Reland "[CUDA][HIP] Fix overloading resolution in global var init" (llvm#65606)

Cherry-pick 9b77638

https://reviews.llvm.org/D158247 caused regressions for HIP on Windows
and was reverted.

A reduced test case is:

```
typedef void (__stdcall* funcTy)();
void invoke(funcTy f);

static void __stdcall callee() noexcept {
}

void foo() {
   invoke(callee);
}
```

It is due to clang missing handling host/device attributes for calling
convention at a few places

This patch fixes that.

Change-Id: Ibbc5cbe232d73ddaeb91f13f6afbff3151c9bf0b

Author: yxsamliu

.gitignore

@@ -70,3 +70,4 @@ pythonenv*
 /clang/utils/analyzer/projects/*/RefScanBuildResults
 # automodapi puts generated documentation files here.
 /lldb/docs/python_api/
+/Debug/

clang/include/clang/Sema/Sema.h

@@ -994,6 +994,14 @@ class Sema final {
     }
   } DelayedDiagnostics;
 
+  enum CUDAFunctionTarget {
+    CFT_Device,
+    CFT_Global,
+    CFT_Host,
+    CFT_HostDevice,
+    CFT_InvalidTarget
+  };
+
   /// A RAII object to temporarily push a declaration context.
   class ContextRAII {
   private:
@@ -4696,8 +4704,13 @@ class Sema final {
   bool isValidPointerAttrType(QualType T, bool RefOkay = false);
 
   bool CheckRegparmAttr(const ParsedAttr &attr, unsigned &value);
+
+  /// Check validaty of calling convention attribute \p attr. If \p FD
+  /// is not null pointer, use \p FD to determine the CUDA/HIP host/device
+  /// target. Otherwise, it is specified by \p CFT.
   bool CheckCallingConvAttr(const ParsedAttr &attr, CallingConv &CC,
-                            const FunctionDecl *FD = nullptr);
+                            const FunctionDecl *FD = nullptr,
+                            CUDAFunctionTarget CFT = CFT_InvalidTarget);
   bool CheckAttrTarget(const ParsedAttr &CurrAttr);
   bool CheckAttrNoArgs(const ParsedAttr &CurrAttr);
   bool checkStringLiteralArgumentAttr(const AttributeCommonInfo &CI,
@@ -13094,14 +13107,6 @@ class Sema final {
   void checkTypeSupport(QualType Ty, SourceLocation Loc,
                         ValueDecl *D = nullptr);
 
-  enum CUDAFunctionTarget {
-    CFT_Device,
-    CFT_Global,
-    CFT_Host,
-    CFT_HostDevice,
-    CFT_InvalidTarget
-  };
-
   /// Determines whether the given function is a CUDA device/host/kernel/etc.
   /// function.
   ///
@@ -13120,6 +13125,29 @@ class Sema final {
   /// Determines whether the given variable is emitted on host or device side.
   CUDAVariableTarget IdentifyCUDATarget(const VarDecl *D);
 
+  /// Defines kinds of CUDA global host/device context where a function may be
+  /// called.
+  enum CUDATargetContextKind {
+    CTCK_Unknown,       /// Unknown context
+    CTCK_InitGlobalVar, /// Function called during global variable
+                        /// initialization
+  };
+
+  /// Define the current global CUDA host/device context where a function may be
+  /// called. Only used when a function is called outside of any functions.
+  struct CUDATargetContext {
+    CUDAFunctionTarget Target = CFT_HostDevice;
+    CUDATargetContextKind Kind = CTCK_Unknown;
+    Decl *D = nullptr;
+  } CurCUDATargetCtx;
+
+  struct CUDATargetContextRAII {
+    Sema &S;
+    CUDATargetContext SavedCtx;
+    CUDATargetContextRAII(Sema &S_, CUDATargetContextKind K, Decl *D);
+    ~CUDATargetContextRAII() { S.CurCUDATargetCtx = SavedCtx; }
+  };
+
   /// Gets the CUDA target for the current context.
   CUDAFunctionTarget CurrentCUDATarget() {
     return IdentifyCUDATarget(dyn_cast<FunctionDecl>(CurContext));

clang/lib/Parse/ParseDecl.cpp

@@ -2479,6 +2479,7 @@ Decl *Parser::ParseDeclarationAfterDeclaratorAndAttributes(
     }
   }
 
