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

[Dmitri Gribenko via libcxx-commits]

gribozavr wrote:

@jyknight I agree with @Artem-B 's analysis. To work around this problem, we need to hide CUDA's `__noinline__` macro definition everywhere where `_LIBCPP_NOINLINE` is used (not just where it is defined) - and it is in potentially in every libc++ header. The only solution that achieves that involves adding `__noinline__` to `_LIBCPP_PUSH_MACROS`, `_LIBCPP_POP_MACROS`, and to `__undef_macros` - like this PR.

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