Horton

Thread

Commits

  1. Eliminate divide in new fast-path locking code

  1. Get rid of integer divide in FAST_PATH_REL_GROUP() macro

    David Rowley <dgrowleyml@gmail.com> — 2025-04-14T02:09:16Z

    I noticed a while ago that the new fast-path locking code uses integer
division to figure out the  fast-path locking group slot.  To me, this
seems a bit unnecessary as FastPathLockGroupsPerBackend is always a
power-of-two value, so we can use bitwise-AND instead.

I don't think FAST_PATH_REL_GROUP() is in any particularly hot code
paths, but still, having the divide in there isn't sitting well with
me. Can we get rid of it?

I've attached a patch for that. I also adjusted the method used to
calculate FastPathLockGroupsPerBackend. Also, the Assert that was
going on at the end of the loop in InitializeFastPathLocks() looked a
little odd as it seems to be verifying something that the loop
condition was checking already. I thought it was better to check that
we end up with a power-of-two.

Please see the attached patch.

David

  2. Re: Get rid of integer divide in FAST_PATH_REL_GROUP() macro

    Tomas Vondra <tomas@vondra.me> — 2025-04-26T20:44:00Z

    On 4/14/25 04:09, David Rowley wrote:
> I noticed a while ago that the new fast-path locking code uses integer
> division to figure out the  fast-path locking group slot.  To me, this
> seems a bit unnecessary as FastPathLockGroupsPerBackend is always a
> power-of-two value, so we can use bitwise-AND instead.
> 
> I don't think FAST_PATH_REL_GROUP() is in any particularly hot code
> paths, but still, having the divide in there isn't sitting well with
> me. Can we get rid of it?
> 

Yes, we can get rid of the divide - if we assume power-of-two value
(which seems fine, we already do that, IIRC).

> I've attached a patch for that. I also adjusted the method used to
> calculate FastPathLockGroupsPerBackend. Also, the Assert that was
> going on at the end of the loop in InitializeFastPathLocks() looked a
> little odd as it seems to be verifying something that the loop
> condition was checking already. I thought it was better to check that
> we end up with a power-of-two.
> 
> Please see the attached patch.
> 

Thanks. Those changes seem fine to me to.

Do you intend to push these, or do you want me to do it?


regards

-- 
Tomas Vondra

  3. Re: Get rid of integer divide in FAST_PATH_REL_GROUP() macro

    David Rowley <dgrowleyml@gmail.com> — 2025-04-26T23:55:13Z

    On Sun, 27 Apr 2025 at 08:44, Tomas Vondra <tomas@vondra.me> wrote:
> Thanks. Those changes seem fine to me to.

Thanks for checking.

> Do you intend to push these, or do you want me to do it?

I made a few tweaks to the comments and pushed.

David