1 files changed, 59 insertions, 17 deletions

diff --git a/clang/lib/Sema/SemaTemplateInstantiateDecl.cpp b/clang/lib/Sema/SemaTemplateInstantiateDecl.cpp
index 468bc1d..28925cc 100644
--- a/clang/lib/Sema/SemaTemplateInstantiateDecl.cpp
+++ b/clang/lib/Sema/SemaTemplateInstantiateDecl.cpp
@@ -707,6 +707,23 @@ static void instantiateDependentAMDGPUMaxNumWorkGroupsAttr(
     S.AMDGPU().addAMDGPUMaxNumWorkGroupsAttr(New, Attr, XExpr, YExpr, ZExpr);
 }
 
+static void instantiateDependentCUDAClusterDimsAttr(
+    Sema &S, const MultiLevelTemplateArgumentList &TemplateArgs,
+    const CUDAClusterDimsAttr &Attr, Decl *New) {
+  EnterExpressionEvaluationContext Unevaluated(
+      S, Sema::ExpressionEvaluationContext::ConstantEvaluated);
+
+  auto SubstElt = [&S, &TemplateArgs](Expr *E) {
+    return E ? S.SubstExpr(E, TemplateArgs).get() : nullptr;
+  };
+
+  Expr *XExpr = SubstElt(Attr.getX());
+  Expr *YExpr = SubstElt(Attr.getY());
+  Expr *ZExpr = SubstElt(Attr.getZ());
+
+  S.addClusterDimsAttr(New, Attr, XExpr, YExpr, ZExpr);
+}
+
 // This doesn't take any template parameters, but we have a custom action that
 // needs to happen when the kernel itself is instantiated. We need to run the
 // ItaniumMangler to mark the names required to name this kernel.
@@ -765,10 +782,18 @@ static bool isRelevantAttr(Sema &S, const Decl *D, const Attr *A) {
 
 static void instantiateDependentHLSLParamModifierAttr(
     Sema &S, const MultiLevelTemplateArgumentList &TemplateArgs,
-    const HLSLParamModifierAttr *Attr, Decl *New) {
-  ParmVarDecl *P = cast<ParmVarDecl>(New);
-  P->addAttr(Attr->clone(S.getASTContext()));
-  P->setType(S.HLSL().getInoutParameterType(P->getType()));
+    const HLSLParamModifierAttr *Attr, const Decl *Old, Decl *New) {
+  ParmVarDecl *NewParm = cast<ParmVarDecl>(New);
+  NewParm->addAttr(Attr->clone(S.getASTContext()));
+
+  const Type *OldParmTy = cast<ParmVarDecl>(Old)->getType().getTypePtr();
+  if (OldParmTy->isDependentType() && Attr->isAnyOut())
+    NewParm->setType(S.HLSL().getInoutParameterType(NewParm->getType()));
+
+  assert(
+      (!Attr->isAnyOut() || (NewParm->getType().isRestrictQualified() &&
+                             NewParm->getType()->isReferenceType())) &&
+      "out or inout parameter type must be a reference and restrict qualified");
 }
 
 void Sema::InstantiateAttrsForDecl(
@@ -921,9 +946,14 @@ void Sema::InstantiateAttrs(const MultiLevelTemplateArgumentList &TemplateArgs,
           *this, TemplateArgs, *AMDGPUMaxNumWorkGroups, New);
     }
 
+    if (const auto *CUDAClusterDims = dyn_cast<CUDAClusterDimsAttr>(TmplAttr)) {
+      instantiateDependentCUDAClusterDimsAttr(*this, TemplateArgs,
+                                              *CUDAClusterDims, New);
+    }
+
     if (const auto *ParamAttr = dyn_cast<HLSLParamModifierAttr>(TmplAttr)) {
       instantiateDependentHLSLParamModifierAttr(*this, TemplateArgs, ParamAttr,
-                                                New);
+                                                Tmpl, New);
       continue;
     }
 
@@ -5282,6 +5312,16 @@ void Sema::InstantiateExceptionSpec(SourceLocation PointOfInstantiation,
   if (Proto->getExceptionSpecType() != EST_Uninstantiated)
     return;
 
