> especially since the original code used highly-optimised assembly incantations from our resident cycle wizard.
He/she might be cool, but I think someone familiar with assembly optimisation would think to look at the generated assembly as well, after noticing the speedup is not what you expect. Don't go hand-writing assembly routines thinking your code is fast without actually checking whether your code is fast. :)
They're certainly familiar with assembly itself, but the `asm!` macro and the related constructs in Clang/GCC is its own beast and has a lot of footguns (like GCC allocating registers as overlapping by default). I don't begrudge them writing suboptimal assembly, since it already beat our Rust implementation by a significant margin.
He/she might be cool, but I think someone familiar with assembly optimisation would think to look at the generated assembly as well, after noticing the speedup is not what you expect. Don't go hand-writing assembly routines thinking your code is fast without actually checking whether your code is fast. :)