Re: locked reads for atomics

On Thu, Feb 22, 2024 at 11:53:50AM -0800, Jeff Davis wrote:
> On Thu, 2024-02-22 at 12:58 +0530, Bharath Rupireddy wrote:
>> There's some immediate use for reads/writes with barrier semantics -
>
> Is this mainly a convenience for safety/readability? Or is it faster in
> some cases than doing an atomic access with separate memory barriers?

The former. Besides the 0002 patch tracked here, there's at least one
other patch [0] that could probably use these new functions. The idea is
to provide an easy way to remove spinlocks, etc. and use atomics for less
performance-sensitive stuff. The implementations are intended to be
relatively inexpensive and might continue to improve in the future, but the
functions are primarily meant to help reason about correctness.

I don't mind prioritizing these patches, especially since there now seems
to be multiple patches waiting on it. IIRC I was worried about not having
enough support for this change, but I might now have it.

[0] https://commitfest.postgresql.org/47/4330/

--
Nathan Bossart
Amazon Web Services: https://aws.amazon.com