+  Sema::CUDATargetContextRAII X(Actions, Sema::CTCK_InitGlobalVar, ThisDecl);
   switch (TheInitKind) {
   // Parse declarator '=' initializer.
   case InitKind::Equal: {

clang/lib/Sema/SemaCUDA.cpp

@@ -105,19 +105,37 @@ Sema::IdentifyCUDATarget(const ParsedAttributesView &Attrs) {
 }
 
 template <typename A>
-static bool hasAttr(const FunctionDecl *D, bool IgnoreImplicitAttr) {
+static bool hasAttr(const Decl *D, bool IgnoreImplicitAttr) {
   return D->hasAttrs() && llvm::any_of(D->getAttrs(), [&](Attr *Attribute) {
            return isa<A>(Attribute) &&
                   !(IgnoreImplicitAttr && Attribute->isImplicit());
          });
 }
 
+Sema::CUDATargetContextRAII::CUDATargetContextRAII(Sema &S_,
+                                                   CUDATargetContextKind K,
+                                                   Decl *D)
+    : S(S_) {
+  SavedCtx = S.CurCUDATargetCtx;
+  assert(K == CTCK_InitGlobalVar);
+  auto *VD = dyn_cast_or_null<VarDecl>(D);
+  if (VD && VD->hasGlobalStorage() && !VD->isStaticLocal()) {
+    auto Target = CFT_Host;
+    if ((hasAttr<CUDADeviceAttr>(VD, /*IgnoreImplicit=*/true) &&
+         !hasAttr<CUDAHostAttr>(VD, /*IgnoreImplicit=*/true)) ||
+        hasAttr<CUDASharedAttr>(VD, /*IgnoreImplicit=*/true) ||
+        hasAttr<CUDAConstantAttr>(VD, /*IgnoreImplicit=*/true))
+      Target = CFT_Device;
+    S.CurCUDATargetCtx = {Target, K, VD};
+  }
+}
+
 /// IdentifyCUDATarget - Determine the CUDA compilation target for this function
 Sema::CUDAFunctionTarget Sema::IdentifyCUDATarget(const FunctionDecl *D,
                                                   bool IgnoreImplicitHDAttr) {
-  // Code that lives outside a function is run on the host.
+  // Code that lives outside a function gets the target from CurCUDATargetCtx.
   if (D == nullptr)
-    return CFT_Host;
+    return CurCUDATargetCtx.Target;
 
   if (D->hasAttr<CUDAInvalidTargetAttr>())
     return CFT_InvalidTarget;

clang/lib/Sema/SemaDeclAttr.cpp

@@ -5030,7 +5030,8 @@ static void handleCallConvAttr(Sema &S, Decl *D, const ParsedAttr &AL) {
   // Diagnostic is emitted elsewhere: here we store the (valid) AL
   // in the Decl node for syntactic reasoning, e.g., pretty-printing.
   CallingConv CC;
-  if (S.CheckCallingConvAttr(AL, CC, /*FD*/nullptr))
+  if (S.CheckCallingConvAttr(AL, CC, /*FD*/ nullptr,
+                             S.IdentifyCUDATarget(dyn_cast<FunctionDecl>(D))))
     return;
 
   if (!isa<ObjCMethodDecl>(D)) {
@@ -5211,7 +5212,8 @@ static void handleNoRandomizeLayoutAttr(Sema &S, Decl *D,
 }
 
 bool Sema::CheckCallingConvAttr(const ParsedAttr &Attrs, CallingConv &CC,
-                                const FunctionDecl *FD) {
+                                const FunctionDecl *FD,
+                                CUDAFunctionTarget CFT) {
   if (Attrs.isInvalid())
     return true;
 
