I should clarify, I meant that ipn_stable and ipn_unstable were separate projects, albeit with some cross-over ideas.
I can't rule out that someone else had that idea for limiting the use of sorting-networks, all I can say is that I came up with that myself. At the same time a lot of the good ideas in ipnsort are ideas from other people, and I try my best to accredit them. As I commented before, my goal is not peak perf for integers at any cost. Here is a commit https://github.com/Voultapher/sort-research-rs/commit/d908fe... that improves perf by 7% across sizes for the random pattern. But that comes with a non negligible binary size and compile time impact even for integers. And while I use heuristics to only use sorting-networks where sensible, they can guess wrong. This is a generic sort implementation with the goal of being a good all-round implementation fit for a standard library and those various uses.
I can't rule out that someone else had that idea for limiting the use of sorting-networks, all I can say is that I came up with that myself. At the same time a lot of the good ideas in ipnsort are ideas from other people, and I try my best to accredit them. As I commented before, my goal is not peak perf for integers at any cost. Here is a commit https://github.com/Voultapher/sort-research-rs/commit/d908fe... that improves perf by 7% across sizes for the random pattern. But that comes with a non negligible binary size and compile time impact even for integers. And while I use heuristics to only use sorting-networks where sensible, they can guess wrong. This is a generic sort implementation with the goal of being a good all-round implementation fit for a standard library and those various uses.