> #63 introduced has double rounding, which can be avoided by using the same algorithm Float16 uses _Originally posted by @oscardssmith in https://github.com/JuliaMath/BFloat16s.jl/issues/63#issuecomment-2355758838_