[clang] Corrections for target_clones multiversion functions.

[tahonermann]

Summary

This change merges code for emit of target and target_clones multiversion
resolver functions and, in doing so, corrects handling of target_clones
functions that are declared but not defined. Previously, a use of such
a target_clones function would result in an attempted emit of an ifunc
that referenced an undefined resolver function. Ifunc references to
undefined resolver functions are not allowed and, when the LLVM verifier
is not disabled (via '-disable-llvm-verifier'), resulted in the verifier
issuing a "IFunc resolver must be a definition" error and aborting the
compilation. With this change, ifuncs and resolver function definitions
are always emitted for used target_clones functions regardless of whether
the target_clones function is defined (if the function is defined, then
the ifunc and resolver are emitted regardless of whether the function is
used).

This change has the side effect of causing target_clones variants and
resolver functions to be emitted in a different order than they were
previously. This is harmless and is reflected in the updated tests.

[tahonermann]

Removed the in-class declaration of CodeGenModule::EmitTargetClonesResolver().

[tahonermann]

clang/lib/CodeGen/CodeGenModule.cpp

3323

This comment seemed oddly placed and didn't seem to convey used information, so I gave it the boot.

3411–3461

This functionality was all merged into CodeGenModule::emitMultiVersionFunctions() where much of it was already present.

3488–3493

Use of GetOrCreateMultiVersionResolver() here avoids duplication of code to determine how resolver functions are named for each multiversion function kind.

3670–3673

This is the change that effectively ensures that a target_clones function resolver is emitted.

clang/test/CodeGen/attr-target-clones.c

26

target_clones resolvers are now emitted consistently with target resolvers.

152–156

Some checks were simply missing previously.

161–165

Declarations emitted cause the function is no defined!

[tahonermann]

Simplified code to select a resolver function.

[tahonermann]

clang/lib/CodeGen/CodeGenModule.cpp

3495

setLinkage() was previously only called in the ifunc case. I don't know why that would be, so I "fixed" it here.

[erichkeane]

clang/lib/CodeGen/CodeGenModule.cpp

3495

Likely that was the only place we 'noticed' it was a problem?

[tahonermann]

Squashed the addition of tests originally made in D122954 to this review.

[erichkeane]

LGTM.

clang/lib/CodeGen/CodeGenModule.h

private:
   void EmitGlobalFunctionDefinition(GlobalDecl GD, llvm::GlobalValue *GV);
   void EmitMultiVersionFunctionDefinition(GlobalDecl GD, llvm::GlobalValue *GV);
 
   void EmitGlobalVarDefinition(const VarDecl *D, bool IsTentative = false);
   void EmitExternalVarDeclaration(const VarDecl *D);
   void EmitAliasDefinition(GlobalDecl GD);
   void emitIFuncDefinition(GlobalDecl GD);
   void emitCPUDispatchDefinition(GlobalDecl GD);
-  void EmitTargetClonesResolver(GlobalDecl GD);
   void EmitObjCPropertyImplementations(const ObjCImplementationDecl *D);
   void EmitObjCIvarInitializations(ObjCImplementationDecl *D);
 
   // C++ related functions.
 
   void EmitDeclContext(const DeclContext *DC);
   void EmitLinkageSpec(const LinkageSpecDecl *D);
 

clang/lib/CodeGen/CodeGenModule.cpp

 void CodeGenModule::EmitMultiVersionFunctionDefinition(GlobalDecl GD,
                                                        llvm::GlobalValue *GV) {
   const auto *FD = cast<FunctionDecl>(GD.getDecl());
 
   if (FD->isCPUSpecificMultiVersion()) {
     auto *Spec = FD->getAttr<CPUSpecificAttr>();
     for (unsigned I = 0; I < Spec->cpus_size(); ++I)
       EmitGlobalFunctionDefinition(GD.getWithMultiVersionIndex(I), nullptr);
-    // Requires multiple emits.
   } else if (FD->isTargetClonesMultiVersion()) {
     auto *Clone = FD->getAttr<TargetClonesAttr>();
     for (unsigned I = 0; I < Clone->featuresStrs_size(); ++I)
       if (Clone->isFirstOfVersion(I))
         EmitGlobalFunctionDefinition(GD.getWithMultiVersionIndex(I), nullptr);
-    EmitTargetClonesResolver(GD);
+    // Ensure that the resolver function is also emitted.
+    GetOrCreateMultiVersionResolver(GD);
   } else
     EmitGlobalFunctionDefinition(GD, GV);
 }
 
 void CodeGenModule::EmitGlobalDefinition(GlobalDecl GD, llvm::GlobalValue *GV) {
   const auto *D = cast<ValueDecl>(GD.getDecl());
 
