Re: SQLFunctionCache and generic plans

čt 27. 2. 2025 v 13:25 odesílatel Alexander Pyhalov <
a(dot)pyhalov(at)postgrespro(dot)ru> napsal:

> Pavel Stehule писал(а) 2025-02-26 22:34:
> > hI
> >
> > I can confirm 60% speedup for execution of function fx and fx3 - both
> > functions are very primitive, so for real code the benefit can be
> > higher
> >
> > Unfortunately, there is about 5% slowdown for inlined code, and for
> > just plpgsql code too.
> >
> > I tested fx4
> >
> > create or replace function fx4(int) returns int immutable as $$ begin
> > return $1 + $1; end $$ language plpgsql;
> >
> > and fx2
> >
> > create or replace function fx2(int) returns int as $$ select 2 * $1;
> > $$
> > language sql immutable;
> >
> > and execution of patched code is about 5% slower. It is strange so
> > this patch has a negative impact on plpgsql execution.
> >
> > Regards
> >
> > Pavel
>
> Hi. I've tried to reproduce slowdown and couldn't.
>
> create or replace function fx4(int) returns int immutable as $$ begin
> return $1 + $1; end $$ language plpgsql;
>
> do $$
> begin
> for i in 1..5000000 loop
> perform fx4((random()*100)::int); -- or fx2
> end loop;
> end;
> $$;
>
> OLD results:
> Time: 8268.614 ms (00:08.269)
> Time: 8178.836 ms (00:08.179)
> Time: 8306.694 ms (00:08.307)
>
> New (patched) results:
> Time: 7743.945 ms (00:07.744)
> Time: 7803.109 ms (00:07.803)
> Time: 7736.735 ms (00:07.737)
>
> Not sure why new is faster (perhaps, some noise?) - looking at perf
> flamegraphs I don't see something evident.
>
> create or replace function fx2(int) returns int as $$ select 2 * $1; $$
> language sql immutable;
> do $$
> begin
> for i in 1..5000000 loop
> perform fx2((random()*100)::int); -- or fx2
> end loop;
> end;
> $$;
>
> OLD results:
> Time: 5346.471 ms (00:05.346)
> Time: 5359.222 ms (00:05.359)
> Time: 5316.747 ms (00:05.317)
>
> New (patched) results:
> Time: 5188.363 ms (00:05.188)
> Time: 5225.322 ms (00:05.225)
> Time: 5203.667 ms (00:05.204)
>

I'll try to get profiles.

Regards

Pavel

>
> --
> Best regards,
> Alexander Pyhalov,
> Postgres Professional
>