Re: call popcount32/64 directly on non-x86 platforms - Mailing list pgsql-hackers

From John Naylor
Subject Re: call popcount32/64 directly on non-x86 platforms
Date
Msg-id CAFBsxsGX48JVBTggHoFWvAXJbBbRZANEqXyP+3kH+dw960BViA@mail.gmail.com
Whole thread Raw
In response to Re: call popcount32/64 directly on non-x86 platforms  (David Rowley <dgrowleyml@gmail.com>)
Responses Re: call popcount32/64 directly on non-x86 platforms  (John Naylor <john.naylor@enterprisedb.com>)
List pgsql-hackers

On Thu, Aug 12, 2021 at 1:26 AM David Rowley <dgrowleyml@gmail.com> wrote:
> Closer, but I don't see why there's any need to make the fast and slow
> functions external.  It should be perfectly fine to keep them static.
>
> I didn't test the performance, but the attached works for me.

Thanks for that! I still get a big improvement to on Power8 / gcc 4.8, but it's not quite as fast as earlier versions, which were around 200ms:

master: 646ms
v3:     312ms

This machine does seem to be pickier about code layout than any other I've tried running benchmarks on, but that's still a bit much. In any case, your version is clearer and has the intended effect, so I plan to commit that, barring other comments.

I think I'll leave my v2-0002 aside for now, since it has wider implications, and I have bigger things to work on.

--
John Naylor
EDB: http://www.enterprisedb.com

pgsql-hackers by date:

Previous
From: Michael Meskes
Date:
Subject: Re: ECPG bug fix: DECALRE STATEMENT and DEALLOCATE, DESCRIBE
Next
From: John Naylor
Date:
Subject: Re: badly calculated width of emoji in psql