Some optimization of exp() and log() is probably possible in all precision ranges.
- Ensure that no excessive divisions/inversions are happening.
- Find the right threshold for cutoff
exp() Taylor series versus Newton iteration (at the moment 2,001 decimal digits).
- Scale argument of
exp(x) down for large x.
- Try for trial implementation of
exp() via AGM methods and see if this beats classic Newton iteration.
- Make dedicated tests for low-precision logarithm.
Some optimization of
exp()andlog()is probably possible in all precision ranges.exp()Taylor series versus Newton iteration (at the moment 2,001 decimal digits).exp(x)down for largex.exp()via AGM methods and see if this beats classic Newton iteration.