Comment by dataflow

> because we now know the exact precision requirement for those functions over all 64-bit doubles

Do you have a link or something on this?