLinkage)
     return true;
   const auto *F = cast<FunctionDecl>(GD.getDecl());
   if (CodeGenOpts.OptimizationLevel == 0 && !F->hasAttr<AlwaysInlineAttr>())
     return false;
 
   if (F->hasAttr<DLLImportAttr>()) {
     // Check whether it would be safe to inline this dllimport function.
     DLLImportFunctionVisitor Visitor;
     Visitor.TraverseFunctionDecl(const_cast<FunctionDecl*>(F));
     if (!Visitor.SafeToInline)
       return false;
 
if (const auto *VD = dyn_cast<VarDecl>(D))
     return EmitGlobalVarDefinition(VD, !VD->hasDefinition());
 
   llvm_unreachable("Invalid argument to EmitGlobalDefinition()");
 }
 
 static void ReplaceUsesOfNonProtoTypeWithRealFunction(llvm::GlobalValue *Old,
                                                       llvm::Function *NewFn);
 
+void CodeGenModule::emitMultiVersionFunctions() {
+  for (GlobalDecl GD : MultiVersionFuncs) {
+    SmallVector<CodeGenFunction::MultiVersionResolverOption, 10> Options;
+    const FunctionDecl *FD = cast<FunctionDecl>(GD.getDecl());
+    getContext().forEachMultiversionedFunctionVersion(
+        FD, [this, &GD, &Options](FunctionDecl *CurFD) {
+          GlobalDecl CurGD{CurFD};
+          StringRef MangledName = getMangledName(CurGD);
+          llvm::Constant *Func = GetGlobalValue(MangledName);
+          if (!Func) {
+            if (CurFD->isDefined()) {
+              EmitGlobalFunctionDefinition(CurGD, nullptr);
+              Func = GetGlobalValue(MangledName);
+            } else {
+              const CGFunctionInfo &FI =
+                  getTypes().arrangeGlobalDeclaration(GD);
+              llvm::FunctionType *Ty = getTypes().GetFunctionType(FI);
+              Func = GetAddrOfFunction(CurGD, Ty, /*ForVTable=*/false,
+                                       /*DontDefer=*/false, ForDefinition);
+            }
+            assert(Func && "This should have just been created");
+          }
+          Options.emplace_back(getTarget(), cast<llvm::Function>(Func),
+                               CurFD->getAttr<TargetAttr>()->parse());
+        });
+
+
+    llvm::Function *ResolverFunc = cast<llvm::Function>(
+        GetGlobalValue((getMangledName(GD) + ".resolver").str()));
+    std::stable_sort(
+        Options.begin(), Options.end(),
+        std::greater<CodeGenFunction::MultiVersionResolverOption>());
+    CodeGenFunction CGF(*this);
+    CGF.EmitMultiVersionResolver(ResolverFunc, Options);
+  }
+}
+
+/// If an ifunc for the specified mangled name is not in the module, create and
+/// return an llvm IFunc Function with the specified type.
+llvm::Constant *
+CodeGenModule::GetOrCreateMultiVersionIFunc(GlobalDecl GD, llvm::Type *DeclTy,
+                                            StringRef MangledName,
+                                            const FunctionDecl *FD) {
+  std::string IFuncName = (MangledName + ".ifunc").str();
+  if (llvm::GlobalValue *IFuncGV = GetGlobalValue(IFuncName))
+    return IFuncGV;
+
+  // Since this is the first time we've created this IFunc, make sure
+  // that we put this multiversioned function into the list to be
+  // replaced later.
+  MultiVersionFuncs.push_back(GD);
+
+  std::string ResolverName = (MangledName + ".resolver").str();
+  llvm::Type *ResolverType = llvm::FunctionType::get(
+      llvm::PointerType::get(DeclTy,
+                             Context.getTargetAddressSpace(FD->getType())),
+      false);
+  llvm::Constant *Resolver =
+      GetOrCreateLLVMFunction(ResolverName, ResolverType, GlobalDecl{},
+                              /*ForVTable=*/false);
+  llvm::GlobalIFunc *GIF = llvm::GlobalIFunc::create(
+      DeclTy, 0, llvm::Function::ExternalLinkage, "", Resolver, &getModule());
+  GIF->setName(IFuncName);
+  SetCommonAttributes(FD, GIF);
+
+  return GIF;
+}
+
 /// GetOrCreateLLVMFunction - If the specified mangled name is not in the
 /// module, create and return an llvm Function with the specified type. If there
 /// is something in the module with the specified name, return it potentially
 /// bitcasted to the right type.
 ///
 /// If D is non-null, it specifies a decl that correspond to this.  This is used
 /// to set the attributes on the function when it is first created.
 llvm::Constant *CodeGenModule::GetOrCreateLLVMFunction(
     StringRef MangledName, llvm::Type *Ty, GlobalDecl GD, bool ForVTable,
     bool DontDefer, bool IsThunk, llvm::AttributeList ExtraAttrs,
     ForDefinition_t IsForDefinition) {
   const Decl *D = GD.getDecl();
 
+  // Any attempts to use a MultiVersion function should result in retrieving
+  // the iFunc instead. Name Mangling will handle the rest of the changes.
+  if (const FunctionDecl *FD = cast_or_null<FunctionDecl>(D)) {
+    if (FD->isMultiVersion() && FD->getAttr<TargetAttr>()->isDefaultVersion()) {
+      UpdateMultiVersionNames(GD, FD);
+      if (!IsForDefinition)
+        return GetOrCreateMultiVersionIFunc(GD, Ty, MangledName, FD);
+    }
+  }
+
   // Lookup the entry, lazily creating it if necessary.
   llvm::GlobalValue *Entry = GetGlobalValue(MangledName);
   if (Entry) {
     if (WeakRefReferences.erase(Entry)) {
       const FunctionDecl *FD = cast_or_null<FunctionDecl>(D);
       if (FD && !FD->hasAttr<WeakAttr>())
         Entry->setLinkage(llvm::Function::ExternalLinkage);
     }

lib/Sema/Sema.cpp

for (UnresolvedSetImpl::iterator It = Overloads.begin(),
     // FIXME: Magic number for max shown overloads stolen from
     // OverloadCandidateSet::NoteCandidates.
     if (ShownOverloads >= 4 && S.Diags.getShowOverloads() == Ovl_Best) {
       ++SuppressedOverloads;
       continue;
     }
 
     NamedDecl *Fn = (*It)->getUnderlyingDecl();
+    // Don't print overloads for non-default multiversioned functions.
+    if (const auto *FD = Fn->getAsFunction()) {
+      if (FD->isMultiVersion() &&
+          !FD->getAttr<TargetAttr>()->isDefaultVersion())
+        continue;
+    }
     S.Diag(Fn->getLocation(), diag::note_possible_target_of_call);
     ++ShownOverloads;
   }
 
   if (SuppressedOverloads)
     S.Diag(FinalNoteLoc, diag::note_ovl_too_many_candidates)
       << SuppressedOverloads;
 }

lib/Sema/SemaDecl.cpp

bool Sema::shouldLinkDependentDeclWithPrevious(Decl *D, Decl *PrevDecl) {
   //       template<typename T> class C2 { friend void func() { } };
   //
   // This code snippet is a valid code unless both templates are instantiated.
   return !(D->getLexicalDeclContext()->isDependentContext() &&
            D->getDeclContext()->isFileContext() &&
            D->getFriendObjectKind() != Decl::FOK_None);
 }
 
