[libcxx-commits] [libcxx] [libc++] Improve performance of std::atomic_flag on Windows (PR #163524)

[Roger Sanders via libcxx-commits]

RogerSanders wrote:

> Since this is a new performance improvement and not a regression or fix in any way, I don't think this will get back-ported to previous releases. Every change has the potential for breaking things, and the aim of the point releases is to remove as many bugs as possible. New features should only be added in major releases unless there is a really compelling reason otherwise.

If that's the policy so be it. I'd argue the performance here might be was bad enough to consider it a defect. I was seeing my program run 3x slower vs MSVC STL, and I only had two threads attempting to synchronize using a pair of std::atomic_flag members. That was my motivation to submit this. I can live with it being a major update though. I've already had to mitigate it on the application side.

https://github.com/llvm/llvm-project/pull/163524