[libcxx-commits] [libcxx] [libc++] Implement P0528R3 `std::atomic` CAS for types with paddings (PR #76180)

[via libcxx-commits]

huixie90 wrote:

I am a bit struggling that the codegen seems to be much worse in some cases.  For example,
https://godbolt.org/z/9K35c36eq
- If I don't capture `success` memory order in the lambda, everything is nicely inlined
- as long as I capture the `success` memory order, it will generate all combination of the memory order in CAS and a switch statement to select the right one. and lambda can no longer inlined due to this complexity

https://github.com/llvm/llvm-project/pull/76180