+/// \brief Check the target attribute of the function for MultiVersion
+/// validity.
+///
+/// Returns true if there was an error, false otherwise.
+static bool CheckMultiVersionValue(Sema &S, const FunctionDecl *FD) {
+  const auto *TA = FD->getAttr<TargetAttr>();
+  assert(TA && "MultiVersion Candidate requires a target attribute");
+  TargetAttr::ParsedTargetAttr ParseInfo = TA->parse();
+  const TargetInfo &TargetInfo = S.Context.getTargetInfo();
+  enum ErrType { Feature = 0, Architecture = 1 };
+
+  if (!ParseInfo.Architecture.empty() &&
+      !TargetInfo.validateCpuIs(ParseInfo.Architecture)) {
+    S.Diag(FD->getLocation(), diag::err_bad_multiversion_option)
+        << Architecture << ParseInfo.Architecture;
+    return true;
+  }
+
+  for (const auto &Feature : ParseInfo.Features) {
+    auto BareFeat = StringRef{Feature}.substr(1);
+    if (Feature[0] == '-') {
+      S.Diag(FD->getLocation(), diag::err_bad_multiversion_option)
+          << Feature << ("no-" + BareFeat).str();
+      return true;
+    }
+
+    if (!TargetInfo.validateCpuSupports(BareFeat) ||
+        !TargetInfo.isValidFeatureName(BareFeat)) {
+      S.Diag(FD->getLocation(), diag::err_bad_multiversion_option)
+          << Feature << BareFeat;
+      return true;
+    }
+  }
+  return false;
+}
+
+static bool CheckMultiVersionAdditionalRules(Sema &S, const FunctionDecl *OldFD,
+                                             const FunctionDecl *NewFD,
+                                             bool CausesMV) {
+  enum DoesntSupport {
+    FuncTemplates = 0,
+    VirtFuncs = 1,
+    DeducedReturn = 2,
+    Constructors = 3,
+    Destructors = 4,
+    DeletedFuncs = 5,
+    DefaultedFuncs = 6
+  };
+  enum Different {
+    CallingConv = 0,
+    ReturnType = 1,
+    ConstexprSpec = 2,
+    InlineSpec = 3,
+    StorageClass = 4,
+    Linkage = 5
+  };
+
+  // For now, disallow all other attributes.  These should be opt-in, but
+  // an analysis of all of them is a future FIXME.
+  if (CausesMV && OldFD &&
+      std::distance(OldFD->attr_begin(), OldFD->attr_end()) != 1) {
+    S.Diag(OldFD->getLocation(), diag::err_multiversion_no_other_attrs);
+    S.Diag(NewFD->getLocation(), diag::note_multiversioning_caused_here);
+    return true;
+  }
+
+  if (std::distance(NewFD->attr_begin(), NewFD->attr_end()) != 1) {
+    S.Diag(NewFD->getLocation(), diag::err_multiversion_no_other_attrs);
+    if (CausesMV)
+      S.Diag(NewFD->getLocation(), diag::note_multiversioning_caused_here);
+    return true;
+  }
+
+  if (NewFD->getTemplatedKind() == FunctionDecl::TK_FunctionTemplate) {
+    S.Diag(NewFD->getLocation(), diag::err_multiversion_doesnt_support)
+        << FuncTemplates;
+    if (CausesMV)
+      S.Diag(NewFD->getLocation(), diag::note_multiversioning_caused_here);
+    return true;
+  }
+
+
+  if (const auto *NewCXXFD = dyn_cast<CXXMethodDecl>(NewFD)) {
+    if (NewCXXFD->isVirtual()) {
+      S.Diag(NewCXXFD->getLocation(), diag::err_multiversion_doesnt_support)
+          << VirtFuncs;
+      if (CausesMV)
+        S.Diag(NewCXXFD->getLocation(), diag::note_multiversioning_caused_here);
+      return true;
+    }
+
+    if (const auto *NewCXXCtor = dyn_cast<CXXConstructorDecl>(NewFD)) {
+      S.Diag(NewCXXCtor->getLocation(), diag::err_multiversion_doesnt_support)
+          << Constructors;
+      if (CausesMV)
+        S.Diag(NewCXXCtor->getLocation(),
+               diag::note_multiversioning_caused_here);
+      return true;
+    }
+
+    if (const auto *NewCXXDtor = dyn_cast<CXXDestructorDecl>(NewFD)) {
+      S.Diag(NewCXXDtor->getLocation(), diag::err_multiversion_doesnt_support)
+          << Destructors;
+      if (CausesMV)
+        S.Diag(NewCXXDtor->getLocation(),
+               diag::note_multiversioning_caused_here);
+      return true;
+    }
+  }
+
+  if (NewFD->isDeleted()) {
+    S.Diag(NewFD->getLocation(), diag::err_multiversion_doesnt_support)
+      << DeletedFuncs;
+    if (CausesMV)
+      S.Diag(NewFD->getLocation(), diag::note_multiversioning_caused_here);
+  }
+  if (NewFD->isDefaulted()) {
+    S.Diag(NewFD->getLocation(), diag::err_multiversion_doesnt_support)
+      << DefaultedFuncs;
+    if (CausesMV)
+      S.Diag(NewFD->getLocation(), diag::note_multiversioning_caused_here);
+  }
+
+  QualType NewQType = S.getASTContext().getCanonicalType(NewFD->getType());
+  const auto *NewType = cast<FunctionType>(NewQType);
+  QualType NewReturnType = NewType->getReturnType();
+
+  if (NewReturnType->isUndeducedType()) {
+    S.Diag(NewFD->getLocation(), diag::err_multiversion_doesnt_support)
+        << DeducedReturn;
+    if (CausesMV)
+      S.Diag(NewFD->getLocation(), diag::note_multiversioning_caused_here);
+    return true;
+  }
+
+  // Only allow transition to MultiVersion if it hasn't been used.
+  if (OldFD && CausesMV && OldFD->isUsed(false)) {
+    S.Diag(NewFD->getLocation(), diag::err_multiversion_after_used);
+    return true;
+  }
+
+  // Ensure the return type is identical.
+  if (OldFD) {
+    QualType OldQType = S.getASTContext().getCanonicalType(OldFD->getType());
+    const auto *OldType = cast<FunctionType>(OldQType);
+    FunctionType::ExtInfo OldTypeInfo = OldType->getExtInfo();
+    FunctionType::ExtInfo NewTypeInfo = NewType->getExtInfo();
+
+    if (OldTypeInfo.getCC() != NewTypeInfo.getCC()) {
+      S.Diag(NewFD->getLocation(), diag::err_multiversion_diff) << CallingConv;
+      S.Diag(NewFD->getLocation(), diag::note_multiversioning_caused_here);
+      return true;
+    }
+
+    QualType OldReturnType = OldType->getReturnType();
+
+    if (OldReturnType != NewReturnType) {
+      S.Diag(NewFD->getLocation(), diag::err_multiversion_diff) << ReturnType;
+      if (CausesMV)
+        S.Diag(NewFD->getLocation(), diag::note_multiversioning_caused_here);
+      return true;
+    }
+
+    if (OldFD->isConstexpr() != NewFD->isConstexpr()) {
+      S.Diag(NewFD->getLocation(), diag::err_multiversion_diff)
+          << ConstexprSpec;
+      if (CausesMV)
+        S.Diag(NewFD->getLocation(), diag::note_multiversioning_caused_here);
+      return true;
+    }
+
+    if (OldFD->isInlineSpecified() != NewFD->isInlineSpecified()) {
+      S.Diag(NewFD->getLocation(), diag::err_multiversion_diff) << InlineSpec;
+      if (CausesMV)
+        S.Diag(NewFD->getLocation(), diag::note_multiversioning_caused_here);
+      return true;
+    }
+
+    if (OldFD->getStorageClass() != NewFD->getStorageClass()) {
+      S.Diag(NewFD->getLocation(), diag::err_multiversion_diff) << StorageClass;
+      if (CausesMV)
+        S.Diag(NewFD->getLocation(), diag::note_multiversioning_caused_here);
+      return true;
+    }
+
+    if (OldFD->isExternC() != NewFD->isExternC()) {
+      S.Diag(NewFD->getLocation(), diag::err_multiversion_diff) << Linkage;
+      if (CausesMV)
+        S.Diag(NewFD->getLocation(), diag::note_multiversioning_caused_here);
+      return true;
+    }
+
+    if (S.CheckEquivalentExceptionSpec(
+            OldFD->getType()->getAs<FunctionProtoType>(), OldFD->getLocation(),
+            NewFD->getType()->getAs<FunctionProtoType>(), NewFD->getLocation()))
+      return true;
+  }
+  return false;
+}
+
+/// \brief Check the validity of a mulitversion function declaration.
+/// Also sets the multiversion'ness' of the function itself.
+///
+/// This sets NewFD->isInvalidDecl() to true if there was an error.
+///
+/// Returns true if there was an error, false otherwise.
+static bool CheckMultiVersionFunction(Sema &S, FunctionDecl *NewFD,
+                                      bool &Redeclaration, NamedDecl *&OldDecl,
+                                      bool &MergeTypeWithPrevious,
+                                      LookupResult &Previous) {
+  const auto *NewTA = NewFD->getAttr<TargetAttr>();
+  if (NewFD->isMain()) {
+    if (NewTA && NewTA->isDefaultVersion()) {
+      S.Diag(NewFD->getLocation(), diag::err_multiversion_not_allowed_on_main);
+      NewFD->isInvalidDecl();
+      return true;
+    }
+    return false;
+  }
+
+  // If there is no matching previous decl, only 'default' can
+  // cause MultiVersioning.
+  if (!OldDecl) {
+    if (NewTA && NewTA->isDefaultVersion()) {
+      if (!NewFD->getType()->getAs<FunctionProtoType>()) {
+        S.Diag(NewFD->getLocation(), diag::err_multiversion_noproto);
+        NewFD->setInvalidDecl();
+        return true;
+      }
+      if (CheckMultiVersionAdditionalRules(S, nullptr, NewFD, true)) {
+        NewFD->setInvalidDecl();
+        return true;
+      }
+      if (!S.getASTContext().getTargetInfo().supportsMultiVersioning()) {
+        S.Diag(NewFD->getLocation(), diag::err_multiversion_not_supported);
+        return true;
+      }
+
+      NewFD->setIsMultiVersion();
+    }
+    return false;
+  }
+
+  if (OldDecl->getDeclContext()->getRedeclContext() !=
+      NewFD->getDeclContext()->getRedeclContext())
+    return false;
+
+  FunctionDecl *OldFD = OldDecl->getAsFunction();
+  // Unresolved 'using' statements (the other way OldDecl can be not a function)
+  // likely cannot cause a problem here.
+  if (!OldFD)
+    return false;
+
+  if (!OldFD->isMultiVersion() && !NewTA)
+    return false;
+
+  if (OldFD->isMultiVersion() && !NewTA) {
+    S.Diag(NewFD->getLocation(), diag::err_target_required_in_redecl);
+    NewFD->setInvalidDecl();
+    return true;
+  }
+
+  TargetAttr::ParsedTargetAttr NewParsed = NewTA->parse();
+  // Sort order doesn't matter, it just needs to be consistent.
+  std::sort(NewParsed.Features.begin(), NewParsed.Features.end());
+
+  const auto *OldTA = OldFD->getAttr<TargetAttr>();
+  if (!OldFD->isMultiVersion()) {
+    // If the old decl is NOT MultiVersioned yet, and we don't cause that
+    // to change, this is a simple redeclaration.
+    if (!OldTA || OldTA->getFeaturesStr() == NewTA->getFeaturesStr())
+      return false;
+
+    // Otherwise, this decl causes MultiVersioning.
+    if (!S.getASTContext().getTargetInfo().supportsMultiVersioning()) {
+      S.Diag(NewFD->getLocation(), diag::err_multiversion_not_supported);
+      S.Diag(OldFD->getLocation(), diag::note_previous_declaration);
+      return true;
+    }
+
+    if (!OldFD->getType()->getAs<FunctionProtoType>()) {
+      S.Diag(OldFD->getLocation(), diag::err_multiversion_noproto);
+      S.Diag(NewFD->getLocation(), diag::note_multiversioning_caused_here);
+      NewFD->setInvalidDecl();
+      return true;
+    }
+
+    if (CheckMultiVersionValue(S, NewFD)) {
+      NewFD->setInvalidDecl();
+      return true;
+    }
+
+    if (CheckMultiVersionValue(S, OldFD)) {
+      S.Diag(NewFD->getLocation(), diag::note_multiversioning_caused_here);
+      NewFD->setInvalidDecl();
+      return true;
+    }
+
+    TargetAttr::ParsedTargetAttr OldParsed =
+        OldTA->parse(std::less<std::string>());
+
+    if (OldParsed == NewParsed) {
+      S.Diag(NewFD->getLocation(), diag::err_multiversion_duplicate);
+      S.Diag(OldFD->getLocation(), diag::note_previous_declaration);
+      NewFD->setInvalidDecl();
+      return true;
+    }
+
+    for (const auto *FD : OldFD->redecls()) {
+      const auto *CurTA = FD->getAttr<TargetAttr>();
+      if (!CurTA || CurTA->isInherited()) {
+        S.Diag(FD->getLocation(), diag::err_target_required_in_redecl);
+        S.Diag(NewFD->getLocation(), diag::note_multiversioning_caused_here);
+        NewFD->setInvalidDecl();
+        return true;
+      }
+    }
+
+    if (CheckMultiVersionAdditionalRules(S, OldFD, NewFD, true)) {
+      NewFD->setInvalidDecl();
+      return true;
+    }
+
+    OldFD->setIsMultiVersion();
+    NewFD->setIsMultiVersion();
+    Redeclaration = false;
+    MergeTypeWithPrevious = false;
+    OldDecl = nullptr;
+    Previous.clear();
+    return false;
+  }
+
+  bool UseMemberUsingDeclRules =
+      S.CurContext->isRecord() && !NewFD->getFriendObjectKind();
+
+  // Next, check ALL non-overloads to see if this is a redeclaration of a
+  // previous member of the MultiVersion set.
+  for (NamedDecl *ND : Previous) {
+    FunctionDecl *CurFD = ND->getAsFunction();
+    if (!CurFD)
+      continue;
+    if (S.IsOverload(NewFD, CurFD, UseMemberUsingDeclRules))
+      continue;
+
+    const auto *CurTA = CurFD->getAttr<TargetAttr>();
+    if (CurTA->getFeaturesStr() == NewTA->getFeaturesStr()) {
+      NewFD->setIsMultiVersion();
+      Redeclaration = true;
+      OldDecl = ND;
+      return false;
+    }
+
+    TargetAttr::ParsedTargetAttr CurParsed =
+        CurTA->parse(std::less<std::string>());
+
+    if (CurParsed == NewParsed) {
+      S.Diag(NewFD->getLocation(), diag::err_multiversion_duplicate);
+      S.Diag(CurFD->getLocation(), diag::note_previous_declaration);
+      NewFD->setInvalidDecl();
+      return true;
+    }
+  }
+
+  // Else, this is simply a non-redecl case.
+  if (CheckMultiVersionValue(S, NewFD)) {
+    NewFD->setInvalidDecl();
+    return true;
+  }
+
+  if (CheckMultiVersionAdditionalRules(S, OldFD, NewFD, false)) {
+    NewFD->setInvalidDecl();
+    return true;
+  }
+
+  NewFD->setIsMultiVersion();
+  Redeclaration = false;
+  MergeTypeWithPrevious = false;
+  OldDecl = nullptr;
+  Previous.clear();
+  return false;
+}
+
 /// \brief Perform semantic checking of a new function declaration.
 ///
 /// Performs semantic analysis of the new function declaration
 /// NewFD. This routine performs all semantic checking that does not
 /// require the actual declarator involved in the declaration, and is
 /// used both for the declaration of functions as they are parsed
 /// (called via ActOnDeclarator) and for the declaration of functions
 /// that have been instantiated via C++ template instantiation (called
if (!Previous.empty()) {
           MayNeedOverloadableChecks = true;
           Redeclaration = false;
           OldDecl = nullptr;
         }
       }
     }
   }
 