   PrettyStackTraceDecl CrashInfo(const_cast<ValueDecl *>(D), D->getLocation(),

 llvm::GlobalValue::LinkageTypes getMultiversionLinkage(CodeGenModule &CGM,
                                                        GlobalDecl GD) {
   const FunctionDecl *FD = cast<FunctionDecl>(GD.getDecl());
   if (FD->getFormalLinkage() == InternalLinkage)
     return llvm::GlobalValue::InternalLinkage;
   return llvm::GlobalValue::WeakODRLinkage;
 }
 
-void CodeGenModule::EmitTargetClonesResolver(GlobalDecl GD) {
-  const auto *FD = cast<FunctionDecl>(GD.getDecl());
-  assert(FD && "Not a FunctionDecl?");
-  const auto *TC = FD->getAttr<TargetClonesAttr>();
-  assert(TC && "Not a target_clones Function?");
-
-  llvm::Function *ResolverFunc;
-  if (getTarget().supportsIFunc()) {
-    auto *IFunc = cast<llvm::GlobalIFunc>(GetOrCreateMultiVersionResolver(GD));
-    ResolverFunc = cast<llvm::Function>(IFunc->getResolver());
-  } else
-    ResolverFunc = cast<llvm::Function>(GetOrCreateMultiVersionResolver(GD));
-
-  SmallVector<CodeGenFunction::MultiVersionResolverOption, 10> Options;
-  for (unsigned VersionIndex = 0; VersionIndex < TC->featuresStrs_size();
-       ++VersionIndex) {
-    if (!TC->isFirstOfVersion(VersionIndex))
-      continue;
-    StringRef Version = TC->getFeatureStr(VersionIndex);
-    StringRef MangledName =
-        getMangledName(GD.getWithMultiVersionIndex(VersionIndex));
-    llvm::Constant *Func = GetGlobalValue(MangledName);
-    assert(Func &&
-           "Should have already been created before calling resolver emit");
-
-    StringRef Architecture;
-    llvm::SmallVector<StringRef, 1> Feature;
-
-    if (Version.startswith("arch="))
-      Architecture = Version.drop_front(sizeof("arch=") - 1);
-    else if (Version != "default")
-      Feature.push_back(Version);
-
-    Options.emplace_back(cast<llvm::Function>(Func), Architecture, Feature);
-  }
-
-  if (supportsCOMDAT())
-    ResolverFunc->setComdat(
-        getModule().getOrInsertComdat(ResolverFunc->getName()));
-
-  const TargetInfo &TI = getTarget();
-  std::stable_sort(
-      Options.begin(), Options.end(),
-      [&TI](const CodeGenFunction::MultiVersionResolverOption &LHS,
-            const CodeGenFunction::MultiVersionResolverOption &RHS) {
-        return TargetMVPriority(TI, LHS) > TargetMVPriority(TI, RHS);
-      });
-  CodeGenFunction CGF(*this);
-  CGF.EmitMultiVersionResolver(ResolverFunc, Options);
-}
-
 void CodeGenModule::emitMultiVersionFunctions() {
   std::vector<GlobalDecl> MVFuncsToEmit;
   MultiVersionFuncs.swap(MVFuncsToEmit);
   for (GlobalDecl GD : MVFuncsToEmit) {
     const auto *FD = cast<FunctionDecl>(GD.getDecl());
     assert(FD && "Expected a FunctionDecl");
 
     SmallVector<CodeGenFunction::MultiVersionResolverOption, 10> Options;
if (FD->isTargetMultiVersion()) {
 
             const auto *TA = CurFD->getAttr<TargetAttr>();
             llvm::SmallVector<StringRef, 8> Feats;
             TA->getAddedFeatures(Feats);
 
