PUSH ECX faster? Or MOV Count, ECX?
Author
Hi guys I'm back, back with a stipid question. So which one is
faster? Considerin' that both can be used and I have to choose
one. Thanks!
Unless you're programming a 486 or earlier, it won't make a difference.
I'd probably use the stack directly with push since its more 'assembly-like' imo. Using variables is for sissies =-P
I'd probably use the stack directly with push since its more 'assembly-like' imo. Using variables is for sissies =-P
Given that (non-register passing) function calls have their parameters passed with PUSHes for every compiler that I've ever seen, rather than a SUB and a bunch of MOVs, I'd say jsut go with PUSH.
Author
Thanks guys, that's helpful. I'm goin' with BUSH, I mean, PUSH.
Anyway my friend told me that PUSH is built in, so it only
occupies a single code byte (0x51), whereas "MOV Count, ECX"
occupies three code bytes and it's slower. Yeah and I saw that
in my dissasembly too :)
Anyway my friend told me that PUSH is built in, so it only
occupies a single code byte (0x51), whereas "MOV Count, ECX"
occupies three code bytes and it's slower. Yeah and I saw that
in my dissasembly too :)
If you have to ask (i.e. can't figure out how to test it for yourself), then you are pretty much automatically not qualified to make use of the answer.
Not necessarily. It may be that he could go forth and do a bunch of benchmarks, and infer, from the data gained, which approach is faster. But instead of doing all that stuffing around, he pops the question on GameDev.Net, on the assumption that with so many asm programmers around, someone's probably already done the groundwork and he'd be wasting his time to repeat it.
Linkay. Which may or may not be useful.
Linkay. Which may or may not be useful.
Quote:Original post by HaywireGuy
Thanks guys, that's helpful. I'm goin' with BUSH, I mean, PUSH.
Anyway my friend told me that PUSH is built in, so it only
occupies a single code byte (0x51), whereas "MOV Count, ECX"
occupies three code bytes and it's slower. Yeah and I saw that
in my dissasembly too :)
These days pretty much all of the ALU operations (including MOV and PUSH/POP) are hardwired into the chip already. Yes the alternate encoding for PUSH register operand onto the stack is only one byte, and I think a MOV mem/reg operation is two bytes followed by up to a four byte memory value.
The Intel Optimization manual gives PUSH with latency of 1.5 and throughput of 1, while MOV has a latency of 0.5 and throughput of 0.5. Remember that these are just estimates, and what usually matters more on modern processors is memory and cache performance. Either way I doubt you'll get much speed difference between the two.
Quote:Original post by ZahlmanAin't that the absolute truth!
If you have to ask (i.e. can't figure out how to test it for yourself), then you are pretty much automatically not qualified to make use of the answer.
I was going to somehow say the same thing, but you couldn't have said it better.
Author
Zahlman, I was waitin' for a reply like that given by
DukeAtreides076 and the rest, which my test application won't
tell me. So, yah, I'm here.
Cheers fractoid, I'm sure that's gonna be useful :)
Duke, I'm sure that's true, maybe I'm just being a lit'
paranoid when I have these MOVs in a very tight loop that
executes millions of times. But anyway, thanks!
DukeAtreides076 and the rest, which my test application won't
tell me. So, yah, I'm here.
Cheers fractoid, I'm sure that's gonna be useful :)
Duke, I'm sure that's true, maybe I'm just being a lit'
paranoid when I have these MOVs in a very tight loop that
executes millions of times. But anyway, thanks!
Author
Thanks iMalc, that's the ultimate truth, that's why I
was "asking a stupid question". See?
was "asking a stupid question". See?
This topic is closed to new replies.
Advertisement
Popular Topics
Advertisement
