Get the latest tech news
Spice: Fine-grained parallelism with sub-nanosecond overhead in Zig
Fine-grained parallelism with sub-nanosecond overhead in Zig - judofyr/spice
The benchmark in the figure above (summing over the nodes in a binary tree) is typically one of the worst cases for parallelism frameworks: The actual operation is extremely fast so any sort of overhead will have a measurable impact. This turns out to be suboptimal because (1) stack usage matters for performance (at least in this benchmark) and (2) there's quite a lot of additional state needed to keep track of futures which are executing. This paper provides excellent information about the concepts, but the implementation is based around integrating this into an interpreter and focus is primarily on the theoretical guarantees as opposed to raw performance.
Or read this on Hacker News