+  if (CheckMultiVersionFunction(*this, NewFD, Redeclaration, OldDecl,
+                                MergeTypeWithPrevious, Previous))
+    return Redeclaration;
+
   // C++11 [dcl.constexpr]p8:
   //   A constexpr specifier for a non-static member function that is not
   //   a constructor declares that member function to be const.
   //
   // This needs to be delayed until we know whether this is an out-of-line
   // definition of a static member function.
   //
   // This rule is not present in C++1y, so we produce a backwards

lib/Sema/SemaOverload.cpp

OverloadCandidate &Candidate =
       CandidateSet.addCandidate(Args.size(), EarlyConversions);
   Candidate.FoundDecl = FoundDecl;
   Candidate.Function = Function;
   Candidate.Viable = true;
   Candidate.IsSurrogate = false;
   Candidate.IgnoreObjectArgument = false;
   Candidate.ExplicitCallArguments = Args.size();
 
+  if (Function->isMultiVersion() &&
+      !Function->getAttr<TargetAttr>()->isDefaultVersion()) {
+    Candidate.Viable = false;
+    Candidate.FailureKind = ovl_non_default_multiversion_function;
+    return;
+  }
+
   if (Constructor) {
     // C++ [class.copy]p3:
     //   A member function template is never instantiated to perform the copy
     //   of a class object to an object of its class type.
     QualType ClassType = Context.getTypeDeclType(Constructor->getParent());
     if (Args.size() == 1 && Constructor->isSpecializationCopyingObject() &&
         (Context.hasSameUnqualifiedType(ClassType, Args[0]->getType()) ||
          IsDerivedFrom(Args[0]->getLocStart(), Args[0]->getType(),
Sema::AddMethodCandidate(CXXMethodDecl *Method, DeclAccessPair FoundDecl,
   }
 
   if (EnableIfAttr *FailedAttr = CheckEnableIf(Method, Args, true)) {
     Candidate.Viable = false;
     Candidate.FailureKind = ovl_fail_enable_if;
     Candidate.DeductionFailure.Data = FailedAttr;
     return;
   }
+
+  if (Method->isMultiVersion() &&
+      !Method->getAttr<TargetAttr>()->isDefaultVersion()) {
+    Candidate.Viable = false;
+    Candidate.FailureKind = ovl_non_default_multiversion_function;
+  }
 }
 
 /// \brief Add a C++ member function template as a candidate to the candidate
 /// set, using template argument deduction to produce an appropriate member
 /// function template specialization.
 void
 Sema::AddMethodTemplateCandidate(FunctionTemplateDecl *MethodTmpl,
                                  DeclAccessPair FoundDecl,
Sema::AddConversionCandidate(CXXConversionDecl *Conversion,
   }
 
   if (EnableIfAttr *FailedAttr = CheckEnableIf(Conversion, None)) {
     Candidate.Viable = false;
     Candidate.FailureKind = ovl_fail_enable_if;
     Candidate.DeductionFailure.Data = FailedAttr;
     return;
   }
+
+  if (Conversion->isMultiVersion() &&
+      !Conversion->getAttr<TargetAttr>()->isDefaultVersion()) {
+    Candidate.Viable = false;
+    Candidate.FailureKind = ovl_non_default_multiversion_function;
+  }
 }
 
 /// \brief Adds a conversion function template specialization
 /// candidate to the overload set, using template argument deduction
 /// to deduce the template arguments of the conversion function
 /// template from the type that we are converting to (C++
 /// [temp.deduct.conv]).
 void
return ::checkAddressOfFunctionIsAvailable(*this, Function, Complain,
                                              Loc);
 }
 
 // Notes the location of an overload candidate.
 void Sema::NoteOverloadCandidate(NamedDecl *Found, FunctionDecl *Fn,
                                  QualType DestType, bool TakingAddress) {
   if (TakingAddress && !checkAddressOfCandidateIsAvailable(*this, Fn))
     return;
+  if (Fn->isMultiVersion() && !Fn->getAttr<TargetAttr>()->isDefaultVersion())
+    return;
 
   std::string FnDesc;
   OverloadCandidateKind K = ClassifyOverloadCandidate(*this, Found, Fn, FnDesc);
   PartialDiagnostic PD = PDiag(diag::note_ovl_candidate)
                              << (unsigned) K << Fn << FnDesc;
 
