Signed overflow optimization hazards in the kernel [LWN.net]

Signed overflow optimization hazards in the kernel

Posted Aug 18, 2012 22:56 UTC (Sat) by jzbiciak (guest, #5246)
In reply to: Signed overflow optimization hazards in the kernel by cmccabe
Parent article: Signed overflow optimization hazards in the kernel

Where I've heard it coming up is when you have code that effectively looks like this:

    for (i = x; i < x + 16; i++)
    {
        /* code that does not modify either x or i */
    }

The compiler wants to know it's safe to assume this loop goes around 16 times. That isn't true if "x + 16" could overflow.

to post comments

Signed overflow optimization hazards in the kernel

Posted Aug 20, 2012 7:22 UTC (Mon) by jezuch (subscriber, #52988) [Link]

> The compiler wants to know it's safe to assume this loop goes around 16 times. That isn't true if "x + 16" could overflow.

If you want to make it explicit, there's a flag for this in GCC: -funsafe-loop-optimizations (along with -Wunsafe-loop-optimizations if you want to know when it happens; it's a warning, though, so beware of build environments which insist on -Werror). AFACT there are two cases handled by this flag: assuming that the loop counter does not overflow, and assuming that the loop is not infinite. Don't know the exact implementation details, though.