@@ -5310,7 +5312,8 @@ bool Sema::CheckCallingConvAttr(const ParsedAttr &Attrs, CallingConv &CC,
   // on their host/device attributes.
   if (LangOpts.CUDA) {
     auto *Aux = Context.getAuxTargetInfo();
-    auto CudaTarget = IdentifyCUDATarget(FD);
+    assert(FD || CFT != CFT_InvalidTarget);
+    auto CudaTarget = FD ? IdentifyCUDATarget(FD) : CFT;
     bool CheckHost = false, CheckDevice = false;
     switch (CudaTarget) {
     case CFT_HostDevice:

clang/lib/Sema/SemaOverload.cpp

@@ -6608,17 +6608,19 @@ void Sema::AddOverloadCandidate(
   }
 
   // (CUDA B.1): Check for invalid calls between targets.
-  if (getLangOpts().CUDA)
-    if (const FunctionDecl *Caller = getCurFunctionDecl(/*AllowLambda=*/true))
-      // Skip the check for callers that are implicit members, because in this
-      // case we may not yet know what the member's target is; the target is
-      // inferred for the member automatically, based on the bases and fields of
-      // the class.
-      if (!Caller->isImplicit() && !IsAllowedCUDACall(Caller, Function)) {
-        Candidate.Viable = false;
-        Candidate.FailureKind = ovl_fail_bad_target;
-        return;
-      }
+  if (getLangOpts().CUDA) {
+    const FunctionDecl *Caller = getCurFunctionDecl(/*AllowLambda=*/true);
+    // Skip the check for callers that are implicit members, because in this
+    // case we may not yet know what the member's target is; the target is
+    // inferred for the member automatically, based on the bases and fields of
+    // the class.
+    if (!(Caller && Caller->isImplicit()) &&
+        !IsAllowedCUDACall(Caller, Function)) {
+      Candidate.Viable = false;
+      Candidate.FailureKind = ovl_fail_bad_target;
+      return;
+    }
+  }
 
   if (Function->getTrailingRequiresClause()) {
     ConstraintSatisfaction Satisfaction;
@@ -7130,12 +7132,11 @@ Sema::AddMethodCandidate(CXXMethodDecl *Method, DeclAccessPair FoundDecl,
 
   // (CUDA B.1): Check for invalid calls between targets.
   if (getLangOpts().CUDA)
-    if (const FunctionDecl *Caller = getCurFunctionDecl(/*AllowLambda=*/true))
-      if (!IsAllowedCUDACall(Caller, Method)) {
-        Candidate.Viable = false;
-        Candidate.FailureKind = ovl_fail_bad_target;
-        return;
-      }
+    if (!IsAllowedCUDACall(getCurFunctionDecl(/*AllowLambda=*/true), Method)) {
+      Candidate.Viable = false;
+      Candidate.FailureKind = ovl_fail_bad_target;
+      return;
+    }
 
   if (Method->getTrailingRequiresClause()) {
     ConstraintSatisfaction Satisfaction;
@@ -12383,10 +12384,12 @@ class AddressOfFunctionResolver {
       return false;
 
     if (FunctionDecl *FunDecl = dyn_cast<FunctionDecl>(Fn)) {
-      if (S.getLangOpts().CUDA)
-        if (FunctionDecl *Caller = S.getCurFunctionDecl(/*AllowLambda=*/true))
-          if (!Caller->isImplicit() && !S.IsAllowedCUDACall(Caller, FunDecl))
-            return false;
+      if (S.getLangOpts().CUDA) {
+        FunctionDecl *Caller = S.getCurFunctionDecl(/*AllowLambda=*/true);
+        if (!(Caller && Caller->isImplicit()) &&
+            !S.IsAllowedCUDACall(Caller, FunDecl))
+          return false;
+      }
       if (FunDecl->isMultiVersion()) {
         const auto *TA = FunDecl->getAttr<TargetAttr>();
         if (TA && !TA->isDefaultVersion())

clang/lib/Sema/SemaType.cpp

@@ -359,12 +359,14 @@ enum TypeAttrLocation {
   TAL_DeclName
 };
 
-static void processTypeAttrs(TypeProcessingState &state, QualType &type,
-                             TypeAttrLocation TAL,
-                             const ParsedAttributesView &attrs);
+static void
+processTypeAttrs(TypeProcessingState &state, QualType &type,
+                 TypeAttrLocation TAL, const ParsedAttributesView &attrs,
+                 Sema::CUDAFunctionTarget CFT = Sema::CFT_HostDevice);
 
 static bool handleFunctionTypeAttr(TypeProcessingState &state, ParsedAttr &attr,
-                                   QualType &type);
+                                   QualType &type,
+                                   Sema::CUDAFunctionTarget CFT);
 
