when is a double float nearest infinity ?

[Christian Bruel <chrbr at free dot fr>]

using the attached example, I realize that adding any number to DBL_MAX 
gives DBL_MAX, which is perfectly compatible with the ieee 764 rounding 
default rounding mode.

However, doubling DBL_MAX gives infinity, not DBL_MAX. If this an error 
from myself or the standard interpretation, what is the rule ?

If this is standard, what is the limit for which the result is 
'infinity' instead of 'nearest' ?

Another question: is there some support in gcc or the glibc to change 
the default rounding mode to be 'infinity' ? I tried 'fesetround' 
without success.

oh, forgot to say that the example is compiled with 
i586-mandrake-linux-gnu gcc 3.4.3.

attached example compiled with gcc testinf.c -pedantic -ansi -O2

thank you for any clue,

Christian


#include <stdio.h>
#include <float.h>
#include <math.h>

const double infd=(double)1.0/(double)0.0;

void testinf(double v) 
{
  if (v == infd)
    puts("infinity");

  else if (v == DBL_MAX)
    puts("ok nearest");

  else
    puts("error?");
}

main()
{
  volatile double max=DBL_MAX;

  if (FLT_ROUNDS == 1)
    puts ("rounding to nearest");

  testinf(DBL_MAX+1);
  testinf(DBL_MAX+DBL_MAX);
  
}