[Bug target/123206] aarch64: Wrong gating of fp8 intrinsics in aarch64-simd-pragma-builtins.def

[acarlotti at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=123206

--- Comment #3 from Alice Carlotti <acarlotti at gcc dot gnu.org> ---
> 1. Make aarch64-builtins.cc:aarch64_pragma_builtins declared constexpr.  This would prevent uses of global_options via TARGET_* macros.

That sounds like reasonable improvement.

However, I'll note that we might some day need to support more flexible gating
here.  I don't think this has come up (architecturally) in SIMD yet, but there
are multiple cases in SVE/SME where we ought to support an intrinsic when
either of two different features is enabled.

> 2. Make the ctor of aarch64_feature_flags (i.e. bbitmap) declared explicit.  

I originally wanted to prevent any initialisation from a single int, but
Richard Sandiford suggested we should allow it - see
https://gcc.gnu.org/pipermail/gcc-patches/2024-May/652191.html for the original
discussion.

However, I think we can do better.  The specific issue we have in this case is
that our constructor allow conversion from a boolean type.  In future we might
be able to narrow the range of accepted types using C++20 type constraints, but
that isn't an option now. Instead, we can address the current issue by adding a
more specific constructor and marking it private - e.g.:

private:
  constexpr bbitmap(bool x) : val{x} {}