 static bool handleMSPointerTypeQualifierAttr(TypeProcessingState &state,
                                              ParsedAttr &attr, QualType &type);
@@ -610,7 +612,8 @@ static void distributeFunctionTypeAttr(TypeProcessingState &state,
 /// distributed, false if no location was found.
 static bool distributeFunctionTypeAttrToInnermost(
     TypeProcessingState &state, ParsedAttr &attr,
-    ParsedAttributesView &attrList, QualType &declSpecType) {
+    ParsedAttributesView &attrList, QualType &declSpecType,
+    Sema::CUDAFunctionTarget CFT) {
   Declarator &declarator = state.getDeclarator();
 
   // Put it on the innermost function chunk, if there is one.
@@ -622,19 +625,20 @@ static bool distributeFunctionTypeAttrToInnermost(
     return true;
   }
 
-  return handleFunctionTypeAttr(state, attr, declSpecType);
+  return handleFunctionTypeAttr(state, attr, declSpecType, CFT);
 }
 
 /// A function type attribute was written in the decl spec.  Try to
 /// apply it somewhere.
-static void distributeFunctionTypeAttrFromDeclSpec(TypeProcessingState &state,
-                                                   ParsedAttr &attr,
-                                                   QualType &declSpecType) {
+static void
+distributeFunctionTypeAttrFromDeclSpec(TypeProcessingState &state,
+                                       ParsedAttr &attr, QualType &declSpecType,
+                                       Sema::CUDAFunctionTarget CFT) {
   state.saveDeclSpecAttrs();
 
   // Try to distribute to the innermost.
   if (distributeFunctionTypeAttrToInnermost(
-          state, attr, state.getCurrentAttributes(), declSpecType))
+          state, attr, state.getCurrentAttributes(), declSpecType, CFT))
     return;
 
   // If that failed, diagnose the bad attribute when the declarator is
@@ -646,14 +650,14 @@ static void distributeFunctionTypeAttrFromDeclSpec(TypeProcessingState &state,
 /// Try to apply it somewhere.
 /// `Attrs` is the attribute list containing the declaration (either of the
 /// declarator or the declaration).
-static void distributeFunctionTypeAttrFromDeclarator(TypeProcessingState &state,
-                                                     ParsedAttr &attr,
-                                                     QualType &declSpecType) {
+static void distributeFunctionTypeAttrFromDeclarator(
+    TypeProcessingState &state, ParsedAttr &attr, QualType &declSpecType,
+    Sema::CUDAFunctionTarget CFT) {
   Declarator &declarator = state.getDeclarator();
 
   // Try to distribute to the innermost.
   if (distributeFunctionTypeAttrToInnermost(
-          state, attr, declarator.getAttributes(), declSpecType))
+          state, attr, declarator.getAttributes(), declSpecType, CFT))
     return;
 
   // If that failed, diagnose the bad attribute when the declarator is
@@ -675,7 +679,8 @@ static void distributeFunctionTypeAttrFromDeclarator(TypeProcessingState &state,
 /// `Attrs` is the attribute list containing the declaration (either of the
 /// declarator or the declaration).
 static void distributeTypeAttrsFromDeclarator(TypeProcessingState &state,
-                                              QualType &declSpecType) {
+                                              QualType &declSpecType,
+                                              Sema::CUDAFunctionTarget CFT) {
   // The called functions in this loop actually remove things from the current
   // list, so iterating over the existing list isn't possible.  Instead, make a
   // non-owning copy and iterate over that.
@@ -692,7 +697,7 @@ static void distributeTypeAttrsFromDeclarator(TypeProcessingState &state,
       break;
 
     FUNCTION_TYPE_ATTRS_CASELIST:
-      distributeFunctionTypeAttrFromDeclarator(state, attr, declSpecType);
+      distributeFunctionTypeAttrFromDeclarator(state, attr, declSpecType, CFT);
       break;
 
     MS_TYPE_ATTRS_CASELIST:
@@ -3510,7 +3515,8 @@ static QualType GetDeclSpecTypeForDeclarator(TypeProcessingState &state,
   // Note: We don't need to distribute declaration attributes (i.e.
   // D.getDeclarationAttributes()) because those are always C++11 attributes,
   // and those don't get distributed.
-  distributeTypeAttrsFromDeclarator(state, T);
+  distributeTypeAttrsFromDeclarator(
+      state, T, SemaRef.IdentifyCUDATarget(D.getAttributes()));
 