             Options.emplace_back(cast<llvm::Function>(Func),
                                  TA->getArchitecture(), Feats);
           });
+    } else if (FD->isTargetClonesMultiVersion()) {
+      const auto *TC = FD->getAttr<TargetClonesAttr>();
+      for (unsigned VersionIndex = 0; VersionIndex < TC->featuresStrs_size();
+           ++VersionIndex) {
+        if (!TC->isFirstOfVersion(VersionIndex))
+          continue;
+        GlobalDecl CurGD{(FD->isDefined() ? FD->getDefinition() : FD),
+                         VersionIndex};
+        StringRef Version = TC->getFeatureStr(VersionIndex);
+        StringRef MangledName = getMangledName(CurGD);
+        llvm::Constant *Func = GetGlobalValue(MangledName);
+        if (!Func) {
+          if (FD->isDefined()) {
+            EmitGlobalFunctionDefinition(CurGD, nullptr);
+            Func = GetGlobalValue(MangledName);
+          } else {
+            const CGFunctionInfo &FI =
+                getTypes().arrangeGlobalDeclaration(CurGD);
+            llvm::FunctionType *Ty = getTypes().GetFunctionType(FI);
+            Func = GetAddrOfFunction(CurGD, Ty, /*ForVTable=*/false,
+                                     /*DontDefer=*/false, ForDefinition);
+          }
+          assert(Func && "This should have just been created");
+        }
+
+        StringRef Architecture;
+        llvm::SmallVector<StringRef, 1> Feature;
+
+        if (Version.startswith("arch="))
+          Architecture = Version.drop_front(sizeof("arch=") - 1);
+        else if (Version != "default")
+          Feature.push_back(Version);
+
+        Options.emplace_back(cast<llvm::Function>(Func), Architecture, Feature);
+      }
     } else {
-      assert(0 && "Expected a target multiversion function");
+      assert(0 && "Expected a target or target_clones multiversion function");
       continue;
     }
 
     llvm::Function *ResolverFunc;
-    const TargetInfo &TI = getTarget();
+    if (getTarget().supportsIFunc()) {
+      auto *IFunc =
+          cast<llvm::GlobalIFunc>(GetOrCreateMultiVersionResolver(GD));
+      ResolverFunc = cast<llvm::Function>(IFunc->getResolver());
+    } else
+      ResolverFunc = cast<llvm::Function>(GetOrCreateMultiVersionResolver(GD));
 
-    if (TI.supportsIFunc() || FD->isTargetMultiVersion()) {
-      ResolverFunc = cast<llvm::Function>(
-          GetGlobalValue((getMangledName(GD) + ".resolver").str()));
-      ResolverFunc->setLinkage(getMultiversionLinkage(*this, GD));
-    } else {
-      ResolverFunc = cast<llvm::Function>(GetGlobalValue(getMangledName(GD)));
-    }
 
     if (supportsCOMDAT())
       ResolverFunc->setComdat(
           getModule().getOrInsertComdat(ResolverFunc->getName()));
 
+    const TargetInfo &TI = getTarget();
     llvm::stable_sort(
         Options, [&TI](const CodeGenFunction::MultiVersionResolverOption &LHS,
                        const CodeGenFunction::MultiVersionResolverOption &RHS) {
           return TargetMVPriority(TI, LHS) > TargetMVPriority(TI, RHS);
         });
     CodeGenFunction CGF(*this);
     CGF.EmitMultiVersionResolver(ResolverFunc, Options);
   }
llvm::Constant *CodeGenModule::GetOrCreateMultiVersionResolver(GlobalDecl GD) {
   // Holds the name of the resolver, in ifunc mode this is the ifunc (which has
   // a separate resolver).
   std::string ResolverName = MangledName;
   if (getTarget().supportsIFunc())
     ResolverName += ".ifunc";
   else if (FD->isTargetMultiVersion())
     ResolverName += ".resolver";
 
-  // If this already exists, just return that one.
+  // If the resolver has already been created, just return it.
   if (llvm::GlobalValue *ResolverGV = GetGlobalValue(ResolverName))
     return ResolverGV;
 
   const CGFunctionInfo &FI = getTypes().arrangeGlobalDeclaration(GD);
   llvm::FunctionType *DeclTy = getTypes().GetFunctionType(FI);
 
-  // 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 if necessary (target multiversioning only).
-  if (FD->isTargetMultiVersion())
+  // The resolver needs to be created. For target and target_clones, defer
+  // creation until the end of the TU.
+  if (FD->isTargetMultiVersion() || FD->isTargetClonesMultiVersion())
     MultiVersionFuncs.push_back(GD);
-  else if (FD->isTargetClonesMultiVersion()) {
-    // In target_clones multiversioning, make sure we emit this if used.
-    auto DDI =
-        DeferredDecls.find(getMangledName(GD.getWithMultiVersionIndex(0)));
-    if (DDI != DeferredDecls.end()) {
-      addDeferredDeclToEmit(GD);
-      DeferredDecls.erase(DDI);
-    } else {
-      // Emit the symbol of the 1st variant, so that the deferred decls know we
-      // need it, otherwise the only global value will be the resolver/ifunc,
-      // which end up getting broken if we search for them with GetGlobalValue'.
-      GetOrCreateLLVMFunction(
-          getMangledName(GD.getWithMultiVersionIndex(0)), DeclTy, FD,
-          /*ForVTable=*/false, /*DontDefer=*/true,
-          /*IsThunk=*/false, llvm::AttributeList(), ForDefinition);
-    }
-  }
 