   HandleFunctionTypeMismatch(PD, Fn->getType(), DestType);
   Diag(Fn->getLocation(), PD);
case ovl_fail_inhctor_slice:
     return;
 
   case ovl_fail_addr_not_available: {
     bool Available = checkAddressOfCandidateIsAvailable(S, Cand->Function);
     (void)Available;
     assert(!Available);
     break;
   }
+  case ovl_non_default_multiversion_function:
+    // Do Nothing, these should simly be ignored.
+    break;
   }
 }
 
 static void NoteSurrogateCandidate(Sema &S, OverloadCandidate *Cand) {
   // Desugar the type of the surrogate down to a function type,
   // retaining as many typedefs as possible while still showing
   // the function type (and, therefore, its parameter types).
   QualType FnType = Cand->Surrogate->getConversionType();
bool AddMatchingNonTemplateFunction(NamedDecl* Fn,
     else if (TargetTypeIsNonStaticMemberFunction)
       return false;
 
     if (FunctionDecl *FunDecl = dyn_cast<FunctionDecl>(Fn)) {
       if (S.getLangOpts().CUDA)
         if (FunctionDecl *Caller = dyn_cast<FunctionDecl>(S.CurContext))
           if (!Caller->isImplicit() && !S.IsAllowedCUDACall(Caller, FunDecl))
             return false;
+      if (FunDecl->isMultiVersion()) {
+        const auto *TA = FunDecl->getAttr<TargetAttr>();
+        assert(TA && "Multiversioned functions require a target attribute");
+        if (!TA->isDefaultVersion())
+          return false;
+      }
 
       // If any candidate has a placeholder return type, trigger its deduction
       // now.
       if (completeFunctionType(S, FunDecl, SourceExpr->getLocStart(),
                                Complain)) {
         HasComplained |= Complain;
         return false;
       }

lib/Serialization/ASTReaderDecl.cpp

void ASTDeclReader::VisitFunctionDecl(FunctionDecl *FD) {
   FD->IsDeleted = Record.readInt();
   FD->IsTrivial = Record.readInt();
   FD->IsDefaulted = Record.readInt();
   FD->IsExplicitlyDefaulted = Record.readInt();
   FD->HasImplicitReturnZero = Record.readInt();
   FD->IsConstexpr = Record.readInt();
   FD->UsesSEHTry = Record.readInt();
   FD->HasSkippedBody = Record.readInt();
+  FD->IsMultiVersion = Record.readInt();
   FD->IsLateTemplateParsed = Record.readInt();
   FD->setCachedLinkage(Linkage(Record.readInt()));
   FD->EndRangeLoc = ReadSourceLocation();
 
   FD->ODRHash = Record.readInt();
   FD->HasODRHash = true;
 
   switch ((FunctionDecl::TemplatedKind)Record.readInt()) {
if (FunctionDecl *FuncX = dyn_cast<FunctionDecl>(X)) {
     FunctionDecl *FuncY = cast<FunctionDecl>(Y);
     if (CXXConstructorDecl *CtorX = dyn_cast<CXXConstructorDecl>(X)) {
       CXXConstructorDecl *CtorY = cast<CXXConstructorDecl>(Y);
       if (CtorX->getInheritedConstructor() &&
           !isSameEntity(CtorX->getInheritedConstructor().getConstructor(),
                         CtorY->getInheritedConstructor().getConstructor()))
         return false;
     }
+    // Multiversioned functions with different feature strings are represented
+    // as separate declarations.
+    if (FuncX->isMultiVersion()) {
+      const auto *TAX = FuncX->getAttr<TargetAttr>();
+      const auto *TAY = FuncY->getAttr<TargetAttr>();
+      assert(TAX && "Multiversion Function without target attribute");
+
+      if (!TAY || TAX->getFeaturesStr() != TAY->getFeaturesStr())
+        return false;
+    }
     ASTContext &C = FuncX->getASTContext();
     if (!C.hasSameType(FuncX->getType(), FuncY->getType())) {
       // We can get functions with different types on the redecl chain in C++17
       // if they have differing exception specifications and at least one of
       // the excpetion specs is unresolved.
       // FIXME: Do we need to check for C++14 deduced return types here too?
       auto *XFPT = FuncX->getType()->getAs<FunctionProtoType>();
       auto *YFPT = FuncY->getType()->getAs<FunctionProtoType>();
       if (C.getLangOpts().CPlusPlus17 && XFPT && YFPT &&
           (isUnresolvedExceptionSpec(XFPT->getExceptionSpecType()) ||

lib/Serialization/ASTWriterDecl.cpp

void ASTDeclWriter::VisitFunctionDecl(FunctionDecl *D) {
   Record.push_back(D->IsDeleted);
   Record.push_back(D->IsTrivial);
   Record.push_back(D->IsDefaulted);
   Record.push_back(D->IsExplicitlyDefaulted);
   Record.push_back(D->HasImplicitReturnZero);
   Record.push_back(D->IsConstexpr);
   Record.push_back(D->UsesSEHTry);
   Record.push_back(D->HasSkippedBody);
+  Record.push_back(D->IsMultiVersion);
   Record.push_back(D->IsLateTemplateParsed);
   Record.push_back(D->getLinkageInternal());
   Record.AddSourceLocation(D->getLocEnd());
 
   Record.push_back(D->getODRHash());
 
   Record.push_back(D->getTemplatedKind());
   switch (D->getTemplatedKind()) {
void ASTWriter::WriteDeclAbbrevs() {
   Abv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Fixed, 1)); // Deleted
   Abv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Fixed, 1)); // Trivial
   Abv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Fixed, 1)); // Defaulted
   Abv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Fixed, 1)); // ExplicitlyDefaulted
   Abv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Fixed, 1)); // ImplicitReturnZero
   Abv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Fixed, 1)); // Constexpr
   Abv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Fixed, 1)); // UsesSEHTry
   Abv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Fixed, 1)); // SkippedBody
+  Abv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Fixed, 1)); // MultiVersion
   Abv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Fixed, 1)); // LateParsed
   Abv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Fixed, 3)); // Linkage
   Abv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::VBR, 6));   // LocEnd
   Abv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Fixed, 32)); // ODRHash
   Abv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Fixed, 3)); // TemplateKind
   // This Array slurps the rest of the record. Fortunately we want to encode
   // (nearly) all the remaining (variable number of) fields in the same way.
   //

test/CodeGen/attr-target-mv-func-ptrs.c

+// RUN: %clang_cc1 -triple x86_64-linux-gnu -emit-llvm %s -o - | FileCheck %s
+int __attribute__((target("sse4.2"))) foo(int i) { return 0; }
+int __attribute__((target("arch=sandybridge"))) foo(int);
+int __attribute__((target("arch=ivybridge"))) foo(int i) {return 1;}
+int __attribute__((target("default"))) foo(int i) { return 2; }
+
+typedef int (*FuncPtr)(int);
+void func(FuncPtr);
+
+int bar() {
+  func(foo);
+  FuncPtr Free = &foo;
+  FuncPtr Free2 = foo;
+
+  return 0;
+  return Free(1) + Free(2);
+}
+
+// CHECK: @foo.ifunc = ifunc i32 (i32), i32 (i32)* ()* @foo.resolver
+// CHECK: define i32 @foo.sse4.2(
+// CHECK: ret i32 0
+// CHECK: define i32 @foo.arch_ivybridge(
+// CHECK: ret i32 1
+// CHECK: define i32 @foo(
+// CHECK: ret i32 2
+
+// CHECK: define i32 @bar()
+// CHECK: call void @func(i32 (i32)* @foo.ifunc)
+// CHECK: store i32 (i32)* @foo.ifunc
+// CHECK: store i32 (i32)* @foo.ifunc
+
+// CHECK: declare i32 @foo.arch_sandybridge(

test/CodeGen/attr-target-mv-va-args.c

+// RUN: %clang_cc1 -triple x86_64-linux-gnu -emit-llvm %s -o - | FileCheck %s
+int __attribute__((target("sse4.2"))) foo(int i, ...) { return 0; }
+int __attribute__((target("arch=sandybridge"))) foo(int i, ...);
+int __attribute__((target("arch=ivybridge"))) foo(int i, ...) {return 1;}
+int __attribute__((target("default"))) foo(int i, ...) { return 2; }
+
+int bar() {
+  return foo(1, 'a', 1.1) + foo(2, 2.2, "asdf");
+}
+
+// CHECK: @foo.ifunc = ifunc i32 (i32, ...), i32 (i32, ...)* ()* @foo.resolver
+// CHECK: define i32 @foo.sse4.2(i32 %i, ...)
+// CHECK: ret i32 0
+// CHECK: define i32 @foo.arch_ivybridge(i32 %i, ...)
+// CHECK: ret i32 1
+// CHECK: define i32 @foo(i32 %i, ...)
+// CHECK: ret i32 2
+// CHECK: define i32 @bar()
+// CHECK: call i32 (i32, ...) @foo.ifunc(i32 1, i32 97, double
+// CHECK: call i32 (i32, ...) @foo.ifunc(i32 2, double 2.2{{[0-9Ee+]+}}, i8* getelementptr inbounds 
+// CHECK: define i32 (i32, ...)* @foo.resolver()
+// CHECK: ret i32 (i32, ...)* @foo.arch_sandybridge
+// CHECK: ret i32 (i32, ...)* @foo.arch_ivybridge
+// CHECK: ret i32 (i32, ...)* @foo.sse4.2
+// CHECK: ret i32 (i32, ...)* @foo
+// CHECK: declare i32 @foo.arch_sandybridge(i32, ...)

