Re: unnecessary executor overheads around seqscans

On Sun, 25 Jan 2026 at 04:36, Andres Freund <andres@anarazel.de> wrote:
> On 2026-01-24 19:36:08 +1300, David Rowley wrote:
> > I also noticed my compiler does not inline SeqNext(). Adding a
> > pg_attribute_always_inline results in it getting inlined and gives a
> > small speedup.
>
> Oh,m that's not good. I think we really had assumed that it would with the 18
> changes around this. It does here, but that's probably because I use -O3.

To reduce the variables here, I've pushed a fix for that after a quick
test showed a 3.9% speedup on a 1 million row table with a single int4
column filtering out all rows. I noticed that clang also didn't inline
with -O2. It does now.

David