This project was created to answer one question: what would be the costs and benefits of making the task scheduling of rayon NUMA-aware, and how would the associated code look like in practice?
Other questions that come up along the way (most recently "What is the very
costly mfence x86 instruction doing in the the generated assembly for
crossbeam::deque's
hot code path?") will hopefully also be answered. But in my ideal world, this
repo will never become a dedicated library on crates.io, rather its ideas will
eventually be integrated into other relevant libraries in the Rust ecosystem.