test/CodeGen/attr-target-mv.c

+// RUN: %clang_cc1 -triple x86_64-linux-gnu -emit-llvm %s -o - | FileCheck %s
+int __attribute__((target("sse4.2"))) foo(void) { return 0; }
+int __attribute__((target("arch=sandybridge"))) foo(void);
+int __attribute__((target("arch=ivybridge"))) foo(void) {return 1;}
+int __attribute__((target("default"))) foo(void) { return 2; }
+
+int bar() {
+  return foo();
+}
+
+inline int __attribute__((target("sse4.2"))) foo_inline(void) { return 0; }
+inline int __attribute__((target("arch=sandybridge"))) foo_inline(void);
+inline int __attribute__((target("arch=ivybridge"))) foo_inline(void) {return 1;}
+inline int __attribute__((target("default"))) foo_inline(void) { return 2; }
+
+int bar2() {
+  return foo_inline();
+}
+
+// CHECK: @foo.ifunc = ifunc i32 (), i32 ()* ()* @foo.resolver
+// CHECK: @foo_inline.ifunc = ifunc i32 (), i32 ()* ()* @foo_inline.resolver
+
+// CHECK: define i32 @foo.sse4.2()
+// CHECK: ret i32 0
+// CHECK: define i32 @foo.arch_ivybridge()
+// CHECK: ret i32 1
+// CHECK: define i32 @foo()
+// CHECK: ret i32 2
+// CHECK: define i32 @bar()
+// CHECK: call i32 @foo.ifunc()
+
+// CHECK: define i32 ()* @foo.resolver()
+// CHECK: call void @__cpu_indicator_init()
+// CHECK: ret i32 ()* @foo.arch_sandybridge
+// CHECK: ret i32 ()* @foo.arch_ivybridge
+// CHECK: ret i32 ()* @foo.sse4.2
+// CHECK: ret i32 ()* @foo
+
+// CHECK: define i32 @bar2()
+// CHECK: call i32 @foo_inline.ifunc()
+
+// CHECK: define i32 ()* @foo_inline.resolver()
+// CHECK: call void @__cpu_indicator_init()
+// CHECK: ret i32 ()* @foo_inline.arch_sandybridge
+// CHECK: ret i32 ()* @foo_inline.arch_ivybridge
+// CHECK: ret i32 ()* @foo_inline.sse4.2
+// CHECK: ret i32 ()* @foo_inline
+
+// CHECK: declare i32 @foo.arch_sandybridge()
+
+// CHECK: define available_externally i32 @foo_inline.sse4.2()
+// CHECK: ret i32 0
+
+// CHECK: declare i32 @foo_inline.arch_sandybridge()
+//
+// CHECK: define available_externally i32 @foo_inline.arch_ivybridge()
+// CHECK: ret i32 1
+// CHECK: define available_externally i32 @foo_inline()
+// CHECK: ret i32 2
+
+

test/CodeGenCXX/attr-target-mv-constexpr.cpp

+// RUN: %clang_cc1 -std=c++11 -triple x86_64-linux-gnu -emit-llvm %s -o - | FileCheck %s
+
+constexpr int __attribute__((target("sse4.2"))) foo(void) { return 0; }
+constexpr int __attribute__((target("arch=sandybridge"))) foo(void);
+constexpr int __attribute__((target("arch=ivybridge"))) foo(void) {return 1;}
+constexpr int __attribute__((target("default"))) foo(void) { return 2; }
+
+int bar() {
+  constexpr int i = foo();
+  return i;
+}
+
+// CHECK-NOT: foo
+// CHECK: define i32 @_Z3barv()
+// CHECK: ret i32 2
+// CHECK-NOT: foo

test/CodeGenCXX/attr-target-mv-diff-ns.cpp

+// RUN: %clang_cc1 -std=c++11 -triple x86_64-linux-gnu -emit-llvm %s -o - | FileCheck %s
+// Test ensures that this properly differentiates between types in different 
+// namespaces.
+int __attribute__((target("sse4.2"))) foo(int) { return 0; }
+int __attribute__((target("arch=sandybridge"))) foo(int);
+int __attribute__((target("arch=ivybridge"))) foo(int) {return 1;}
+int __attribute__((target("default"))) foo(int) { return 2; }
+
+namespace ns {
+int __attribute__((target("sse4.2"))) foo(int) { return 0; }
+int __attribute__((target("arch=sandybridge"))) foo(int);
+int __attribute__((target("arch=ivybridge"))) foo(int) {return 1;}
+int __attribute__((target("default"))) foo(int) { return 2; }
+}
+
+int bar() {
+  return foo(1) + ns::foo(2);
+}
+
+// CHECK: @_Z3fooi.ifunc = ifunc i32 (i32), i32 (i32)* ()* @_Z3fooi.resolver
+// CHECK: @_ZN2ns3fooEi.ifunc = ifunc i32 (i32), i32 (i32)* ()* @_ZN2ns3fooEi.resolver
+
+// CHECK: define i32 @_Z3fooi.sse4.2(i32)
+// CHECK: ret i32 0
+// CHECK: define i32 @_Z3fooi.arch_ivybridge(i32)
+// CHECK: ret i32 1
+// CHECK: define i32 @_Z3fooi(i32)
+// CHECK: ret i32 2
+
+// CHECK: define i32 @_ZN2ns3fooEi.sse4.2(i32)
+// CHECK: ret i32 0
+// CHECK: define i32 @_ZN2ns3fooEi.arch_ivybridge(i32)
+// CHECK: ret i32 1
+// CHECK: define i32 @_ZN2ns3fooEi(i32)
+// CHECK: ret i32 2
+
+// CHECK: define i32 @_Z3barv()
+// CHECK: call i32 @_Z3fooi.ifunc(i32 1)
+// CHECK: call i32 @_ZN2ns3fooEi.ifunc(i32 2)
+
+// CHECK: define i32 (i32)* @_Z3fooi.resolver()
+// CHECK: ret i32 (i32)* @_Z3fooi.arch_sandybridge
+// CHECK: ret i32 (i32)* @_Z3fooi.arch_ivybridge
+// CHECK: ret i32 (i32)* @_Z3fooi.sse4.2
+// CHECK: ret i32 (i32)* @_Z3fooi
+//
+// CHECK: define i32 (i32)* @_ZN2ns3fooEi.resolver()
+// CHECK: ret i32 (i32)* @_ZN2ns3fooEi.arch_sandybridge
+// CHECK: ret i32 (i32)* @_ZN2ns3fooEi.arch_ivybridge
+// CHECK: ret i32 (i32)* @_ZN2ns3fooEi.sse4.2
+// CHECK: ret i32 (i32)* @_ZN2ns3fooEi
+
+// CHECK: declare i32 @_Z3fooi.arch_sandybridge(i32)
+// CHECK: declare i32 @_ZN2ns3fooEi.arch_sandybridge(i32)

test/CodeGenCXX/attr-target-mv-func-ptrs.cpp

+// RUN: %clang_cc1 -std=c++11 -triple x86_64-linux-gnu -emit-llvm %s -o - | FileCheck %s
+void temp();
+void temp(int);
+using FP = void(*)(int);
+void b() {
+  FP f = temp; 
+}
+
+int __attribute__((target("sse4.2"))) foo(int) { return 0; }
+int __attribute__((target("arch=sandybridge"))) foo(int);
+int __attribute__((target("arch=ivybridge"))) foo(int) {return 1;}
+int __attribute__((target("default"))) foo(int) { return 2; }
+
+struct S {
+int __attribute__((target("sse4.2"))) foo(int) { return 0; }
+int __attribute__((target("arch=sandybridge"))) foo(int);
+int __attribute__((target("arch=ivybridge"))) foo(int) {return 1;}
+int __attribute__((target("default"))) foo(int) { return 2; }
+};
+
+using FuncPtr = int (*)(int);
+using MemFuncPtr = int (S::*)(int);
+
+void f(FuncPtr, MemFuncPtr);
+
+int bar() {
+  FuncPtr Free = &foo;
+  MemFuncPtr Member = &S::foo;
+  S s;
+  f(foo, &S::foo);
+  return Free(1) + (s.*Member)(2);
+}
+
+
+// CHECK: @_Z3fooi.ifunc 
+// CHECK: @_ZN1S3fooEi.ifunc
+
+// CHECK: define i32 @_Z3barv()
+// Store to Free of ifunc
+// CHECK: store i32 (i32)* @_Z3fooi.ifunc
+// Store to Member of ifunc
+// CHECK: store { i64, i64 } { i64 ptrtoint (i32 (%struct.S*, i32)* @_ZN1S3fooEi.ifunc to i64), i64 0 }, { i64, i64 }* [[MEMBER:%[a-z]+]]
+
+// Call to 'f' with the ifunc
+// CHECK: call void @_Z1fPFiiEM1SFiiE(i32 (i32)* @_Z3fooi.ifunc