   // For cpu_specific, don't create an ifunc yet because we don't know if the
   // cpu_dispatch will be emitted in this translation unit.
   if (getTarget().supportsIFunc() && !FD->isCPUSpecificMultiVersion()) {
     llvm::Type *ResolverType = llvm::FunctionType::get(
         llvm::PointerType::get(
             DeclTy, getContext().getTargetAddressSpace(FD->getType())),
         false);

clang/test/CodeGen/attr-target-clones.c

 // RUN: %clang_cc1 -triple x86_64-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefixes=LINUX,CHECK
 // RUN: %clang_cc1 -triple x86_64-windows-pc -emit-llvm %s -o - | FileCheck %s --check-prefixes=WINDOWS,CHECK
-// XFAIL: *
 
 // LINUX: $foo.resolver = comdat any
 // LINUX: $foo_dupes.resolver = comdat any
 // LINUX: $unused.resolver = comdat any
 // LINUX: $foo_inline.resolver = comdat any
 // LINUX: $foo_inline2.resolver = comdat any
 
 // WINDOWS: $foo = comdat any
 // WINDOWS: $foo_dupes = comdat any
 // WINDOWS: $unused = comdat any
 // WINDOWS: $foo_inline = comdat any
 // WINDOWS: $foo_inline2 = comdat any
 
 // LINUX: @foo.ifunc = weak_odr ifunc i32 (), i32 ()* ()* @foo.resolver
 // LINUX: @foo_dupes.ifunc = weak_odr ifunc void (), void ()* ()* @foo_dupes.resolver
 // LINUX: @unused.ifunc = weak_odr ifunc void (), void ()* ()* @unused.resolver
 // LINUX: @foo_inline.ifunc = weak_odr ifunc i32 (), i32 ()* ()* @foo_inline.resolver
 // LINUX: @foo_inline2.ifunc = weak_odr ifunc i32 (), i32 ()* ()* @foo_inline2.resolver
 // LINUX: @foo_used_no_defn.ifunc = weak_odr ifunc i32 (), i32 ()* ()* @foo_used_no_defn.resolver
 
 int __attribute__((target_clones("sse4.2, default"))) foo(void) { return 0; }
 // LINUX: define {{.*}}i32 @foo.sse4.2.0()
 // LINUX: define {{.*}}i32 @foo.default.1()
-// LINUX: define i32 ()* @foo.resolver() comdat
+// LINUX: define weak_odr i32 ()* @foo.resolver() comdat
 // LINUX: ret i32 ()* @foo.sse4.2.0
 // LINUX: ret i32 ()* @foo.default.1
 
 // WINDOWS: define dso_local i32 @foo.sse4.2.0()
 // WINDOWS: define dso_local i32 @foo.default.1()
-// WINDOWS: define dso_local i32 @foo() comdat
+// WINDOWS: define weak_odr dso_local i32 @foo() comdat
 // WINDOWS: musttail call i32 @foo.sse4.2.0
 // WINDOWS: musttail call i32 @foo.default.1
 
 __attribute__((target_clones("default,default ,sse4.2"))) void foo_dupes(void) {}
 // LINUX: define {{.*}}void @foo_dupes.default.1()
 // LINUX: define {{.*}}void @foo_dupes.sse4.2.0()
-// LINUX: define void ()* @foo_dupes.resolver() comdat
+// LINUX: define weak_odr void ()* @foo_dupes.resolver() comdat
 // LINUX: ret void ()* @foo_dupes.sse4.2.0
 // LINUX: ret void ()* @foo_dupes.default.1
 
 // WINDOWS: define dso_local void @foo_dupes.default.1()
 // WINDOWS: define dso_local void @foo_dupes.sse4.2.0()
-// WINDOWS: define dso_local void @foo_dupes() comdat
+// WINDOWS: define weak_odr dso_local void @foo_dupes() comdat
 // WINDOWS: musttail call void @foo_dupes.sse4.2.0
 // WINDOWS: musttail call void @foo_dupes.default.1
 
 void bar2(void) {
   // LINUX: define {{.*}}void @bar2()
   // WINDOWS: define dso_local void @bar2()
   foo_dupes();
   // LINUX: call void @foo_dupes.ifunc()
   // WINDOWS: call void @foo_dupes()
 }
 
 int bar(void) {
   // LINUX: define {{.*}}i32 @bar() #[[DEF:[0-9]+]]
   // WINDOWS: define dso_local i32 @bar() #[[DEF:[0-9]+]]
   return foo();
   // LINUX: call i32 @foo.ifunc()
   // WINDOWS: call i32 @foo()
 }
 
 void __attribute__((target_clones("default, arch=ivybridge"))) unused(void) {}
 // LINUX: define {{.*}}void @unused.default.1()
 // LINUX: define {{.*}}void @unused.arch_ivybridge.0()
-// LINUX: define void ()* @unused.resolver() comdat
+// LINUX: define weak_odr void ()* @unused.resolver() comdat
 // LINUX: ret void ()* @unused.arch_ivybridge.0
 // LINUX: ret void ()* @unused.default.1
 
