Q_UNLIKELY and compiler optimization

[kshegunov]

Hello,
If I put a function call inside Q_UNLIKELY can I be sure that it won't be optimized out? Am I gaining the cache branching reduction one'd expect this way, or it's better to use a variable? E.g.

if (Q_UNLIKELY(myfunction() != SUCCESS))
     ; // Do stuff

or should I use:

register int result = myfunction();
if (Q_UNLIKELY(result != SUCCESS))
     ; // Do stuff

Kind regards.

[SGaist]

Hi,

What is the value of success ? Depending on it you can use Q_LIKELY but in any case AFAIK, it won't be optimized out.

[kshegunov]

@SGaist
Success is equal to zero. I'm not expecting MPI_Iprobe (the function in question) to ever return an error, unless something is very wrong with a node. However the constant is external to my application and while it's value probably won't change I can't be certain of it.

[SGaist]

That's the kind of constant that shouldn't change. Usually success = 0 and anything else (usually below 0) is an error code. Sometimes a positive number can be an information like the size of something.

I'd go for the first version of your if in case you have doubts that SUCCESS might change