test/CodeGenCXX/attr-target-mv-member-funcs.cpp

+// RUN: %clang_cc1 -std=c++11 -triple x86_64-linux-gnu -emit-llvm %s -o - | FileCheck %s
+
+struct S {
+  int __attribute__((target("sse4.2"))) foo(int) { return 0; }
+  int __attribute__((target("arch=sandybridge"))) foo(int);
+  int __attribute__((target("arch=ivybridge"))) foo(int) { return 1; }
+  int __attribute__((target("default"))) foo(int) { return 2; }
+
+  S &__attribute__((target("arch=ivybridge"))) operator=(const S &) {
+    return *this;
+  }
+  S &__attribute__((target("default"))) operator=(const S &) {
+    return *this;
+  }
+  // TODO:
+  // befriending mv functions
+};
+
+struct ConvertTo {
+  __attribute__((target("arch=ivybridge"))) operator S() const {
+    return S{};
+  }
+  __attribute__((target("default"))) operator S() const {
+    return S{};
+  }
+};
+
+int bar() {
+  S s;
+  S s2;
+  s2 = s;
+
+  ConvertTo C;
+  s2 = static_cast<S>(C);
+
+  return s.foo(0);
+}
+
+template<typename T>
+struct templ {
+  int __attribute__((target("sse4.2"))) foo(int) { return 0; }
+  int __attribute__((target("arch=sandybridge"))) foo(int);
+  int __attribute__((target("arch=ivybridge"))) foo(int) { return 1; }
+  int __attribute__((target("default"))) foo(int) { return 2; }
+};
+
+int templ_use() {
+  templ<int> a;
+  templ<double> b;
+  return a.foo(1) + b.foo(2);
+}
+
+
+
+// CHECK: @_ZN1SaSERKS_.ifunc = ifunc %struct.S* (%struct.S*, %struct.S*), %struct.S* (%struct.S*, %struct.S*)* ()* @_ZN1SaSERKS_.resolver
+// CHECK: @_ZNK9ConvertTocv1SEv.ifunc = ifunc void (%struct.ConvertTo*), void (%struct.ConvertTo*)* ()* @_ZNK9ConvertTocv1SEv.resolver
+// CHECK: @_ZN1S3fooEi.ifunc = ifunc i32 (%struct.S*, i32), i32 (%struct.S*, i32)* ()* @_ZN1S3fooEi.resolver
+// Templates:
+// CHECK: @_ZN5templIiE3fooEi.ifunc = ifunc i32 (%struct.templ*, i32), i32 (%struct.templ*, i32)* ()* @_ZN5templIiE3fooEi.resolver
+// CHECK: @_ZN5templIdE3fooEi.ifunc = ifunc i32 (%struct.templ.0*, i32), i32 (%struct.templ.0*, i32)* ()* @_ZN5templIdE3fooEi.resolver
+
+// CHECK: define i32 @_Z3barv()
+// CHECK: %s = alloca %struct.S, align 1
+// CHECK: %s2 = alloca %struct.S, align 1
+// CHECK: %C = alloca %struct.ConvertTo, align 1
+// CHECK: call dereferenceable(1) %struct.S* @_ZN1SaSERKS_.ifunc(%struct.S* %s2
+// CHECK: call void @_ZNK9ConvertTocv1SEv.ifunc(%struct.ConvertTo* %C)
+// CHECK: call dereferenceable(1) %struct.S* @_ZN1SaSERKS_.ifunc(%struct.S* %s2
+// CHECK: call i32 @_ZN1S3fooEi.ifunc(%struct.S* %s, i32 0)
+
+// CHECK: define %struct.S* (%struct.S*, %struct.S*)* @_ZN1SaSERKS_.resolver()
+// CHECK: ret %struct.S* (%struct.S*, %struct.S*)* @_ZN1SaSERKS_.arch_ivybridge
+// CHECK: ret %struct.S* (%struct.S*, %struct.S*)* @_ZN1SaSERKS_
+
+// CHECK: define void (%struct.ConvertTo*)* @_ZNK9ConvertTocv1SEv.resolver()
+// CHECK: ret void (%struct.ConvertTo*)* @_ZNK9ConvertTocv1SEv.arch_ivybridge
+// CHECK: ret void (%struct.ConvertTo*)* @_ZNK9ConvertTocv1SEv
+
+// CHECK: define i32 (%struct.S*, i32)* @_ZN1S3fooEi.resolver()
+// CHECK: ret i32 (%struct.S*, i32)* @_ZN1S3fooEi.arch_sandybridge
+// CHECK: ret i32 (%struct.S*, i32)* @_ZN1S3fooEi.arch_ivybridge
+// CHECK: ret i32 (%struct.S*, i32)* @_ZN1S3fooEi.sse4.2
+// CHECK: ret i32 (%struct.S*, i32)* @_ZN1S3fooEi
+
+// CHECK: define i32 @_Z9templ_usev()
+// CHECK:  call i32 @_ZN5templIiE3fooEi.ifunc
+// CHECK:  call i32 @_ZN5templIdE3fooEi.ifunc
+
+
+// CHECK: define i32 (%struct.templ*, i32)* @_ZN5templIiE3fooEi.resolver()
+// CHECK: ret i32 (%struct.templ*, i32)* @_ZN5templIiE3fooEi.arch_sandybridge
+// CHECK: ret i32 (%struct.templ*, i32)* @_ZN5templIiE3fooEi.arch_ivybridge
+// CHECK: ret i32 (%struct.templ*, i32)* @_ZN5templIiE3fooEi.sse4.2
+// CHECK: ret i32 (%struct.templ*, i32)* @_ZN5templIiE3fooEi
+//
+// CHECK: define i32 (%struct.templ.0*, i32)* @_ZN5templIdE3fooEi.resolver()
+// CHECK: ret i32 (%struct.templ.0*, i32)* @_ZN5templIdE3fooEi.arch_sandybridge
+// CHECK: ret i32 (%struct.templ.0*, i32)* @_ZN5templIdE3fooEi.arch_ivybridge
+// CHECK: ret i32 (%struct.templ.0*, i32)* @_ZN5templIdE3fooEi.sse4.2
+// CHECK: ret i32 (%struct.templ.0*, i32)* @_ZN5templIdE3fooEi
+
+// CHECK: define linkonce_odr i32 @_ZN1S3fooEi.sse4.2(%struct.S* %this, i32)
+// CHECK: ret i32 0
+
+// CHECK: declare i32 @_ZN1S3fooEi.arch_sandybridge(%struct.S*, i32)
+
+// CHECK: define linkonce_odr i32 @_ZN1S3fooEi.arch_ivybridge(%struct.S* %this, i32)
+// CHECK: ret i32 1
+
+// CHECK: define linkonce_odr i32 @_ZN1S3fooEi(%struct.S* %this, i32)
+// CHECK: ret i32 2
+

test/CodeGenCXX/attr-target-mv-out-of-line-defs.cpp

+// RUN: %clang_cc1 -std=c++11 -triple x86_64-linux-gnu -emit-llvm %s -o - | FileCheck %s
+struct S {
+  int __attribute__((target("sse4.2"))) foo(int);
+  int __attribute__((target("arch=sandybridge"))) foo(int);
+  int __attribute__((target("arch=ivybridge"))) foo(int);
+  int __attribute__((target("default"))) foo(int);
+};
+
+int __attribute__((target("default"))) S::foo(int) { return 2; }
+int __attribute__((target("sse4.2"))) S::foo(int) { return 0; }
+int __attribute__((target("arch=ivybridge"))) S::foo(int) { return 1; }
+
+int bar() {
+  S s;
+  return s.foo(0);
+}
+
+// CHECK: @_ZN1S3fooEi.ifunc = ifunc i32 (%struct.S*, i32), i32 (%struct.S*, i32)* ()* @_ZN1S3fooEi.resolver
+
+// CHECK: define i32 @_ZN1S3fooEi(%struct.S* %this, i32)
+// CHECK: ret i32 2
+
+// CHECK: define i32 @_ZN1S3fooEi.sse4.2(%struct.S* %this, i32)
+// CHECK: ret i32 0
+
+// CHECK: define i32 @_ZN1S3fooEi.arch_ivybridge(%struct.S* %this, i32)
+// CHECK: ret i32 1
+
+// CHECK: define i32 @_Z3barv()
+// CHECK: %s = alloca %struct.S, align 1
+// CHECK: %call = call i32 @_ZN1S3fooEi.ifunc(%struct.S* %s, i32 0)
+
+// CHECK: define i32 (%struct.S*, i32)* @_ZN1S3fooEi.resolver()
+// CHECK: ret i32 (%struct.S*, i32)* @_ZN1S3fooEi.arch_sandybridge
+// CHECK: ret i32 (%struct.S*, i32)* @_ZN1S3fooEi.arch_ivybridge
+// CHECK: ret i32 (%struct.S*, i32)* @_ZN1S3fooEi.sse4.2
+// CHECK: ret i32 (%struct.S*, i32)* @_ZN1S3fooEi
+
+// CHECK: declare i32 @_ZN1S3fooEi.arch_sandybridge(%struct.S*, i32)

