algorithm to calculate pi

[hypercube1]

i'm working on an implementation of the gauss-legendre algorithm to calculate pi and all runs quite smoothly but i was wondering how to get more digits outputted, cout.precision() seems to top out at 17?

#include <iostream>
#include <math.h>
using namespace std;

int main()
{
    bool run=true;
    double a=1, aa, b=pow(2,-.5), t=.25, p=1, r;
    int iterations, exit;
    cin >> iterations;
    while(run)
    {
         for(int i=0;i<iterations;i++)
         {
             aa=(a+b)/2;
             b=pow((a*b),.5);
             t=t-p*pow((a-aa),2);
             p=2*p;
             r=(pow((a+b),2))/(4*t);
             a=aa;
         }
         cout.precision(17);
         cout << r;
         cin >> a;
         run=false;
    }
    return 0;
}

i was thinking maybe some sort set of conditionals that will kick it over to another variable or maybe an use an array to hold each digit seperately. any ideas?

[Disch]

precision is not topping out. The capacity for 'double' is.

doubles can only hold so much information. The more decimal places you put on a number, the more information you need. 17 digits requires a lot of memory for a single primitive type. It's more than double can offer you.

You have the following options:

1) accept that double can't get any more precise and settle on a 17 digit value for pi

2) upgrade to 'long double' and get a few more digits for precision

3) abandon primitive types and use something like a bignum library

[hypercube1]

Thanks, what would you suggest for a bignum library?

[Disch]

I've never used any as I haven't had the need.

But this one sounds good:

http://gmplib.org/

It's also the first hit on google for "bignum" (well, 2nd... the first was wikipedia) ;P

[Duthomhas]

GNU MP Bignum +1

It has a C++ class interface to the system, which makes it really easy to use in C++.

[edit] BTW, please #include <cmath>