Hacker News Reader

Comment by heyjudy | Hacker News Reader

I still remember the olden days when

   XOR AX,AX

was faster than

   MOV AX,0

and Abrash'es book was the magnum opus of the pre-P6 era.

I guess I better throw most of those assumptions out for modern microarchitectures.

View on HN

The XOR is still idiomatic because it is smaller than the MOV, because the immediate zero has to be encoded as the same size as the destination register.

Probably not directly faster, but "xor %rax, %rax" is still smaller (and thus faster if you're icache constrained) than "mov $0x0, %rax"

It is still faster because the register renamer handles it --- ditto for "sub reg, reg":

https://randomascii.wordpress.com/2012/12/29/the-surprising-...

"xor %eax, %eax" is even smaller (2 bytes vs. 3) and does the same thing since operations on the 32-bit subregisters are defined to zero the upper half [1]. This is what compilers tend to do when they want to zero a register.

[1] This is not true for 8- or 16-bit subregisters, causing all kinds of fun issues (see "partial register stalls", also discussed in Agner Fog's materials).

>"and Abrash'es book was the magnum opus of the pre-P6 era."

What book is that?

Probably Zen of Code Optimization. Good times.

The Zen of Assembler? Does that sound correct? That's the closest I could find.

Zen of Code Optimization 1994 ( https://www.amazon.com/Zen-Code-Optimization-Ultimate-Softwa...) is the follow up to Zen of Assembly Language 1990 (https://www.amazon.com/Zen-Assembly-Language-Knowledge-Progr...)

XOR was never faster, it is 2 bytes-vs 3 bytes instruction size.

Reply on news.ycombinator.com