Re: Fix performance of generic atomics

Jesper Pedersen <jesper(dot)pedersen(at)redhat(dot)com> writes:
> On 09/05/2017 02:24 PM, Tom Lane wrote:
>> Hm, so if we can't demonstrate a performance win, it's hard to justify
>> risking touching this code. What test case(s) did you use?

> I ran pgbench (-M prepared) with synchronous_commit 'on' and 'off' using
> both logged and unlogged tables. Also ran an internal benchmark which
> didn't show anything either.

That may just mean that pgbench isn't stressing any atomic ops very
hard (at least in the default scenario).

I'm tempted to write a little C function that just hits the relevant
atomic ops in a tight loop, and see how long it takes to do a few
million iterations. That would be erring in the opposite direction,
of overstating the importance of atomic ops to real-world scenarios
--- but if we didn't get any win that way, then it's surely in the noise.

regards, tom lane