Re: Cygwin GMP 4.1.4-3 does not work correctly

["Angelo Graziosi (D. Zanello)" <Angelo dot Graziosi at roma1 dot infn dot it>]

Hi,

with the package gmp-4.1.4-1 I have the same problems that were present
with 4.1.3 and that WERE NOT present with 4.1.2!

I have downloaded the source gmp-4.1.4.tar.bz2 from
http://www.swox.com/gmp/
and have built it and uninstalled the Cygwin GMP package.
With this source, all works fine (as Cygwin package 4.1.2)!

The simple LiDIA-GMP application is:

// ==================================== LiDIA_picalc.cpp

#include <iostream>
#include <iomanip>
#include <LiDIA/bigfloat.h>

#ifdef LIDIA_NAMESPACE
using namespace LiDIA;
#endif

using namespace std;

int main()
{
   int nDigits = 10;

   cout << endl
	<< "Number of digits: ";
   cin >> nDigits;

   bigfloat::set_precision(nDigits);

   bigfloat x = Pi();

   cout << setprecision(nDigits) << endl
	<< "PI = " << x << endl;

   return 0;
}

// ====================================

With Cygwin packages it is built with

g++ -O3 -Wall -I/usr/local/include LiDIA_picalc.cpp \
    -s -L/usr/local/lib -lLiDIA -lgmp -o LiDIA_picalc

(/usr/local is the directory where I have installed my builds
of LiDIA-2.1.2 and gmp-4.1.4 from http://www.swox.com/gmp/)

--------------------------------------------

RESULTS with Cygwin packages 4.1.3, 4.1.4

Number of digits: 20

PI = 0.48602824728463545305e-4


--------------------------------------------

RESULTS with Cygwin package 4.1.2,
             build 4.1.4 from http://www.swox.com/gmp/ source


Number of digits: 20

PI = 3.1415926535897932385


With these buids I have verified almost 2000 digits of PI

----------------------------------------------




The simple strictly GMP application is:

// ==================================== picalc.cpp ===================

#include <iostream>
#include <cstdio>
#include <gmp.h>

using namespace std;

int main()
{
   int nBits;
   mpf_t x,y,t,sx,sx1,y1,pi,TWO,HALF,ONE;

   cout << endl
	<< "Number of bits precision (0, default): ";
   cin >> nBits;

   if (nBits > 0) mpf_set_default_prec(nBits);

   // Constant definition
   mpf_init_set_str(ONE,"1e0",10);
   mpf_init_set_str(TWO,"2e0",10);
   mpf_init_set_str(HALF,"5e-1",10);

   mpf_init(x);
   mpf_init(y);
   mpf_init(t);
   mpf_init(sx);
   mpf_init(sx1);
   mpf_init(y1);
   mpf_init(pi);

   cout << "Current precision: " << mpf_get_default_prec() << endl
	<< endl;

   mpf_sqrt(x,TWO);                    // X(0) = sqrt(2)
   mpf_add(pi,TWO,x);                  // PI(0) = 2+sqrt(2)
   mpf_sqrt(y,x);                      // Y(0) = 2^(1/4) = sqrt(X(0))

   mpf_set(sx,y);                      // SX(0) = sqrt(X(0))
   mpf_div(sx1,ONE,sx);                // SX1(0) = 1/SX(0), i = 0

   do
   {
      mpf_add(t,sx,sx1);               // T = SX(i)+SX1(i)
      mpf_mul(x,HALF,t);               // X(i+1) = 0.5*(SX(i)+SX1(i))

      mpf_sqrt(sx,x);                  // SX(i+1) = sqrt(X(i+1))
      mpf_div(sx1,ONE,sx);             // SX1(i+1) = 1/SX(i+1)

      mpf_mul(t,y,sx);                 // T = Y(i)*SX(i+1)+SX1(i+1)
      mpf_add(t,t,sx1);

      mpf_add(x,x,ONE);                // X = X(i+1) + 1
      mpf_add(y,y,ONE);                // Y = Y(i) + 1
      mpf_div(y1,ONE,y);               // Y1 = 1/Y(i)

      // Y(i+1) = [Y(i)*SX(i+1)+SX1(i+1)]/[Y(i)+1]
      mpf_mul(y,t,y1);

      mpf_mul(t,x,y1);                 // T = [X(i+1)+1]/[Y(i)+1]
      mpf_mul(pi,pi,t);                // PI(i+1) = PI(i)*T

      // For i -> inf, PI(i+1) == PI(i) == PI, T == 1
   } while(mpf_eq(ONE,t,1) == 0);

   // To output only significant digits use %.Ff
   gmp_printf("T = %.Ff\nPI= %.Ff",t,pi);

/*
   cout << endl
	   << "T = ";
   mpf_out_str(stdout,10,0,t);
   cout << endl
	<< "PI= ";
   mpf_out_str(stdout,10,0,pi);
*/

   mpf_clear(ONE);
   mpf_clear(TWO);
   mpf_clear(HALF);
   mpf_clear(x);
   mpf_clear(y);
   mpf_clear(t);
   mpf_clear(sx);
   mpf_clear(sx1);
   mpf_clear(y1);
   mpf_clear(pi);
}

// ====================================

With Cygwin packages it is built with

g++ -O3 -Wall -I/usr/local/include picalc.cpp \
    -s -L/usr/local/lib -lgmp -o picalc

(/usr/local is the directory where I have installed my build
of gmp-4.1.4 from http://www.swox.com/gmp/)

--------------------------------------------

RESULTS with Cygwin packages 4.1.3, 4.1.4

Number of bits precision (0, default): 0
Current precision: 60

T = 1.00000000000000000043
PI= 3.14159265358979323743


in this case the bits precision is always a multiple of 30
(this is strange: one expect a multiple of a power of 2!)
--------------------------------------------

RESULTS with Cygwin package 4.1.2,
             build 4.1.4 from http://www.swox.com/gmp/ source


Number of bits precision (0, default): 0
Current precision: 64

T = 1.00000000000000000003
PI= 3.14159265358979323857


in this case the bits precision is always a multiple of 32

With these buids I have verified almost 2000 digits of PI

----------------------------------------------

angelo
angelo.garziosi@roma1.infn.it

On Mon, 11 Oct 2004, Gerrit P. Haase wrote:

> Angelo schrieb:
> 
> Oh not again...
> 
> > I have rebuilt my previous applications (which use also LiDIA).
> > With version 4.1.3-3 of the package they give incorrect results
> > (Pi() prints 0.4...E-4 instead of 3.14..., for example).
> 
> I saw the same problems, but I wasn't able to track this down, so I
> thought it might bbe the program I wanted to do something useful...
> 
> I never looked at the GMP site, oh shit.
> 
> > I have noted at GMP site (http://www.swox.com/gmp/) that 4.1.3 has serious
> > bugs. There it is suggested to pass at 4.1.4.
> 
>  
> > When the 4.1.4 for Cygwin?
> 
> Laop is the maintainer, I have CC'ed him.
> Unfortunately the previous version was also broken (at least for me
> which usually means nothing).
> 
> Lapo, I'll try to build 4.1.4 now, for heavens sake;)
> 
> Gerrit
> -- 
> =^..^=
> 
> 


--
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple
Problem reports:       http://cygwin.com/problems.html
Documentation:         http://cygwin.com/docs.html
FAQ:                   http://cygwin.com/faq/