[CUDA] Reject the alias attribute in CUDA device code.

[jlebar]

Summary

CUDA (well, strictly speaking, NVPTX) doesn't support aliases.

[echristo]

OK.

-eric

[echristo]

OK. One question inline, but feel free to commit after answering one way or the other :)

-eric

test/SemaCUDA/alias.cu

1–2 ↗

(On Diff #45783)

Why does this require a registered target? Pretty sure TargetInfo should be able to cope with all of these.

[jlebar]

test/SemaCUDA/alias.cu

1–2 ↗

(On Diff #45783)

Honestly I've been cargo-culting those lines from other SemaCUDA tests. Would you mind elaborating on what they're for?

[jlebar]

test/SemaCUDA/alias.cu

1–2 ↗

(On Diff #45783)

Removed them in the meantime.

cfe/trunk/include/clang/Basic/DiagnosticSemaKinds.td

 def warn_kern_is_method : Extension<
   "kernel function %0 is a member function; this may not be accepted by nvcc">,
   InGroup<CudaCompat>;
 def warn_kern_is_inline : Warning<
   "ignored 'inline' attribute on kernel function %0">,
   InGroup<CudaCompat>;
 def err_va_arg_in_device : Error<
   "CUDA device code does not support va_arg">;
+def err_alias_not_supported_on_nvptx : Error<"CUDA does not support aliases">;
 
 def warn_non_pod_vararg_with_format_string : Warning<
   "cannot pass %select{non-POD|non-trivial}0 object of type %1 to variadic "
   "%select{function|block|method|constructor}2; expected type from format "
   "string was %3">, InGroup<NonPODVarargs>, DefaultError;
 // The arguments to this diagnostic should match the warning above.
 def err_cannot_pass_objc_interface_to_vararg_format : Error<
   "cannot pass object with interface type %1 by value to variadic "

cfe/trunk/lib/Sema/SemaDeclAttr.cpp

static void handleAliasAttr(Sema &S, Decl *D, const AttributeList &Attr) {
   StringRef Str;
   if (!S.checkStringLiteralArgumentAttr(Attr, 0, Str))
     return;
 
   if (S.Context.getTargetInfo().getTriple().isOSDarwin()) {
     S.Diag(Attr.getLoc(), diag::err_alias_not_supported_on_darwin);
     return;
   }
+  if (S.Context.getTargetInfo().getTriple().isNVPTX()) {
+    S.Diag(Attr.getLoc(), diag::err_alias_not_supported_on_nvptx);
+  }
 
   // Aliases should be on declarations, not definitions.
   if (const auto *FD = dyn_cast<FunctionDecl>(D)) {
     if (FD->isThisDeclarationADefinition()) {
       S.Diag(Attr.getLoc(), diag::err_alias_is_definition) << FD;
       return;
     }
   } else {

cfe/trunk/test/SemaCUDA/alias.cu

+// RUN: %clang_cc1 -triple nvptx-unknown-cuda -fsyntax-only -fcuda-is-device -verify -DEXPECT_ERR %s
+// RUN: %clang_cc1 -triple x86_64-unknown-linux-gnu -fsyntax-only -verify %s
+
+// The alias attribute is not allowed in CUDA device code.
+void bar();
+__attribute__((alias("bar"))) void foo();
+#ifdef EXPECT_ERR
+// expected-error@-2 {{CUDA does not support aliases}}
+#else
+// expected-no-diagnostics
+#endif