[Patch][OpenCL] Custom atomic Builtin check ignores address space of a non-atomic pointer

[Anastasia Stulova via cfe-commits]

Thanks!

Yes, I agree it might make sense to drop the default AS parameter value in  getPointerTo(). It doesn't seem like there is an enormous amount of places that need to be changed!

Cheers,
Anastasia

-----Original Message-----
From: Pekka Jääskeläinen [mailto:pekka.jaaskelainen at tut.fi] 
Sent: 02 October 2015 10:20
To: Anastasia Stulova; cfe-commits at lists.llvm.org
Subject: Re: [Patch][OpenCL] Custom atomic Builtin check ignores address space of a non-atomic pointer

LGTM.

Related to it:
There has been so many getPointerTo() issues with multi-AS in the past that I wonder if it'd be time to drop the default value from it, and go through all the places where it's called with the default AS, thus breaking multi-AS.  Might be a worthwhile job to do at some point.

On 09/30/2015 01:23 PM, Anastasia Stulova via cfe-commits wrote:
> Hi all,
>
> Address spaces are not handled in custom semantic checks of atomic Builtins.
>
> If there are two pointers passed to the Builtin, it doesn't allow the 
> second
>
> (non-atomic) one to be qualified with an address space.
>
> This patch removed this restriction by recording the address space of 
> the
>
> passed pointers while checking its type correctness.
>
> Currently, the following code:
>
> _Atomic int __attribute__((address_space(1))) *A;
>
> int __attribute__((address_space(2))) *B;
>
> ...
>
> ... = __c11_atomic_compare_exchange_strong(A, B, 1, 
> memory_order_seq_cst, memory_order_seq_cst);
>
> fails to compile with an error:
>
> "passing '__attribute__((address_space(2))) int *' to parameter of 
> type 'int *' changes address space of pointer".
>
> Please, review the attached fix for it!
>
> Cheers,
>
> Anastasia
>
>
>
> _______________________________________________
> cfe-commits mailing list
> cfe-commits at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
>

--
Pekka