So, I'm not able to edit the original post myself, but I want new readers to get good info more quickly, which means Zerbina's post and others, rather than my OP.
If I were able to contact an admin, I'd ask that the below be pasted at the top of the OP, subject of course to corrections and improvements. \--- BEGIN EDIT: Some good news: the concerns raised in this post have been addressed. See post by Zerbina below. Nim resizable arrays, whether native in the form of sequences or non-native in the form of imported C++ "std::vector"s (via the cppstl library), can be just as fast as Nim fixed-length arrays, which can be expected to be just as fast as C arrays. Pointer indirection also isn't slower. The slow-down that prompted the OP's concern was, instead - as pointed out by Zerbina - due to: Sequences being used outside of a procedure (at the module level), which made them globals, which placed them in the application's data section. This defeats automatic vectorization (the usage of SIMD assembly), by at least the tested C and C++ compilers, for - apparently - things other than than fixed arrays. So, put your sequences in a procedure. Do it like Dennis Felsing: <https://hookrace.net/blog/what-is-special-about-nim/#nim-runtime-26-s> (and, when you've got some time to invest, here's W. Gordon Goodsman working Nim hardcore on the same problem <https://nim-lang.org/blog/2021/07/28/Nim-Efficient-Expressive-Elegant-Benchmarking.html>) Also consider turning off bounds-checking (again, see Zerbina's post for details). Remember that this is only suitable for well-tested code blocks that you absolutely need to be fast; Nim defaults to automatic bounds-checks for good reasons! \--- ORIGINAL POST FOLLOWS ---
