[Mingw-users] GCC 3.4 Optimization Problem?

[<ad...@ne...>]

/**
 * Calculates the updated value of a checksum.  The values are 4 bytes.
 *
 * @param uOrigSum  the original checksum
 * @param uOldValue a long value to subtract
 * @param uNewValue a long value to add
 * @return          the updated checksum
 */
inline u_short in_cksum_replace_long(u_short uOrigSum,
                                     u_long uOldValue,
                                     u_long uNewValue)
{
    u_long uSum = uOrigSum;
#if 1
    uSum += *((u_short*)&uOldValue) + (~(*((u_short*)&uNewValue)) & 
0xFFFF) + *((u_short*)&uOldValue + 1) + (~(*((u_short*)&uNewValue + 1)) & 
0xFFFF);
#else
    uNewValue = ~uNewValue;
    uSum += (uOldValue & 0xFFFF) + (uOldValue >> 16);
    uSum += (uNewValue & 0xFFFF) + (uNewValue >> 16);
#endif
    uSum = (uSum >> 16) + (uSum & 0xFFFF);
    uSum += (uSum >> 16);
    return uSum;
}

The above function updates the checksum in a packet. The two alternatives 
should be identical in functionality, but the active version cannot give 
reliable results when -O2 is used with GCC 3.4.4. Attached is a complete 
test case (the last two lines of output should be identical). Compiling 
with 3.2 or 2.95.3 is OK.

Removing `inline' will remove the wrong behaviour. Even commenting out one 
`cout << sum << endl;' will make the problem disappear!

Does anyone know whether it is a known bug of GCC 3.4 (regarding inlining 
and addressof)?

Best regards,

Yongwei