Horton

Re: Broken atomics code on PPC with FreeBSD 10.3

Tom Lane <tgl@sss.pgh.pa.us>

View thread View on postgr.es
From: Tom Lane <tgl@sss.pgh.pa.us>
To: Robert Haas <robertmhaas@gmail.com>
Cc: Andres Freund <andres@anarazel.de>, "pgsql-hackers@postgresql.org" <pgsql-hackers@postgresql.org>
Date: 2017-01-03T22:03:59Z
Lists: pgsql-hackers 
Robert Haas <robertmhaas@gmail.com> writes:
> The only point I'm making here is that the width of a spinlock is not
> irrelevant.

Sure, but it does not follow that we need to get all hot and bothered
about the width of pg_atomic_flag, which has yet to find its first
use-case.  When and if its width becomes a demonstrable issue,
we'll have some work to do in this area ... but that was true already.

> All things being equal
> I still think a narrower one is significantly better than a wider one,
> but we can always leave solving that problem to a day when the
> difference can be proved out by benchmarks.

I think we can agree on that conclusion.

			regards, tom lane

Commits

  1. Prefer int-wide pg_atomic_flag over char-wide when using gcc intrinsics.

  2. Allow Pin/UnpinBuffer to operate in a lockfree manner.

  3. Avoid the use of a separate spinlock to protect a LWLock's wait queue.

  4. Move buffer I/O and content LWLocks out of the main tranche.

  5. Improve spinlock performance for HP-UX, ia64, non-gcc.