   // Find the deduced type in this type. Look in the trailing return type if we
   // have one, otherwise in the DeclSpec type.
@@ -4021,7 +4027,8 @@ static CallingConv getCCForDeclaratorChunk(
       // function type.  We'll diagnose the failure to apply them in
       // handleFunctionTypeAttr.
       CallingConv CC;
-      if (!S.CheckCallingConvAttr(AL, CC) &&
+      if (!S.CheckCallingConvAttr(AL, CC, /*FunctionDecl=*/nullptr,
+                                  S.IdentifyCUDATarget(D.getAttributes())) &&
           (!FTI.isVariadic || supportsVariadicCall(CC))) {
         return CC;
       }
@@ -5665,7 +5672,8 @@ static TypeSourceInfo *GetFullTypeForDeclarator(TypeProcessingState &state,
     }
 
     // See if there are any attributes on this declarator chunk.
-    processTypeAttrs(state, T, TAL_DeclChunk, DeclType.getAttrs());
+    processTypeAttrs(state, T, TAL_DeclChunk, DeclType.getAttrs(),
+                     S.IdentifyCUDATarget(D.getAttributes()));
 
     if (DeclType.Kind != DeclaratorChunk::Paren) {
       if (ExpectNoDerefChunk && !IsNoDerefableChunk(DeclType))
@@ -7709,7 +7717,8 @@ static Attr *getCCTypeAttr(ASTContext &Ctx, ParsedAttr &Attr) {
 /// Process an individual function attribute.  Returns true to
 /// indicate that the attribute was handled, false if it wasn't.
 static bool handleFunctionTypeAttr(TypeProcessingState &state, ParsedAttr &attr,
-                                   QualType &type) {
+                                   QualType &type,
+                                   Sema::CUDAFunctionTarget CFT) {
   Sema &S = state.getSema();
 
   FunctionTypeUnwrapper unwrapped(S, type);
@@ -7891,7 +7900,7 @@ static bool handleFunctionTypeAttr(TypeProcessingState &state, ParsedAttr &attr,
 
   // Otherwise, a calling convention.
   CallingConv CC;
-  if (S.CheckCallingConvAttr(attr, CC))
+  if (S.CheckCallingConvAttr(attr, CC, /*FunctionDecl=*/nullptr, CFT))
     return true;
 
   const FunctionType *fn = unwrapped.get();
@@ -8369,7 +8378,8 @@ static void HandleLifetimeBoundAttr(TypeProcessingState &State,
 
 static void processTypeAttrs(TypeProcessingState &state, QualType &type,
                              TypeAttrLocation TAL,
-                             const ParsedAttributesView &attrs) {
+                             const ParsedAttributesView &attrs,
+                             Sema::CUDAFunctionTarget CFT) {
 
   state.setParsedNoDeref(false);
   if (attrs.empty())
@@ -8603,7 +8613,7 @@ static void processTypeAttrs(TypeProcessingState &state, QualType &type,
       // Attributes with standard syntax have strict rules for what they
       // appertain to and hence should not use the "distribution" logic below.
       if (attr.isStandardAttributeSyntax()) {
-        if (!handleFunctionTypeAttr(state, attr, type)) {
+        if (!handleFunctionTypeAttr(state, attr, type, CFT)) {
           diagnoseBadTypeAttribute(state.getSema(), attr, type);
           attr.setInvalid();
         }
@@ -8613,10 +8623,10 @@ static void processTypeAttrs(TypeProcessingState &state, QualType &type,
       // Never process function type attributes as part of the
       // declaration-specifiers.
       if (TAL == TAL_DeclSpec)
-        distributeFunctionTypeAttrFromDeclSpec(state, attr, type);
+        distributeFunctionTypeAttrFromDeclSpec(state, attr, type, CFT);
 
       // Otherwise, handle the possible delays.
-      else if (!handleFunctionTypeAttr(state, attr, type))
+      else if (!handleFunctionTypeAttr(state, attr, type, CFT))
         distributeFunctionTypeAttr(state, attr, type);
       break;
     case ParsedAttr::AT_AcquireHandle: {