 // WINDOWS: define dso_local void @unused.default.1()
 // WINDOWS: define dso_local void @unused.arch_ivybridge.0()
-// WINDOWS: define dso_local void @unused() comdat
+// WINDOWS: define weak_odr dso_local void @unused() comdat
 // WINDOWS: musttail call void @unused.arch_ivybridge.0
 // WINDOWS: musttail call void @unused.default.1
 
 
 inline int __attribute__((target_clones("arch=sandybridge,default,sse4.2")))
 foo_inline(void) { return 0; }
 inline int __attribute__((target_clones("arch=sandybridge,default,sse4.2")))
 foo_inline2(void);
 
 int bar3(void) {
   // LINUX: define {{.*}}i32 @bar3()
   // WINDOWS: define dso_local i32 @bar3()
   return foo_inline() + foo_inline2();
   // LINUX: call i32 @foo_inline.ifunc()
   // LINUX: call i32 @foo_inline2.ifunc()
   // WINDOWS: call i32 @foo_inline()
   // WINDOWS: call i32 @foo_inline2()
 }
 
-// Deferred emission of foo_inline, which got delayed because it is inline.
-// LINUX: define i32 ()* @foo_inline.resolver() comdat
+// LINUX: define weak_odr i32 ()* @foo_inline.resolver() comdat
 // LINUX: ret i32 ()* @foo_inline.arch_sandybridge.0
 // LINUX: ret i32 ()* @foo_inline.sse4.2.1
 // LINUX: ret i32 ()* @foo_inline.default.2
 
-// WINDOWS: define dso_local i32 @foo_inline() comdat
+// WINDOWS: define weak_odr dso_local i32 @foo_inline() comdat
 // WINDOWS: musttail call i32 @foo_inline.arch_sandybridge.0
 // WINDOWS: musttail call i32 @foo_inline.sse4.2.1
 // WINDOWS: musttail call i32 @foo_inline.default.2
 
 inline int __attribute__((target_clones("arch=sandybridge,default,sse4.2")))
 foo_inline2(void){ return 0; }
-// LINUX: define linkonce i32 @foo_inline2.arch_sandybridge.0() #[[SB:[0-9]+]]
-// LINUX: define i32 ()* @foo_inline2.resolver() comdat
+// LINUX: define weak_odr i32 ()* @foo_inline2.resolver() comdat
 // LINUX: ret i32 ()* @foo_inline2.arch_sandybridge.0
 // LINUX: ret i32 ()* @foo_inline2.sse4.2.1
 // LINUX: ret i32 ()* @foo_inline2.default.2
 
-// WINDOWS: define linkonce_odr dso_local i32 @foo_inline2.arch_sandybridge.0() #[[SB:[0-9]+]]
-// WINDOWS: define dso_local i32 @foo_inline2() comdat
+// WINDOWS: define weak_odr dso_local i32 @foo_inline2() comdat
 // WINDOWS: musttail call i32 @foo_inline2.arch_sandybridge.0
 // WINDOWS: musttail call i32 @foo_inline2.sse4.2.1
 // WINDOWS: musttail call i32 @foo_inline2.default.2
 
 
 int __attribute__((target_clones("default", "sse4.2")))
 foo_unused_no_defn(void);
 
 int __attribute__((target_clones("default", "sse4.2")))
 foo_used_no_defn(void);
 
 int test_foo_used_no_defn(void) {
   // LINUX: define {{.*}}i32 @test_foo_used_no_defn()
   // WINDOWS: define dso_local i32 @test_foo_used_no_defn()
   return foo_used_no_defn();
   // LINUX: call i32 @foo_used_no_defn.ifunc()
   // WINDOWS: call i32 @foo_used_no_defn()
 }
-// LINUX: declare i32 @foo_used_no_defn.default.1()
-// LINUX: declare i32 ()* @foo_used_no_defn.resolver()
 
