NO_DBL_TRAPS.cpp

Go to the documentation of this file.

// checking if floating point math traps
 
 
/* IEEE 754 (IEC 559 supeseded by IEC 60559) specifies that
 * the following operations return NaN unless they trap:
 *
 *     1.   sqrt (n); n < 0
 *     2.   rem (x, 0.0), rem (inf, x)
 *     3.   0 * inf
 *     4.   0.0 / 0.0, inf / inf
 *     5.   inf - inf (when both inf have the same sign)
 *
 * Traps are defined as kind of an exception, typically
 * causing the delivery of SIGFPE (e.g., with Compaq C++
 * on Tru64 UNIX, except when using the -ieee flag which
 * disables traps as required by C).
 */
 
#include "nodbg.h"
 
double dbl_get_zero ();
double dbl_get_one ();
 
int main (int argc, char*[])
{
    nodbg ();
 
    double dbl_inf;
    double dbl_nan;
 
    double dbl_zero = dbl_get_zero ();
    double dbl_one  = dbl_get_one ();
 
    // compute infinity (division by zero may trap)
    dbl_inf = dbl_one / dbl_zero;
 
    // computation involving infinities that might trap
    dbl_nan = dbl_zero * dbl_inf;    // #3
    dbl_nan = dbl_zero / dbl_zero;   // #4a
    dbl_nan = dbl_inf / dbl_inf;     // #4b
    dbl_nan = dbl_inf - dbl_inf;     // #5
 
    if (argc > 2)
        return dbl_nan;
 
    if (argc > 1)
        return dbl_inf;
 
    // NEGATIVE test: successful exit status indicates a failure
    return 0;
}
 
// foil optimizers
double dbl_get_zero ()
{
    static double value = 0.0;
 
    return value;
}
 
 
double dbl_get_one ()
{
    static double value = 1.0;
 
    return value;
}