CODEWITHSUNDEEP
assemblygccx86-64micro-optimizationabsolute-value
ugo_capeto
ugo_capeto

Reputation: 338

x86 assembly abs() implementation? (revisited)

Extending this question, I was looking for a fast way to abs %ax.

I have two options, I will call them arithmetic and conditional, and avoid stack management for semplicity:

    absax:
        cwd
        xor %dx, %ax
        sub %dx, %ax
        ret

The arithmetic version extends the sign of %ax into %dx, so the way it works is: x = (~x) -(-1) in case the sign bit is 1, essentially performing a neg operation.

    absax:
        mov %ax, %dx
        neg %ax
        cmovs %dx, %ax
        ret

This version is in my opinion simpler to understand and remember, and actually it is what gcc prefers. However I was not able to easily find human opinion about the conditional version. I was only able to find the arithmetic version, so I want to know once and for all what the best version is, and if my preferred one is inferior in any way.

NOTE: After searching for a bit more I found some discussion on the conditional version as well, actually it was present in the original answer, I am deeply sorry. However, I still feel that this question is an attempt to better address the differences between the two approaches.

Upvotes: 0

Views: 170

Answers (0)

Related Questions