Re: [boost] [SoC] multi precision arithmetic library proposal.

From: Stefan Slapeta (stefan_at_[hidden])
Date: 2006-05-01 07:31:03

John Maddock wrote:

> A big int should be too hard, an arbitary precision floating point arithmetc 
> type would be very welcome too.  That's harder as it needs std lib support 
> to be useful (exp, pow, log trig functions etc), so it's rather more work, 
> but then again the student has all summer right ? :-)
> 

IMO a big int would be the most welcome as it would provide the _base_ 
for an arbitrary-length floating point type.

> 
> Very fabuluous indeed, but have you read the spec?  It make me quiver just 
> to think about it!  Maybe as in all things familiarity would simplify things 
> somewhat.

I don't think so. Integer arithmetics is a well-described field and 
there is a lot of good material about it, all well listed under 
http://www2.hursley.ibm.com/decimal. It's not necessary to provide a 
full-optimized implementation with fourier transformations for 
multiplication etc. - this can be added later as well.
A type that does the basic operations and maybe some of exp/pow/log 
would be very helpful, too, to start with!

I'm rather afraid that the design of such a type could become quite 
complex, as you have to provide a good separation of interface and 
actual calculation: this is the only possibility to use the same one 
later for n1977 as well - but this is also a great chance!

Stefan