Kalman VS nanobench

Since you mentioned C++ and Kalman filters, I author this Kalman filter library which helped me to get reacquainted to control theory, tries to be approachable, and lists a variety of sources to ease in the topic with examples.

Yes, and here's one for sanitizers. Other tools, documentation, and coverage in the neigbhorong files.

I've been learning FectContent as well with some successes and failures.

My goal with this Kalman filter for C++ is to solve your exact question.

I released a first version of a generic Kalman filter.

I use fmt and others in my project with CMake fetch and it's been a good experience so far: fetch, declare, link in a few lines. Hope this can be useful to someome.

Gratuitous French codebase self-promotion though. /s

Would you be open to exploring an implementation collaboration? I author a C++ Kalman library and would like to expand demonstrators.

A façade (example) included only for your GCC builds and with the fmt library would allow you to generically use the std::format support in your code. Avoiding the fmt:: and dependencies with MSVC. When the support lands in GCC, only that file would need to be deleted.

I use it with boilerplate similar to this: https://github.com/FrancoisCarouge/Kalman/blob/develop/benchmark/benchmark.cpp and run the executables with priority and pining: nice -n 20 tasker -- cpu-list 0

An alternative is tracking the number of instructions a test executes: https://github.com/martinus/nanobench

Nano bench is a great library with low overhead. https://github.com/martinus/nanobench

I've done a relatively simple test of taking random doubles (between 0 and 1), converting them to a C string via std::to_chars and then converting that C string back to a double via std::from_chars vs his xeerx::chars_to and got the following results on my machine via nanobench:

I like https://nanobench.ankerl.com/

Of course, the best way to benchmark is nanobench: https://nanobench.ankerl.com/

I've created robin-hood-hashing and nanobench, and recently made some contributions to Bitcoin and doxygen

Right. Regex runtime construction is known to be slow, so ideally the state machinery construction is built at compile time (boost.xpressive, ctre). Also, boost.regex is faster than most of the std implementations if compile time isn’t possible. And if that’s no good rewrite without regex. Since it sounds like it’s all encapsulated at least it would be easy to measure the options. These days I use this one to compare https://nanobench.ankerl.com/

You can check the API of nanobench which also doesn't use macros, as far as I have used it.

I've tried again with your latest changes and decided to use https://github.com/martinus/nanobench for a better benchmark and got the following output: