LLVM FP atomic min/max variants alignment

[Jonathan Wakely]

On Mon, 30 Mar 2026 at 16:05, Gonzalo Brito via Gcc <gcc@gcc.gnu.org> wrote:
>
> Thank you for the introduction Matthew.
>
> I’ve proposed following clang builtins for libc++ C++26 P3008 implementation:
> https://github.com/llvm/llvm-project/pull/187139
>
>
>   *   __atomic_fetch_fminimum
>   *   __atomic_fetch_fmaximum
>   *   __atomic_fetch_fminimum_num
>   *   __atomic_fetch_fmaximum_num
>
> These builtin names are analogous to __atomic_fetch_min/max .
>
> I’d like to avoid introducing any incompatibilities between gcc and clang here.
> Would the proposed builtins cause any?

Those names look good to me. I think our main concern on the GCC side
was consistently using the __atomic_ prefix that is used for other
existing builtins. These names do that, and the rest of the names are
consistent with the name from the standard. So __atomic_xxx maps to
std::atomic::xxx, which is good.

Thanks!