Re: [ql-users] Sbasic and numbers

[Lau]

By way of a complete diversion, I did produce a *very* high precision 
package for the QL. All in assembler, very optimised, and/but requiring 
the precision (in multiples of 16 bit) to be stated in advance (anything 
from 64 bit up to a ludicrous number of digits). Not IEEE, but close, 
and easily converted. It took the form of a library that operated on a 
stack (very similarly to the QL's FP stack and its operations) and a 
front end to that.

I later converted the whole thing to Delphi, allowing up to 315,643 
digit precision, which you can find on my web site. Converting the 
Pascal to C wouldn't be a huge task, and I might just do that anyway.

The reason I mention it is that it gets to be very inefficient doing 
every operation as a separate *BASIC call - the call overhead itself 
will be as great as the computation time (for precisions less that about 
256 bit, IIRC).

Certainly it's a bad idea to convert to/from decimal! That typically 
takes longer than the calculation itself.

Doing the whole thing in decimal strings is not such a bad idea. If 
anyone is familiar with the "bc" calculator (Linux/*nix), that does it 
that way, and I used some of its ideas for my calculator. There's 
already a QL package (via DP) that does this (quickly enough for 
moderate precision data).

PS. Amongst the tests I did on my calculator's accuracy were verifying 
that IEEE 64 bit operations on an Intel co-processor actually were 
correct (!) and matching my computation of pi (two inbuilt special 
methods and 4*arctan(1)) against various programs that just calculate pi 
(up to 32K digits). Ie. the code is pretty much stable and correct.

--
Lau
http://www.bergbland.info
Get a domain from http://oneandone.co.uk/xml/init?k_id=5165217 and I'll 
get the commission!