Useful code and benchmarks on floating point accuracy and performance: https://orlp.net/blog/taming-float-sums/ - relevant for numeric-loops subcrate https://github.com/rust-ndarray/ndarray/issues/46#issuecomment-438458138