Re: Using __builtin_expect (likely/unlikely macros)

[Paul Eggert]

On 4/16/19 9:10 AM, Alex Gramiak wrote:
> It would still be good to use it for default builds, no? The GCC
> documentation states:
>
>   It is thus useful to mark functions used to handle unlikely
>   conditions, such as perror, as cold to improve optimization of hot
>   functions that do call marked functions in rare occasions.

That argument would be stronger if functions like 'perror' actually used
__attribute__ ((cold)), which they don't (at least, not in GNU systems).

There is a cost and a benefit to adding these attributes. The benefit is
that they should improve runtime performance very slightly, and (more
important) they should help nonexpert human readers know that a function
is rarely called. The cost is that they make the code harder to
maintain, thus placing a burden on maintainers. Perhaps I'm biased as I
would bear the cost and get none of the benefit; still, the overall
cost-benefit ratio doesn't look all that favorable for Emacs.

That being said, it might make sense for a few obviously-rarely-called
functions like 'emacs-abort' to be marked with __attribute__ ((cold)),
so long as we don't turn this into a mission to mark all cold functions
(which would cost us more than it would benefit). That is what GCC
itself does, with its own functions. However, I'd like to see
performance figures. Could you try it out on the benchmark of 'cd lisp
&& time make compile-always'?