+  RecursiveInstGuard AlreadyInstantiating(
+      *this, Decl, RecursiveInstGuard::Kind::ExceptionSpec);
+  if (AlreadyInstantiating) {
+    // This exception specification indirectly depends on itself. Reject.
+    // FIXME: Corresponding rule in the standard?
+    Diag(PointOfInstantiation, diag::err_exception_spec_cycle) << Decl;
+    UpdateExceptionSpec(Decl, EST_None);
+    return;
+  }
+
   InstantiatingTemplate Inst(*this, PointOfInstantiation, Decl,
                              InstantiatingTemplate::ExceptionSpecification());
   if (Inst.isInvalid()) {
@@ -5290,13 +5330,6 @@ void Sema::InstantiateExceptionSpec(SourceLocation PointOfInstantiation,
     UpdateExceptionSpec(Decl, EST_None);
     return;
   }
-  if (Inst.isAlreadyInstantiating()) {
-    // This exception specification indirectly depends on itself. Reject.
-    // FIXME: Corresponding rule in the standard?
-    Diag(PointOfInstantiation, diag::err_exception_spec_cycle) << Decl;
-    UpdateExceptionSpec(Decl, EST_None);
-    return;
-  }
 
   // Enter the scope of this instantiation. We don't use
   // PushDeclContext because we don't have a scope.
@@ -5356,8 +5389,6 @@ TemplateDeclInstantiator::InitFunctionInstantiation(FunctionDecl *New,
   if (ActiveInst.Kind == ActiveInstType::ExplicitTemplateArgumentSubstitution ||
       ActiveInst.Kind == ActiveInstType::DeducedTemplateArgumentSubstitution) {
     if (isa<FunctionTemplateDecl>(ActiveInst.Entity)) {
-      SemaRef.InstantiatingSpecializations.erase(
-          {ActiveInst.Entity->getCanonicalDecl(), ActiveInst.Kind});
       atTemplateEnd(SemaRef.TemplateInstCallbacks, SemaRef, ActiveInst);
       ActiveInst.Kind = ActiveInstType::TemplateInstantiation;
       ActiveInst.Entity = New;
@@ -5515,6 +5546,12 @@ void Sema::InstantiateFunctionDefinition(SourceLocation PointOfInstantiation,
     Function = const_cast<FunctionDecl*>(ExistingDefn);
   }
 
+#ifndef NDEBUG
+  RecursiveInstGuard AlreadyInstantiating(*this, Function,
+                                          RecursiveInstGuard::Kind::Template);
+  assert(!AlreadyInstantiating && "should have been caught by caller");
+#endif
+
   // Find the function body that we'll be substituting.
   const FunctionDecl *PatternDecl = Function->getTemplateInstantiationPattern();
   assert(PatternDecl && "instantiating a non-template");
@@ -5654,7 +5691,7 @@ void Sema::InstantiateFunctionDefinition(SourceLocation PointOfInstantiation,
   }
 
   InstantiatingTemplate Inst(*this, PointOfInstantiation, Function);
-  if (Inst.isInvalid() || Inst.isAlreadyInstantiating())
+  if (Inst.isInvalid())
     return;
   PrettyDeclStackTraceEntry CrashInfo(Context, Function, SourceLocation(),
                                       "instantiating function definition");
@@ -6223,6 +6260,11 @@ void Sema::InstantiateVariableDefinition(SourceLocation PointOfInstantiation,
   if (TSK == TSK_ExplicitSpecialization)
     return;
 
+  RecursiveInstGuard AlreadyInstantiating(*this, Var,
+                                          RecursiveInstGuard::Kind::Template);
+  if (AlreadyInstantiating)
+    return;
+
   // Find the pattern and the arguments to substitute into it.
   VarDecl *PatternDecl = Var->getTemplateInstantiationPattern();
   assert(PatternDecl && "no pattern for templated variable");
@@ -6246,7 +6288,7 @@ void Sema::InstantiateVariableDefinition(SourceLocation PointOfInstantiation,
       // FIXME: Factor out the duplicated instantiation context setup/tear down
       // code here.
       InstantiatingTemplate Inst(*this, PointOfInstantiation, Var);
-      if (Inst.isInvalid() || Inst.isAlreadyInstantiating())
+      if (Inst.isInvalid())
         return;
       PrettyDeclStackTraceEntry CrashInfo(Context, Var, SourceLocation(),
                                           "instantiating variable initializer");
@@ -6350,7 +6392,7 @@ void Sema::InstantiateVariableDefinition(SourceLocation PointOfInstantiation,
   }
 
   InstantiatingTemplate Inst(*this, PointOfInstantiation, Var);
-  if (Inst.isInvalid() || Inst.isAlreadyInstantiating())
+  if (Inst.isInvalid())
     return;
   PrettyDeclStackTraceEntry CrashInfo(Context, Var, SourceLocation(),
                                       "instantiating variable definition");