Subject: quad double arithmetic; was Re: [Maxima] linalg
From: Richard Fateman
Date: Mon, 6 Feb 2006 08:31:19 -0800
I have now working, in a fairly direct fashion, arithmetic for
dd and qd float formats.
dd = double double, twice the fraction size
qd = quad double 4 times the fraction size (approx. 64 decimal digits)
It might be neat to make this part of maxima, and show some demos
where extra-length accumulation makes a difference.
There is software under development to do this using C++, Fortran,
etc, but this is for lisp.
The source code is open, and compiles under linux etc. I compiled
it under Windows (visual studio), for a windows dll, which is on my
directory under
www.cs.berkeley.edu/~fateman/generic
along with other code.
(In addition to arithmetic, there's also sin/cos/log/etc)
If anyone out there is interested, contact me. I have one student here
at Berkeley looking at a term project; his part would probably
not involve Maxima per se, just some lisp interface, perfected.
I wrote the interface in ff for Allegro, but uffi should work for
any other lisp.
For more info on qd, look at the home page of Yozo Hida.
Please do keep in touch though!
Oh, why qd and dd when bigfloat or gmpf or gmpfr does it too?
Speed. Quad double is, when you think about it, lots of bits:
far more than you usually have any justification for thinking are
correct in any physical measurement situation.
RJF