Improve square root efficiency #1311
Description
The performance of the square root function is not good. The purpose of this issue is to improve it. This issue is partially motivated by discussion in the ckormanyos/soft_double repo.
The current sqrt_impl implementation can be found in boost/decimal/detail/cmath/sqrt.hpp. It uses a 2nd-order Padé approximation followed by 3-5 Newton-Raphson iterations, which works but has room for performance improvement.
Attempt to use a table-driven method with higher-order corrections and no iterations.
Cc: @mborland and @justinzhuguangwen