test/CodeGenCXX/attr-target-mv-overloads.cpp

+// RUN: %clang_cc1 -std=c++11 -triple x86_64-linux-gnu -emit-llvm %s -o - | FileCheck %s
+
+int __attribute__((target("sse4.2"))) foo_overload(int) { return 0; }
+int __attribute__((target("arch=sandybridge"))) foo_overload(int);
+int __attribute__((target("arch=ivybridge"))) foo_overload(int) {return 1;}
+int __attribute__((target("default"))) foo_overload(int) { return 2; }
+int __attribute__((target("sse4.2"))) foo_overload(void) { return 0; }
+int __attribute__((target("arch=sandybridge"))) foo_overload(void);
+int __attribute__((target("arch=ivybridge"))) foo_overload(void) {return 1;}
+int __attribute__((target("default"))) foo_overload(void) { return 2; }
+
+int bar2() {
+  return foo_overload() + foo_overload(1);
+}
+
+// CHECK: @_Z12foo_overloadv.ifunc = ifunc i32 (), i32 ()* ()* @_Z12foo_overloadv.resolver
+// CHECK: @_Z12foo_overloadi.ifunc = ifunc i32 (i32), i32 (i32)* ()* @_Z12foo_overloadi.resolver
+
+
+// CHECK: define i32 @_Z12foo_overloadi.sse4.2(i32)
+// CHECK: ret i32 0
+// CHECK: define i32 @_Z12foo_overloadi.arch_ivybridge(i32)
+// CHECK: ret i32 1
+// CHECK: define i32 @_Z12foo_overloadi(i32)
+// CHECK: ret i32 2
+// CHECK: define i32 @_Z12foo_overloadv.sse4.2()
+// CHECK: ret i32 0
+// CHECK: define i32 @_Z12foo_overloadv.arch_ivybridge()
+// CHECK: ret i32 1
+// CHECK: define i32 @_Z12foo_overloadv()
+// CHECK: ret i32 2
+
+// CHECK: define i32 @_Z4bar2v()
+// CHECK: call i32 @_Z12foo_overloadv.ifunc()
+// CHECK: call i32 @_Z12foo_overloadi.ifunc(i32 1)
+
+// CHECK: define i32 ()* @_Z12foo_overloadv.resolver()
+// CHECK: ret i32 ()* @_Z12foo_overloadv.arch_sandybridge
+// CHECK: ret i32 ()* @_Z12foo_overloadv.arch_ivybridge
+// CHECK: ret i32 ()* @_Z12foo_overloadv.sse4.2
+// CHECK: ret i32 ()* @_Z12foo_overloadv
+
+// CHECK: define i32 (i32)* @_Z12foo_overloadi.resolver()
+// CHECK: ret i32 (i32)* @_Z12foo_overloadi.arch_sandybridge
+// CHECK: ret i32 (i32)* @_Z12foo_overloadi.arch_ivybridge
+// CHECK: ret i32 (i32)* @_Z12foo_overloadi.sse4.2
+// CHECK: ret i32 (i32)* @_Z12foo_overloadi
+
+// CHECK: declare i32 @_Z12foo_overloadv.arch_sandybridge()
+// CHECK: declare i32 @_Z12foo_overloadi.arch_sandybridge(i32)

test/Sema/attr-target-mv-bad-target.c

+// RUN: %clang_cc1 -triple x86_64-windows-pc  -fsyntax-only -verify %s
+// RUN: %clang_cc1 -triple arm-none-eabi  -fsyntax-only -verify %s
+
+int __attribute__((target("sse4.2"))) redecl1(void) { return 1; }
+//expected-error@+2 {{function multiversioning is not supported on the current target}}
+//expected-note@-2 {{previous declaration is here}}
+int __attribute__((target("avx")))  redecl1(void) { return 2; }
+
+//expected-error@+1 {{function multiversioning is not supported on the current target}}
+int __attribute__((target("default"))) with_def(void) { return 1;}

test/Sema/attr-target-mv.c

+// RUN: %clang_cc1 -triple x86_64-linux-gnu  -fsyntax-only -verify %s
+
+void __attribute__((target("sse4.2"))) no_default(void);
+void __attribute__((target("arch=sandybridge")))  no_default(void);
+
+void use1(void){
+  // expected-error@+1 {{no matching function for call to 'no_default'}}
+  no_default();
+}
+
+void __attribute__((target("sse4.2"))) has_def(void);
+void __attribute__((target("default")))  has_def(void);
+
+void use2(void){
+  // expected-error@+2 {{reference to overloaded function could not be resolved; did you mean to call it?}}
+  // expected-note@-4 {{possible target for call}}
+  +has_def;
+}
+
+int __attribute__((target("sse4.2"))) no_proto();
+// expected-error@-1 {{multiversioned function must have a prototype}}
+// expected-note@+1 {{function multiversioning caused by this declaration}}
+int __attribute__((target("arch=sandybridge"))) no_proto();
+
+// The following should all be legal, since they are just redeclarations.
+int __attribute__((target("sse4.2"))) redecl1(void);
+int __attribute__((target("sse4.2"))) redecl1(void) { return 1; }
+int __attribute__((target("arch=sandybridge")))  redecl1(void) { return 2; }
+
+int __attribute__((target("sse4.2"))) redecl2(void) { return 1; }
+int __attribute__((target("sse4.2"))) redecl2(void);
+int __attribute__((target("arch=sandybridge")))  redecl2(void) { return 2; }
+
+int __attribute__((target("sse4.2"))) redecl3(void) { return 0; }
+int __attribute__((target("arch=ivybridge"))) redecl3(void) { return 1; }
+int __attribute__((target("arch=sandybridge")))  redecl3(void);
+int __attribute__((target("arch=sandybridge")))  redecl3(void) { return 2; }
+
+int __attribute__((target("sse4.2"))) redecl4(void) { return 1; }
+int __attribute__((target("arch=sandybridge")))  redecl4(void) { return 2; }
+int __attribute__((target("arch=sandybridge")))  redecl4(void);
+
+int __attribute__((target("sse4.2"))) redef(void) { return 1; }
+int __attribute__((target("arch=ivybridge"))) redef(void) { return 1; }
+int __attribute__((target("arch=sandybridge")))  redef(void) { return 2; }
+// expected-error@+2 {{redefinition of 'redef'}}
+// expected-note@-2 {{previous definition is here}}
+int __attribute__((target("arch=sandybridge")))  redef(void) { return 2; }
+
+int __attribute__((target("default"))) redef2(void) { return 1;}
+// expected-error@+2 {{redefinition of 'redef2'}}
+// expected-note@-2 {{previous definition is here}}
+int __attribute__((target("default"))) redef2(void) { return 1;}
+
+int __attribute__((target("sse4.2"))) mv_after_use(void) { return 1; }
+int use3(void) {
+  return mv_after_use();
+}
+
+// expected-error@+1 {{function declaration cannot become a multiversioned function after first usage}}
+int __attribute__((target("arch=sandybridge")))  mv_after_use(void) { return 2; }
+
+int __attribute__((target("sse4.2,arch=sandybridge"))) mangle(void) { return 1; }
+//expected-error@+2 {{multiversioned function redeclarations require identical target attributes}}
+//expected-note@-2 {{previous declaration is here}}
+int __attribute__((target("arch=sandybridge,sse4.2")))  mangle(void) { return 2; }
+
+int prev_no_target(void);
+int __attribute__((target("arch=sandybridge")))  prev_no_target(void) { return 2; }
+// expected-error@-2 {{function declaration is missing 'target' attribute in a multiversioned function}}
+// expected-note@+1 {{function multiversioning caused by this declaration}}
+int __attribute__((target("arch=ivybridge")))  prev_no_target(void) { return 2; }
+
+int __attribute__((target("arch=sandybridge")))  prev_no_target2(void);
+int prev_no_target2(void);
+// expected-error@-1 {{function declaration is missing 'target' attribute in a multiversioned function}}
+// expected-note@+1 {{function multiversioning caused by this declaration}}
+int __attribute__((target("arch=ivybridge")))  prev_no_target2(void);
+
+void __attribute__((target("sse4.2"))) addtl_attrs(void);
+//expected-error@+2 {{attribute 'target' multiversioning cannot be combined}}
+//expected-note@+1 {{function multiversioning caused by this declaration}}
+void __attribute__((used,target("arch=sandybridge")))  addtl_attrs(void);
+
+//expected-error@+2 {{attribute 'target' multiversioning cannot be combined}}
+//expected-note@+1 {{function multiversioning caused by this declaration}}
+void __attribute__((target("default"), used)) addtl_attrs2(void);
+
+//expected-error@+2 {{attribute 'target' multiversioning cannot be combined}}
+//expected-note@+2 {{function multiversioning caused by this declaration}}
+void __attribute__((used,target("sse4.2"))) addtl_attrs3(void);
+void __attribute__((target("arch=sandybridge")))  addtl_attrs3(void);
+
+void __attribute__((target("sse4.2"))) addtl_attrs4(void);
+void __attribute__((target("arch=sandybridge")))  addtl_attrs4(void);
+//expected-error@+1 {{attribute 'target' multiversioning cannot be combined}}
+void __attribute__((used,target("arch=ivybridge")))  addtl_attrs4(void);
+
+int __attribute__((target("sse4.2"))) diff_cc(void);
+// expected-error@+2 {{multiversioned function declaration has a different calling convention}}
+// expected-note@+1 {{function multiversioning caused by this declaration}}
+__vectorcall int __attribute__((target("arch=sandybridge")))  diff_cc(void);
+
+int __attribute__((target("sse4.2"))) diff_ret(void);
+// expected-error@+2 {{multiversioned function declaration has a different return type}}
+// expected-note@+1 {{function multiversioning caused by this declaration}}
+short __attribute__((target("arch=sandybridge")))  diff_ret(void);

