#INFs and casts

[Shoban Jayaraj]

Hi,

I have the following code:

#include <stdio.h>

int main(int argc, char **argv)
{
double d = 1.0e100;

float f = 1.0e100;

float f1 = d;

printf("%6.4f\n", static_cast<float>(d));

printf("%6.4f\n", (float)(d));

printf("%6.4f\n", f);

printf("%6.4f\n", f1);

return 0;
}

that prints

10000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000 0.0000
10000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000 0.0000
1.#INF
1.#INF

My assumption is, as I am casting the doubles to float, it should als
print 1.#INF, but not so. Why?

Thanks in advance,
Shoban Jayaraj

[Pete Becker]

On 2008-07-26 08:05:16 -0400, Shoban Jayaraj <sh****@gmail.comsaid:

>
that prints

10000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000 0.0000

10000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000 0.0000
1.#INF
1.#INF

My

>
assumption is, as I am casting the doubles to float, it should als
print 1.#INF, but not so. Why?

There are two things that interact to produce this result.

First, when you pass an argument of type float as one of the extra
arguments to a function that takes a variable-length argument list,
that argument is promoted to double.

Second, unless you store a value in a variable, the compiler is allowed
to handle it at higher precision than its type calls for. On x86, for
example, floats and doubles are generally handled as 80-bit values,
rather than the 64 bits that a double uses, or the 32 bits that a float
uses.

So when you pass (float)d to printf, the compiler doesn't have to
truncate the value to float; it can simply passes it as a double, and
it looks like that's what's happening here. Same thing for the one with
the static_cast.

When you pass arguments that were stored as floats they are promoted to
double. Both f and f1 have the value infinity, and when they get
promoted to double they remain infinity.

--
Pete
Roundhouse Consulting, Ltd. (www.versatilecoding.com) Author of "The
Standard C++ Library Extensions: a Tutorial and Reference
(www.petebecker.com/tr1book)