Re: Finding the 'double' resolution

[Luiz Henrique de Figueiredo <lhf@...>]

>-- val1 = 9007199254740992
>
>val1 is the first value, on my system (Windows NT on Duron), where an
>increment of 1 isn't signifiant.

Yes, 9007199254740992 = 2^53.
On Intel machines (and most others) doubles are represented using the IEEE 754
standard, which has 52 bits for the mantissa. (There's a hidden bit and so
the precision is actually 53 bits.) For an explantion, see
  http://www.research.microsoft.com/~hollasch/cgindex/coding/ieeefloat.html

For single-precision floats, the limit would be 2^24.
--lhf