test/SemaCXX/attr-target-mv.cpp

+// RUN: %clang_cc1 -triple x86_64-linux-gnu  -fsyntax-only -verify -fexceptions -fcxx-exceptions %s -std=c++14
+void __attribute__((target("sse4.2"))) no_default(void);
+void __attribute__((target("arch=sandybridge")))  no_default(void);
+
+void use1(void){
+  // expected-error@+1 {{no matching function for call to 'no_default'}}
+  no_default();
+}
+constexpr int __attribute__((target("sse4.2"))) foo(void) { return 0; }
+constexpr int __attribute__((target("arch=sandybridge"))) foo(void);
+//expected-error@+1 {{multiversioned function declaration has a different constexpr specification}}
+int __attribute__((target("arch=ivybridge"))) foo(void) {return 1;}
+constexpr int __attribute__((target("default"))) foo(void) { return 2; }
+
+int __attribute__((target("sse4.2"))) foo2(void) { return 0; }
+//expected-error@+2 {{multiversioned function declaration has a different constexpr specification}}
+//expected-note@+1 {{function multiversioning caused by this declaration}}
+constexpr int __attribute__((target("arch=sandybridge"))) foo2(void);
+int __attribute__((target("arch=ivybridge"))) foo2(void) {return 1;}
+int __attribute__((target("default"))) foo2(void) { return 2; }
+
+static int __attribute__((target("sse4.2"))) bar(void) { return 0; }
+static int __attribute__((target("arch=sandybridge"))) bar(void);
+//expected-error@+1 {{multiversioned function declaration has a different storage class}}
+int __attribute__((target("arch=ivybridge"))) bar(void) {return 1;}
+static int __attribute__((target("default"))) bar(void) { return 2; }
+
+int __attribute__((target("sse4.2"))) bar2(void) { return 0; }
+//expected-error@+2 {{multiversioned function declaration has a different storage class}}
+//expected-note@+1 {{function multiversioning caused by this declaration}}
+static int __attribute__((target("arch=sandybridge"))) bar2(void);
+int __attribute__((target("arch=ivybridge"))) bar2(void) {return 1;}
+int __attribute__((target("default"))) bar2(void) { return 2; }
+
+
+inline int __attribute__((target("sse4.2"))) baz(void) { return 0; }
+inline int __attribute__((target("arch=sandybridge"))) baz(void);
+//expected-error@+1 {{multiversioned function declaration has a different inline specification}}
+int __attribute__((target("arch=ivybridge"))) baz(void) {return 1;}
+inline int __attribute__((target("default"))) baz(void) { return 2; }
+
+int __attribute__((target("sse4.2"))) baz2(void) { return 0; }
+//expected-error@+2 {{multiversioned function declaration has a different inline specification}}
+//expected-note@+1 {{function multiversioning caused by this declaration}}
+inline int __attribute__((target("arch=sandybridge"))) baz2(void);
+int __attribute__((target("arch=ivybridge"))) baz2(void) {return 1;}
+int __attribute__((target("default"))) baz2(void) { return 2; }
+
+float __attribute__((target("sse4.2"))) bock(void) { return 0; }
+//expected-error@+2 {{multiversioned function declaration has a different return type}}
+//expected-note@+1 {{function multiversioning caused by this declaration}}
+int __attribute__((target("arch=sandybridge"))) bock(void);
+//expected-error@+2 {{multiversioned function declaration has a different return type}}
+//expected-note@+1 {{function multiversioning caused by this declaration}}
+int __attribute__((target("arch=ivybridge"))) bock(void) {return 1;}
+//expected-error@+2 {{multiversioned function declaration has a different return type}}
+//expected-note@+1 {{function multiversioning caused by this declaration}}
+int __attribute__((target("default"))) bock(void) { return 2; }
+
+int __attribute__((target("sse4.2"))) bock2(void) { return 0; }
+//expected-error@+2 {{multiversioned function declaration has a different return type}}
+//expected-note@+1 {{function multiversioning caused by this declaration}}
+float __attribute__((target("arch=sandybridge"))) bock2(void);
+int __attribute__((target("arch=ivybridge"))) bock2(void) {return 1;}
+int __attribute__((target("default"))) bock2(void) { return 2; }
+
+auto __attribute__((target("sse4.2"))) bock3(void) -> int { return 0; }
+//expected-error@+2 {{multiversioned function declaration has a different return type}}
+//expected-note@+1 {{function multiversioning caused by this declaration}}
+auto __attribute__((target("arch=sandybridge"))) bock3(void) -> short { return (short)0;}
+
+int __attribute__((target("sse4.2"))) bock4(void) noexcept(false) { return 0; }
+//expected-error@+2 {{exception specification in declaration does not match previous declaration}}
+//expected-note@-2 {{previous declaration is here}}
+int __attribute__((target("arch=sandybridge"))) bock4(void) noexcept(true) { return 1;}
+
+// FIXME: Add support for templates and virtual functions!
+template<typename T>
+int __attribute__((target("sse4.2"))) foo(T) { return 0; }
+// expected-error@+3 {{multiversioned functions do not yet support function templates}}
+// expected-note@+2 {{function multiversioning caused by this declaration}}
+template<typename T>
+int __attribute__((target("arch=sandybridge"))) foo(T);
+
+// expected-error@+3 {{multiversioned functions do not yet support function templates}}
+// expected-note@+2 {{function multiversioning caused by this declaration}}
+template<typename T>
+int __attribute__((target("default"))) foo(T) { return 2; }
+
+struct S {
+  template<typename T>
+  int __attribute__((target("sse4.2"))) foo(T) { return 0; }
+  // expected-error@+3 {{multiversioned functions do not yet support function templates}}
+  // expected-note@+2 {{function multiversioning caused by this declaration}}
+  template<typename T>
+  int __attribute__((target("arch=sandybridge"))) foo(T);
+
+  // expected-error@+3 {{multiversioned functions do not yet support function templates}}
+  // expected-note@+2 {{function multiversioning caused by this declaration}}
+  template<typename T>
+  int __attribute__((target("default"))) foo(T) { return 2; }
+
+  // expected-error@+2 {{multiversioned functions do not yet support virtual functions}}
+  // expected-note@+1 {{function multiversioning caused by this declaration}}
+  virtual void __attribute__((target("default"))) virt();
+};
+
+extern "C" {
+int __attribute__((target("sse4.2"))) diff_mangle(void) { return 0; }
+}
+//expected-error@+2 {{multiversioned function declaration has a different linkage}}
+//expected-note@+1 {{function multiversioning caused by this declaration}}
+int __attribute__((target("arch=sandybridge"))) diff_mangle(void) { return 0; }
+
+// expected-error@+2 {{multiversioned functions do not yet support deduced return types}}
+// expected-note@+1 {{function multiversioning caused by this declaration}}
+auto __attribute__((target("default"))) deduced_return(void) { return 0; }
+// expected-error@-1 {{cannot initialize return object of type 'auto' with an rvalue of type 'int'}}
+
+auto __attribute__((target("default"))) trailing_return(void)-> int { return 0; }
+
+__attribute__((target("default"))) void DiffDecl();
+namespace N {
+using ::DiffDecl;
+// expected-error@+3 {{declaration conflicts with target of using declaration already in scope}}
+// expected-note@-4 {{target of using declaration}}
+// expected-note@-3 {{using declaration}}
+__attribute__((target("arch=sandybridge"))) void DiffDecl();
+} // namespace N
+
+struct SpecialFuncs {
+  // expected-error@+2 {{multiversioned functions do not yet support constructors}}
+  // expected-note@+1 {{function multiversioning caused by this declaration}}
+  __attribute__((target("default"))) SpecialFuncs();
+  // expected-error@+2 {{multiversioned functions do not yet support destructors}}
+  // expected-note@+1 {{function multiversioning caused by this declaration}}
+  __attribute__((target("default"))) ~SpecialFuncs();
+
+  // expected-error@+2 {{multiversioned functions do not yet support defaulted functions}}
+  // expected-note@+1 {{function multiversioning caused by this declaration}}
+  SpecialFuncs& __attribute__((target("default"))) operator=(const SpecialFuncs&) = default;
+  // expected-error@+2 {{multiversioned functions do not yet support deleted functions}}
+  // expected-note@+1 {{function multiversioning caused by this declaration}}
+  SpecialFuncs& __attribute__((target("default"))) operator=(SpecialFuncs&&) = delete;
+};