mpf_get_d

[Torbjorn Granlund]

Paul Zimmermann <Paul.Zimmermann at loria.fr> writes:

         Hi,
  
  it seems mpf_get_d does not truncate its input in the subnormal range.
  See the example program below. On a 64-bit Core 2 under Fedora Core 12,
  I get:
  
  tarte% ./bug3
  lib: 4.3.1
  include: 4.3.1
  0.25903268932681546e-317
  d=1.0612569276778973e-314
  
  Paul
  
  #include <stdio.h>
  #include <stdlib.h>
  #include "gmp.h"
  
  int
  main()
  {
    mpf_t a;
    double d;
  
    printf ("lib: %s\n", gmp_version);
    printf ("include: %d.%d.%d\n", __GNU_MP_VERSION,
            __GNU_MP_VERSION_MINOR, __GNU_MP_VERSION_PATCHLEVEL);
    mpf_init2 (a, 1024);
    mpf_set_str (a, "0.25903268932681546e-317", 10);
    mpf_out_str (stdout, 10, 0, a); printf ("\n");
    d = mpf_get_d (a);
    printf ("d=%.16e\n", d);
    mpf_clear (a);
    return 0;
  }

This was fixed a while ago:

http://shell.gmplib.org:8000/gmp-4.3/rev/32cc41c2fdea
  
Please try the patch.

(GMP 4.3.2 should be released within a few days, with this fix, as well
as an important fix to FFT multiplication.)

-- 
Torbjörn