Thank you Mr Plouffe! I'll look it over more closely -- but right away I notice that all of the coefficients in both specfun.f90 and specfun.f have between 18 and 25 digits. There was a time when there were a variety of double-precision hardware implementations (most of them more precise than out present "double precision" but not nearly so precise as today's "quad precision"). For example, the Cray C90's double precision provided about 28 decimal digits. [1] For IEEE binary128, we would need to enhance the precision of the coefficients to 34 digits and perhaps generate new minimax models (for example). - Robert [1] I have a list of many machines, but unfortunately not many of the old mainframes, at http://mrob.com/pub/math/floatformats.html On Mon, Jan 9, 2012 at 12:05, Simon Plouffe <simon.plouffe@gmail.com> wrote:
Hello mr Munafo,
maybe this source code is of interest to you ?
It is about the computation of the gamma function to quadruple precision, if I recall this problem has been solved, I mean there are well established routines for doing that in Fortran or C, you can look at the code here : http://people.sc.fsu.edu/~**jburkardt/f77_src/specfun/**specfun.html<http://people.sc.fsu.edu/~jburkardt/f77_src/specfun/specfun.html>
best regards, simon plouffe
--
Robert Munafo -- mrob.com Follow me at: gplus.to/mrob - fb.com/mrob27 - twitter.com/mrob_27 - mrob27.wordpress.com - youtube.com/user/mrob143 - rilybot.blogspot.com