In C++, which is better i>-1 or i>=0

Question

This might be a silly question to ask, but this kind of optimization is sometimes boost performance of your application. Here I am asking specifically for C++, because the way C++ compile code is a lot different that c# or Java.

The question is which one performs better, if variable i is int.

i > -1
i >= 0

I am looking for performance in terms of memory block or registers required and CPU cycles required for both conditions.

Thanks in advance.

The generated assembly code is nearly same. I don't think there is any difference here. — anishsane, May 22 '15 at 11:01
Check [this](http://stackoverflow.com/questions/12135518/is-faster-than) out. — shauryachats, May 22 '15 at 11:01
Depends a lot on the instruction set. Some have special instructions for test against zero. — cup, May 22 '15 at 11:03
Check this too: http://stackoverflow.com/questions/14520830/x-1-vs-x-0-is-there-a-performance-difference — Drax, May 22 '15 at 11:10
The compiler will use whichever one is faster anyway so it doesn't matter. — puppydrum64, Nov 23 '22 at 12:49

Aracthor · Accepted Answer · 2015-05-22T11:05:05.233

16

In assembly language, both are on the same structure:

i > -1

cmp   [register with i value],-1
jg    [somewhere]

i >= 0

cmp   [register with i value],0
jge   [somewhere]

According to used jump flags, the instruction jg make two flags comparaisons (ZF = 0 and SF = OF) but jge does only one (SF = OF).

So I'm tempted to say that both use almost same registers and CPU cycles, with maybe a very little quicker comparaison for i >= 0.

edited May 22 '15 at 11:05

answered May 22 '15 at 11:03

Aracthor

5,757
6
31
59

Even 2 flag comparisons instead of one would take the same number of cycles, no? – RSinohara May 22 '15 at 11:07
@RSinohara I don't think so. Take the same sample in programming code: what is the slower between `if (something)` and `if (something && something_else)` ? – Aracthor May 22 '15 at 11:10
Loading the immediate value -1 could be more expensive or require more bytes. Unclear, whether the JIT optimized one form to the other. – usr May 22 '15 at 11:10
@usr No. Take a look: http://www.intel.com/content/dam/www/public/us/en/documents/manuals/64-ia-32-architectures-optimization-manual.pdf Every jcc (jump if condition) are grouped together in the Latency/Throughput table in appendix C. – RSinohara May 22 '15 at 11:36
@RSinohara I'm talking about loading the constant -1, not about the jump. Clearly, all jumps and comparisons have the same cost (a tiny one). – usr May 22 '15 at 11:43
Both -1 and 0 will take a byte in machine code. The byte is read in full from memory, taking the very same time (same instruction). It would take longer if you compared to someting >255 – RSinohara May 22 '15 at 11:48

score 2 · Answer 2 · answered May 22 '15 at 11:02

Well, according to logic > operation may be "cheaper" than >=, but I guess you are compiling with Optimization option enabled, so probably the compiler do whatever he wants to optimize your code, so I would say that doesn't matter, even if one is really faster, probably the compiler change it to the best option

score 0 · Answer 3 · answered May 22 '15 at 11:11

When writing a for loop it could be beneficial to convert it from for (i = 0; i < 1000; i++) to for (i = 1000; i > 0; i--) because on some architectures the compiler can skip the compare instruction as a flag will be set when i reaches 0. On modern architectures I'm not sure it matters.

In C++, which is better i>-1 or i>=0

3 Answers3