[PATCH] D11950: [CUDA] Check register names on appropriate side of cuda compilation only.

[Eli Bendersky via cfe-commits]

eliben added inline comments.

================
Comment at: lib/Sema/SemaDecl.cpp:5944
@@ -5943,3 +5943,3 @@
   ProcessDeclAttributes(S, NewVD, D);
-
+  bool ShouldHandleTargetErrors = DeclAttrsMatchCUDAMode(getLangOpts(), NewVD);
   if (getLangOpts().CUDA) {
----------------
Since this is a CUDA-only thing, ShouldHandleTargetErrors can perhaps be named better. The checks you added are very selective now - some diags are disabled, some are not. It's not clear why some fall under the "should handle target error" umbrella. A clearer name like MatchingCUDAMode or something of the sort, may help? 

================
Comment at: lib/Sema/SemaDecl.cpp:5971
@@ -5970,3 +5970,3 @@
   // Handle GNU asm-label extension (encoded as an attribute).
   if (Expr *E = (Expr*)D.getAsmLabel()) {
     // The parser guarantees this is a string.
----------------
Do we plan to support this for CUDA at all? Why not disable here on top?


http://reviews.llvm.org/D11950