Ray Toy has a translator from Fortran to Lisp.
Converting the resulting Lisp code to use arbitrary precision may be much
easier!
I agree that some of this is pointless without changing the algorithm.
I agree: the usual cosine routine that is good to 53 bits is not going to be
more accurate simply
by running the arithmetic in higher precision. You need a more accurate
algorithm.
(Like the one in maxima bigfloat package, or in mpfr).
RJF
> -----Original Message-----
> From: Alexey Beshenov [mailto:al at beshenov.ru]
... snip....
> I have quite big experience with programming arbitrary
> precision numerics with
> MPFR (in C/C++). I've translated some Netlib routines and I
> should say that
>
> (1) the manual translation from FORTRAN to C/C++ and MPFR
> is rather annoying
> (I'm not sure that it's easy to implement a translator
> since I have no
> experience with parsing and analyzing the FORTRAN code);
>
> (2) the literal mapping from the built-in floating-point
> types of a quite
> small precision to more precise ones isn't very useful,
> so implementing
> an effective arbitrary precision routine isn't so primitive.
>
> MPFR-related stuff is interesting for me, although now I'm
> trying to move to
> the symbolic research (namely, genetic programming in LISP).
>
> --
> Alexey Beshenov <al at beshenov.ru>
> http://beshenov.ru/
>