-// WINDOWS: declare dso_local i32 @foo_used_no_defn.default.1()
-// WINDOWS: declare dso_local i32 @foo_used_no_defn()
+
+// LINUX: define weak_odr i32 ()* @foo_used_no_defn.resolver() comdat
+// LINUX: ret i32 ()* @foo_used_no_defn.sse4.2.0
+// LINUX: ret i32 ()* @foo_used_no_defn.default.1
+
+// WINDOWS: define weak_odr dso_local i32 @foo_used_no_defn() comdat
+// WINDOWS: musttail call i32 @foo_used_no_defn.sse4.2.0
+// WINDOWS: musttail call i32 @foo_used_no_defn.default.1
 
 
 // Deferred emission of inline definitions.
-// LINUX: define linkonce i32 @foo_inline.arch_sandybridge.0() #[[SB]]
-// LINUX: define linkonce i32 @foo_inline.default.2() #[[DEF]]
+
+// LINUX: define linkonce i32 @foo_inline.arch_sandybridge.0() #[[SB:[0-9]+]]
+// LINUX: define linkonce i32 @foo_inline.default.2() #[[DEF:[0-9]+]]
 // LINUX: define linkonce i32 @foo_inline.sse4.2.1() #[[SSE42:[0-9]+]]
 
-// WINDOWS: define linkonce_odr dso_local i32 @foo_inline.arch_sandybridge.0() #[[SB]]
+// WINDOWS: define linkonce_odr dso_local i32 @foo_inline.arch_sandybridge.0() #[[SB:[0-9]+]]
 // WINDOWS: define linkonce_odr dso_local i32 @foo_inline.default.2() #[[DEF]]
 // WINDOWS: define linkonce_odr dso_local i32 @foo_inline.sse4.2.1() #[[SSE42:[0-9]+]]
 
 
+// LINUX: define linkonce i32 @foo_inline2.arch_sandybridge.0() #[[SB]]
 // LINUX: define linkonce i32 @foo_inline2.default.2() #[[DEF]]
 // LINUX: define linkonce i32 @foo_inline2.sse4.2.1() #[[SSE42]]
 
+// WINDOWS: define linkonce_odr dso_local i32 @foo_inline2.arch_sandybridge.0() #[[SB]]
 // WINDOWS: define linkonce_odr dso_local i32 @foo_inline2.default.2() #[[DEF]]
 // WINDOWS: define linkonce_odr dso_local i32 @foo_inline2.sse4.2.1() #[[SSE42]]
 
 
+// LINUX: declare i32 @foo_used_no_defn.default.1()
+// LINUX: declare i32 @foo_used_no_defn.sse4.2.0()
+
+// WINDOWS: declare dso_local i32 @foo_used_no_defn.default.1()
+// WINDOWS: declare dso_local i32 @foo_used_no_defn.sse4.2.0()
+
+
 // CHECK: attributes #[[SSE42]] =
 // CHECK-SAME: "target-features"="+crc32,+cx8,+mmx,+popcnt,+sse,+sse2,+sse3,+sse4.1,+sse4.2,+ssse3,+x87"
 // CHECK: attributes #[[DEF]] =
 // Don't bother checking features, we verified it is the same as a normal function.
 // CHECK: attributes #[[SB]] =
 // CHECK-SAME: "target-features"="+avx,+crc32,+cx16,+cx8,+fxsr,+mmx,+pclmul,+popcnt,+sahf,+sse,+sse2,+sse3,+sse4.1,+sse4.2,+ssse3,+x87,+xsave,+xsaveopt"

clang/test/CodeGenCXX/attr-target-clones.cpp

 // RUN: %clang_cc1 -std=c++11 -triple x86_64-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix=LINUX
 // RUN: %clang_cc1 -std=c++11 -triple x86_64-windows-pc -emit-llvm %s -o - | FileCheck %s --check-prefix=WINDOWS
 
 // Overloaded ifuncs
 // LINUX: @_Z10overloadedi.ifunc = weak_odr ifunc i32 (i32), i32 (i32)* ()* @_Z10overloadedi.resolver
 // LINUX: @_Z10overloadedPKc.ifunc = weak_odr ifunc i32 (i8*), i32 (i8*)* ()* @_Z10overloadedPKc.resolver
 // struct 'C' ifuncs, note the 'float, U' one doesn't get one.
 // LINUX: @_ZN1CIssE3fooEv.ifunc = weak_odr ifunc i32 (%struct.C*), i32 (%struct.C*)* ()* @_ZN1CIssE3fooEv.resolver
 // LINUX: @_ZN1CIisE3fooEv.ifunc = weak_odr ifunc i32 (%struct.C.0*), i32 (%struct.C.0*)* ()* @_ZN1CIisE3fooEv.resolver
 // LINUX: @_ZN1CIdfE3fooEv.ifunc = weak_odr ifunc i32 (%struct.C.2*), i32 (%struct.C.2*)* ()* @_ZN1CIdfE3fooEv.resolver
 
