Horton

Re: spinlock support on loongarch64

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

View thread View on postgr.es
From: Tom Lane <tgl@sss.pgh.pa.us>
To: Andres Freund <andres@anarazel.de>
Cc: 吴亚飞 <wuyf41619@hundsun.com>, "pgsql-hackers@lists.postgresql.org" <pgsql-hackers@lists.postgresql.org>
Date: 2022-11-02T18:29:44Z
Lists: pgsql-hackers

Attachments

Andres Freund <andres@anarazel.de> writes:
> On 2022-11-02 11:37:35 -0400, Tom Lane wrote:
>> I wonder if we shouldn't just do that (ie, try to use
>> __sync_lock_test_and_set) as a generic fallback on any unsupported
>> architecture.  We could get rid of the separate stanza for RISC-V
>> that way.  The main thing that an arch-specific stanza could bring
>> is knowledge of the best data type width to use for a spinlock;
>> but I don't see a big problem with defaulting to "int".  We can
>> always add arch-specific stanzas for any machines where that's
>> shown to be a seriously poor choice.

> Yes, please. It might not be perfect for all architectures, and it might not
> be good for some very old architectures. But for anything new it'll be vastly
> better than not having spinlocks at all.

So about like this, then.

			regards, tom lane

Commits

  1. Allow use of __sync_lock_test_and_set for spinlocks on any machine.