What Challenges and Trade-Offs Do Optimising Compilers Face?

Re:What Challenges and Trade-Offs Do Optimising Compilers Face? by phkahler on Thursday June 22, 2017 @11:17AM (#14612814)

Regarding undefined behavior, one thing I very often assume is 2's complement math with wraparound on overflow. That assumption has never failed for me, so it really bothers me that this is undefined in the standard.

I have used processors that have saturating arithmetic and I even used that feature, but even there compiler defaulted to the more common rollover behavior.

[ Parent | on HN ]

Re:What Challenges and Trade-Offs Do Optimising Compilers Face? by usefulcat on Thursday June 22, 2017 @11:41AM (#14613003)

Assuming you're using gcc or something with similar options, you should have a look at -fwrapv if you haven't already.

[ Parent | on HN ]

Re:What Challenges and Trade-Offs Do Optimising Compilers Face? by Animats on Thursday June 22, 2017 @02:02PM (#14614073)

For C, the standard specifies that unsigned arithmetic wraps around, while integer arithmetic overflow is undefined.

I once suggested, back when there were machines that weren't byte-oriented or twos complement, that wraparound arithmetic should be requested with

    unsigned int x;
    x = (x + 1) % 0xffff;

or something similar. That way, you get the same result on all platforms, regardless of word length. If the hardware can truncate cheaply, the compiler can optimize "%" such expressions easily. Other than that, overflow should be an error.

[ Parent | on HN ]