 int __attribute__((target_clones("sse4.2", "default"))) overloaded(int) { return 1; }
 // LINUX: define {{.*}}i32 @_Z10overloadedi.sse4.2.0(i32{{.+}})
 // LINUX: define {{.*}}i32 @_Z10overloadedi.default.1(i32{{.+}})
-// LINUX: define i32 (i32)* @_Z10overloadedi.resolver
+// LINUX: define weak_odr i32 (i32)* @_Z10overloadedi.resolver() comdat
 // LINUX: ret i32 (i32)* @_Z10overloadedi.sse4.2.0
 // LINUX: ret i32 (i32)* @_Z10overloadedi.default.1
 
 // WINDOWS: define dso_local noundef i32 @"?overloaded@@[e-mail]4.2.0"(i32{{.+}})
 // WINDOWS: define dso_local noundef i32 @"?overloaded@@[e-mail].1"(i32{{.+}})
-// WINDOWS: define dso_local i32 @"?overloaded@@YAHH@Z"(i32{{.+}})
+// WINDOWS: define weak_odr dso_local i32 @"?overloaded@@YAHH@Z"(i32{{.+}}) comdat
 // WINDOWS: call i32 @"?overloaded@@[e-mail]4.2.0"
 // WINDOWS: call i32 @"?overloaded@@[e-mail].1"
 
 int __attribute__((target_clones("arch=ivybridge", "default"))) overloaded(const char *) { return 2; }
 // LINUX: define {{.*}}i32 @_Z10overloadedPKc.arch_ivybridge.0(i8*{{.+}})
 // LINUX: define {{.*}}i32 @_Z10overloadedPKc.default.1(i8*{{.+}})
-// LINUX: define i32 (i8*)* @_Z10overloadedPKc.resolver
+// LINUX: define weak_odr i32 (i8*)* @_Z10overloadedPKc.resolver() comdat
 // LINUX: ret i32 (i8*)* @_Z10overloadedPKc.arch_ivybridge.0
 // LINUX: ret i32 (i8*)* @_Z10overloadedPKc.default.1
 
 // WINDOWS: define dso_local noundef i32 @"?overloaded@@[e-mail]_ivybridge.0"(i8*{{.+}})
 // WINDOWS: define dso_local noundef i32 @"?overloaded@@[e-mail].1"(i8*{{.+}})
-// WINDOWS: define dso_local i32 @"?overloaded@@YAHPEBD@Z"(i8*{{.+}})
+// WINDOWS: define weak_odr dso_local i32 @"?overloaded@@YAHPEBD@Z"(i8*{{.+}}) comdat
 // WINDOWS: call i32 @"?overloaded@@[e-mail]_ivybridge.0"
 // WINDOWS: call i32 @"?overloaded@@[e-mail].1"
-//
+
 void use_overloaded() {
   overloaded(1);
   // LINUX: call noundef i32 @_Z10overloadedi.ifunc
   // WINDOWS: call noundef i32 @"?overloaded@@YAHH@Z"
   overloaded(nullptr);
   // LINUX: call noundef i32 @_Z10overloadedPKc.ifunc 
   // WINDOWS: call noundef i32 @"?overloaded@@YAHPEBD@Z"
 }
void uses_specialized() {
   // LINUX: call noundef i32 @_ZN1CIfsE3fooEv(%struct.C
   // WINDOWS: call noundef i32 @"?foo@?$C@MF@@QEAAHXZ"(%struct.C
   C<double, float> c4;
   c4.foo();
   // LINUX: call noundef i32 @_ZN1CIdfE3fooEv.ifunc(%struct.C
   // WINDOWS: call noundef i32 @"?foo@?$C@NM@@QEAAHXZ"(%struct.C
 }
 
-// LINUX: define {{.*}}i32 @_ZN1CIssE3fooEv.sse4.2.0(%struct.C{{.+}})
-// WINDOWS: define {{.*}}i32 @"?foo@?$C@FF@@QEAAHXZ.sse4.2.0"(%struct.C{{.+}})
-// LINUX: define i32 (%struct.C*)* @_ZN1CIssE3fooEv.resolver
+// LINUX: define weak_odr i32 (%struct.C*)* @_ZN1CIssE3fooEv.resolver() comdat
 // LINUX: ret i32 (%struct.C*)* @_ZN1CIssE3fooEv.sse4.2.0
 // LINUX: ret i32 (%struct.C*)* @_ZN1CIssE3fooEv.default.1
+
 // WINDOWS: define {{.*}}i32 @"?foo@?$C@FF@@QEAAHXZ"(%struct.C{{.+}})
 // WINDOWS: call i32 @"?foo@?$C@FF@@QEAAHXZ.sse4.2.0"
 // WINDOWS: call i32 @"?foo@?$C@FF@@QEAAHXZ.default.1"
 
