[libcxx-commits] [libcxx] [libc++] Protect the libc++ implementation from CUDA SDK's `__noinline__` macro (PR #73838)

[Nikolas Klauser via libcxx-commits]

https://github.com/philnik777 requested changes to this pull request.

It actually looks like nvc++ isn't working around the issue. They simply parse GNU attributes slightly differently: https://godbolt.org/z/MKKf4Wdan. I think we can fix this all by simply using `[[__gnu__::__noinline__]]` instead. This works on every compiler we support except Clang 16 in C++03 mode, so I don't think that's a huge problem. If anybody complains that the CodeGen is different if cuda is enabled, we can tell them to complain to the cuda sdk people and hopefully it'll be fixed at some point.

https://github.com/llvm/llvm-project/pull/73838