Optimize common path of Once::doit

[stepancheg]

Submitting PR again, because I cannot reopen #13349, and github does not attach new patch to that PR.

Optimize Once::doit: perform optimistic check that initializtion is
already completed. load is much cheaper than fetch_add at least
on x86_64.

Verified with this test:

static mut o: one::Once = one::ONCE_INIT;
unsafe {
    loop {
        let start = time::precise_time_ns();
        let iters = 50000000u64;
        for _ in range(0, iters) {
            o.doit(|| { println!("once!"); });
        }
        let end = time::precise_time_ns();
        let ps_per_iter = 1000 * (end - start) / iters;
        println!("{} ps per iter", ps_per_iter);

        // confuse the optimizer
        o.doit(|| { println!("once!"); });
    }
}

Test executed on Mac, Intel Core i7 2GHz. Result is:

• 20ns per iteration without patch
• 4ns per iteration with this patch applied

Once.doit could be even faster (800ps per iteration), if doit function
was split into a pair of doit/doit_slow, and doit marked as
#[inline] like this:

#[inline(always)]
pub fn doit(&self, f: ||) {
    if self.cnt.load(atomics::SeqCst) < 0 {
        return
    }

    self.doit_slow(f);
}

fn doit_slow(&self, f: ||) { ... }

[lilyball]

The message on this PR doesn't describe what's actually being landed (both the PR description and the commit message). Can you please rewrite it to actually describe what's going on? You can put the extra information back in a comment for posterity's sake, but I'd rather not have the message suggest that it's being marked as #[inline] or being split.

[stepancheg]

Updated the patch and PR description to make it clear, that doit is not split and not inlined.