[SYCL] Fix check for reqd_sub_group_size attribute mismatches

Fixed an issue with comparing pointers to Attr objects rather than
values passed to the attribute in source code.
Refactored IntelReqdSubGroupSizeAttr class to store evaluated value of
attribute value for further uses to avoid evaluating it each time we
need that value to check something.

The bug were uncovered after #1778 landed, but it had been
introduced in #1807

Author: AlexeySachkov

clang/include/clang/Basic/Attr.td

@@ -1289,11 +1289,28 @@ def LoopUnrollHint : InheritableAttr {
 
 def IntelReqdSubGroupSize: InheritableAttr {
   let Spellings = [GNU<"intel_reqd_sub_group_size">, CXX11<"cl", "intel_reqd_sub_group_size">];
-  let Args = [ExprArgument<"SubGroupSize">];
+  let Args = [ExprArgument<"SubGroupSizeExpr">];
   let Subjects = SubjectList<[Function, CXXMethod], ErrorDiag>;
   let Documentation = [IntelReqdSubGroupSizeDocs];
   let LangOpts = [OpenCL, SYCLIsDevice, SYCLIsHost];
   let PragmaAttributeSupport = 0;
+  let AdditionalMembers = [{
+  private:
+    int SubGroupSize = 0;
+
+  public:
+    int getSubGroupSizeEvaluated(ASTContext &Ctx) {
+      if (SubGroupSize != 0)
+        return SubGroupSize;
+      llvm::APSInt Val(32);
+      bool Succeedded =
+          this->getSubGroupSizeExpr()->isIntegerConstantExpr(Val, Ctx);
+      assert(Succeedded && "expression must be constant integer");
+      SubGroupSize = Val.getSExtValue();
+      assert(SubGroupSize >= 0 && "invalid value of the attribute");
+      return SubGroupSize;
+    }
+  }];
 }
 
 // This attribute is both a type attribute, and a declaration attribute (for

clang/lib/CodeGen/CodeGenFunction.cpp

@@ -565,16 +565,10 @@ void CodeGenFunction::EmitOpenCLKernelMetadata(const FunctionDecl *FD,
     Fn->setMetadata("reqd_work_group_size", llvm::MDNode::get(Context, AttrMDArgs));
   }
 
-  if (const IntelReqdSubGroupSizeAttr *A =
-          FD->getAttr<IntelReqdSubGroupSizeAttr>()) {
-    llvm::APSInt ArgVal(32);
+  if (IntelReqdSubGroupSizeAttr *A = FD->getAttr<IntelReqdSubGroupSizeAttr>()) {
     llvm::LLVMContext &Context = getLLVMContext();
-    bool IsValid = A->getSubGroupSize()->isIntegerConstantExpr(
-        ArgVal, FD->getASTContext());
-    assert(IsValid && "Not an integer constant expression");
-    (void)IsValid;
-    llvm::Metadata *AttrMDArgs[] = {
-        llvm::ConstantAsMetadata::get(Builder.getInt32(ArgVal.getSExtValue()))};
+    llvm::Metadata *AttrMDArgs[] = {llvm::ConstantAsMetadata::get(
+        Builder.getInt32(A->getSubGroupSizeEvaluated(FD->getASTContext())))};
     Fn->setMetadata("intel_reqd_sub_group_size",
                     llvm::MDNode::get(Context, AttrMDArgs));
   }

clang/lib/Sema/SemaSYCL.cpp

@@ -308,14 +308,6 @@ static void reportConflictingAttrs(Sema &S, FunctionDecl *F, const Attr *A1,
   F->setInvalidDecl();
 }
 
-// Returns the signed constant integer value represented by given expression.
-static int64_t getIntExprValue(Sema &S, const Expr *E) {
-  llvm::APSInt Val(32);
-  bool IsValid = E->isIntegerConstantExpr(Val, S.getASTContext());
-  assert(IsValid && "expression must be constant integer");
-  return Val.getSExtValue();
-}
-
 class MarkDeviceFunction : public RecursiveASTVisitor<MarkDeviceFunction> {
 public:
   MarkDeviceFunction(Sema &S)
@@ -1696,15 +1688,16 @@ void Sema::MarkDevice(void) {
               KernelBody ? KernelBody->getAttr<SYCLSimdAttr>() : nullptr;
           if (auto *Existing =
                   SYCLKernel->getAttr<IntelReqdSubGroupSizeAttr>()) {
-            if (Existing->getSubGroupSize() != Attr->getSubGroupSize()) {
+            if (Existing->getSubGroupSizeEvaluated(getASTContext()) !=
+                Attr->getSubGroupSizeEvaluated(getASTContext())) {
               Diag(SYCLKernel->getLocation(),
                    diag::err_conflicting_sycl_kernel_attributes);
               Diag(Existing->getLocation(), diag::note_conflicting_attribute);
               Diag(Attr->getLocation(), diag::note_conflicting_attribute);
               SYCLKernel->setInvalidDecl();
             }
           } else if (KBSimdAttr &&
-                     (getIntExprValue(*this, Attr->getSubGroupSize()) != 1)) {
+                     (Attr->getSubGroupSizeEvaluated(getASTContext()) != 1)) {
             reportConflictingAttrs(*this, KernelBody, KBSimdAttr, Attr);
           } else {
             SYCLKernel->addAttr(A);

clang/lib/Sema/SemaTemplateInstantiateDecl.cpp

@@ -547,7 +547,7 @@ static void instantiateIntelReqdSubGroupSizeAttr(
   // The SubGroupSize expression is a constant expression.
   EnterExpressionEvaluationContext Unevaluated(
       S, Sema::ExpressionEvaluationContext::ConstantEvaluated);
-  ExprResult Result = S.SubstExpr(Attr->getSubGroupSize(), TemplateArgs);
+  ExprResult Result = S.SubstExpr(Attr->getSubGroupSizeExpr(), TemplateArgs);
   if (!Result.isInvalid())
     S.addIntelReqdSubGroupSizeAttr(New, *Attr, Result.getAs<Expr>());
 }

clang/test/SemaSYCL/reqd-sub-group-size-device.cpp

@@ -45,7 +45,16 @@ void bar() {
     baz();
   });
 #endif
+
   kernel<class kernel_name5>([]() [[cl::intel_reqd_sub_group_size(2)]] { });
+  kernel<class kernel_name6>([]() [[cl::intel_reqd_sub_group_size(4)]] { foo(); });
+}
+
+[[cl::intel_reqd_sub_group_size(16)]] SYCL_EXTERNAL void B();
+[[cl::intel_reqd_sub_group_size(16)]] void A() {
+}
+[[cl::intel_reqd_sub_group_size(16)]] SYCL_EXTERNAL void B() {
+  A();
 }
 
 #ifdef TRIGGER_ERROR