[PATCH] [OPENMP] Parsing and sema support for #pragma omp target data" directive.

[Bataev, Alexey]

Kelvin, you don't need fake OMPD_data, instead mark it as OMPD_unknown 
and check explicitly that 'data' identifier is expected

Best regards,
Alexey Bataev
=============
Software Engineer
Intel Compiler Team

01.07.2015 18:03, Kelvin Li пишет:
> ================
> Comment at: lib/Parse/ParseOpenMP.cpp:43-55
> @@ -42,2 +42,15 @@
>             : getOpenMPDirectiveKind(P.getPreprocessor().getSpelling(Tok));
> +
> +  switch (DKind) {
> +  case OMPD_target: {
> +    auto SavedToken = P.getPreprocessor().LookAhead(0);
> +    if (!SavedToken.isAnnotation()) {
> +      if (P.getPreprocessor().getSpelling(SavedToken) == "data") {
> +        DKind = OMPD_target_data;
> +        P.ConsumeToken();
> +      }
> +    }
> +    break;
> +  }
> +  default:
>     for (unsigned i = 0; i < llvm::array_lengthof(F); ++i) {
> ----------------
> ABataev wrote:
>> This code is copied from clang-omp, we're using another solution in official clang. See the patch for 'declare simd' construct (http://reviews.llvm.org/D10599)
> Perhaps I misunderstand it.  The issue of the suggested approach is that it treats 'data' as a directive in order to have { OMPD_target, OMPD_data, OMPD_target_data }.  Creating a fake OMPD_data may be confusing.  That is why I took this approach.
>
> http://reviews.llvm.org/D10765
>
> EMAIL PREFERENCES
>    http://reviews.llvm.org/settings/panel/emailpreferences/
>
>