Re: [v3] conditional tweak

[Martin Sebor <sebor at roguewave dot com>]

Gabriel Dos Reis wrote:

...
> | This conditional actually does happen to have a significant impact
> | on performance -- fill() is called every time a number is inserted
> | into a stream, for example.
>
> compared to the actual work for formatting?

Believe it or not (although it depends if you consider an improvement
of a few tenths of a percent significant and worth the uglification;
I don't mind __builtin_expect() so much but I would probably have
second thoughts about the #pragma since it would have to be guarded
in portable code).

>
...
> I have no idea of how HP aCC $pragma works.  Can you display the actual
> work or is it covered by proprietary rules?

The #pragma is documented here(*). It can be set up to work analogously
to __builtin_expect. This use, for instance

    int foo (int n) {
        if (n < 2) {
    #pragma ESTIMATED_FREQUENCY 0
            return 1;
        }
        return n * foo (n - 1);
    }

is equivalent to gcc's

    int foo (int n) {
        if (__builtin_expect (n < 2, false)) {
            return 1;
        }
        return n * foo (n - 1);
    }

http://docs.hp.com/hpux/onlinedocs/B3901-90008/00/00/44-con.html#bgedagfi

Martin