-// LINUX: define {{.*}}i32 @_ZN1CIisE3fooEv.sse4.2.0(%struct.C{{.+}})
-// WINDOWS: define {{.*}}i32 @"?foo@?$C@HF@@QEAAHXZ.sse4.2.0"(%struct.C{{.+}})
-// LINUX: define i32 (%struct.C{{.+}})* @_ZN1CIisE3fooEv.resolver
+// LINUX: define weak_odr i32 (%struct.C{{.+}})* @_ZN1CIisE3fooEv.resolver() comdat
 // LINUX: ret i32 (%struct.C{{.+}})* @_ZN1CIisE3fooEv.sse4.2.0
 // LINUX: ret i32 (%struct.C{{.+}})* @_ZN1CIisE3fooEv.default.1
+
 // WINDOWS: define {{.*}}i32 @"?foo@?$C@HF@@QEAAHXZ"(%struct.C{{.+}})
 // WINDOWS: call i32 @"?foo@?$C@HF@@QEAAHXZ.sse4.2.0"
 // WINDOWS: call i32 @"?foo@?$C@HF@@QEAAHXZ.default.1"
 
-// LINUX: define i32 (%struct.C{{.+}})* @_ZN1CIdfE3fooEv.resolver
+// LINUX: define weak_odr i32 (%struct.C{{.+}})* @_ZN1CIdfE3fooEv.resolver() comdat
 // LINUX: ret i32 (%struct.C{{.+}})* @_ZN1CIdfE3fooEv.sse4.2.0
 // LINUX: ret i32 (%struct.C{{.+}})* @_ZN1CIdfE3fooEv.default.1
+
 // WINDOWS: define {{.*}}i32 @"?foo@?$C@NM@@QEAAHXZ"(%struct.C{{.+}})
 // WINDOWS: call i32 @"?foo@?$C@NM@@QEAAHXZ.sse4.2.0"
 // WINDOWS: call i32 @"?foo@?$C@NM@@QEAAHXZ.default.1"
 
+// LINUX: define {{.*}}i32 @_ZN1CIssE3fooEv.sse4.2.0(%struct.C{{.+}})
+// LINUX: define {{.*}}i32 @_ZN1CIssE3fooEv.default.1(%struct.C{{.+}})
+// LINUX: define {{.*}}i32 @_ZN1CIisE3fooEv.sse4.2.0(%struct.C{{.+}})
+// LINUX: define {{.*}}i32 @_ZN1CIisE3fooEv.default.1(%struct.C{{.+}})
 // LINUX: define {{.*}}i32 @_ZN1CIdfE3fooEv.sse4.2.0(%struct.C{{.+}})
-// WINDOWS: define {{.*}}i32 @"?foo@?$C@NM@@QEAAHXZ.sse4.2.0"(%struct.C{{.+}})
 // LINUX: define {{.*}}i32 @_ZN1CIdfE3fooEv.default.1(%struct.C{{.+}})
-// WINDOWS: define {{.*}}i32 @"?foo@?$C@NM@@QEAAHXZ.default.1"(%struct.C{{.+}})
-// LINUX: define {{.*}}i32 @_ZN1CIssE3fooEv.default.1(%struct.C{{.+}})
+
+// WINDOWS: define {{.*}}i32 @"?foo@?$C@FF@@QEAAHXZ.sse4.2.0"(%struct.C{{.+}})
 // WINDOWS: define {{.*}}i32 @"?foo@?$C@FF@@QEAAHXZ.default.1"(%struct.C{{.+}})
-// LINUX: define {{.*}}i32 @_ZN1CIisE3fooEv.default.1(%struct.C{{.+}})
+// WINDOWS: define {{.*}}i32 @"?foo@?$C@HF@@QEAAHXZ.sse4.2.0"(%struct.C{{.+}})
 // WINDOWS: define {{.*}}i32 @"?foo@?$C@HF@@QEAAHXZ.default.1"(%struct.C{{.+}})
+// WINDOWS: define {{.*}}i32 @"?foo@?$C@NM@@QEAAHXZ.sse4.2.0"(%struct.C{{.+}})
+// WINDOWS: define {{.*}}i32 @"?foo@?$C@NM@@QEAAHXZ.default.1"(%struct.C{{.+}})