double to int64 loses 1 problem

[Tao Ma]

Hi, everybody:

The following code will get different result in gcc3.2.3 and 
gcc3.4 (or higher, like gcc4.0 or gcc4.1.0).

Anybody can give me some idea? Thank you in advance.

-- 
Kind Regards,
Tao Ma

-----------------------------------------
class Integral {
 typedef long long int int64;
 typedef unsigned long long int ullong;

public:
 static const double TWO_EXP63 = ((ullong)1 << 63);

};


#include <iostream>
#include <math.h>
using namespace std;
int main() {
 typedef long long int int64;
 typedef unsigned long long int ullong;

 int64 num_4 = (int64)pow((double)2, (double)63);
 int64 num_5 = (int64)Integral::TWO_EXP63;

 cout << "gcc3.2.3 will get: 0x8000000000000000" << endl;
 cout << "gcc4.1.0 will get: 0x7fffffffffffffff" << endl;
 cout << hex << num_4 << endl;
 cout << hex << num_5 << endl;
}