[cfe-commits] patch for review

[Reed Kotler]

Okay. Thanks.

The reason why I added the

-triple x86_64-unknown-linux-gnu

is because it's possible to run make check on a Mips platform. We even 
have a build bot connected
to llvm what runs make-check on a mips board. This is why I had two 
different test cases.

mips16, nomips16 should emit warnings except when it is a mips target.

I will add Subjects=. I had that originally but it seemed to be causing 
problems. There might have been some other reason for that. I will try 
adding it back in. It might have been some kind of make bug for clang. 
Long story... but I will recheck.

On 01/12/2013 06:44 AM, Dmitri Gribenko wrote:
> On Sat, Jan 12, 2013 at 4:32 PM, Reed Kotler <rkotler at mips.com> wrote:
>> I don't have commit access to the clang list (as far as I know; just llvm).
>>
>> Please commit if you approve the patch.
> Since you have commit access to llvm, you have commit access to clang,
> and every other repository.
>
>> This patch is the first step to adding function attributes mips16 and
>> nomips16 to the mips
>> target.
> Please add
> let Subjects = [Function];
> just for documentation purposes now.  Somebody will come along in
> future and will write the necessary TableGen magic to make
> lib/Sema/SemaDeclAttr.cpp autogenerated.  (We hope!)
>
> +    bool ProcessDeclAttribute(Scope *scope, Decl *D, const AttributeList &Attr,
> +                              Sema &S) const {
>
> 1. scope -> Scope
>
> 2. Please add checks to ensure that there are no parameters passed to
> the attributes.
>
> 3. Please add tests for (2), like:
>
> void __attribute__((mips16(foo))) foo32(); // expected-error {{whatever}}
> void __attribute__((mips16(10))) foo32(); // expected-error {{whatever}}
>
> and same for nomips16.
>
> +// RUN: %clang_cc1 -triple x86_64-unknown-linux-gnu -fsyntax-only -verify %s
>
> Triple is not needed in a target-independent test.
>
> +int main() {
> +  foo32();
> +  foo16();
> +}
>
> That is not needed.
>
> Another suggestion: should we reject these attributes when we are not
> targeting mips?
>
> Dmitri
>