Re: Optimize PL/Perl function argument passing [PATCH] - Mailing list pgsql-hackers

From Alex Hunsaker
Subject Re: Optimize PL/Perl function argument passing [PATCH]
Date
Msg-id AANLkTimP+YFW3x5qsi2B7kre8h-gUb89er=GqyNyjRu7@mail.gmail.com
Whole thread Raw
In response to Optimize PL/Perl function argument passing [PATCH]  (Tim Bunce <Tim.Bunce@pobox.com>)
Responses Re: Optimize PL/Perl function argument passing [PATCH]  (Andrew Dunstan <andrew@dunslane.net>)
List pgsql-hackers
On Tue, Dec 7, 2010 at 07:24, Tim Bunce <Tim.Bunce@pobox.com> wrote:
> Changes:
>
>    Sets the local $_TD via C instead of passing an extra argument.
>    So functions no longer start with "our $_TD; local $_TD = shift;"
>
>    Pre-extend stack for trigger arguments for slight performance gain.
>
> Passes installcheck.

Cool, surprisingly in the non trigger case I saw up to an 18% speedup.
The trigger case remained about the same, I suppose im I/O bound.

Find attached a v2 with some minor fixes, If it looks good to you Ill
mark this as "Ready for Commit".

Changes:
- move up a declaration to make it c90 safe
- avoid using tg_trigger before it was initialized
- only extend the stack to the size we need (there was + 1 which
unless I am missing something was needed because we used to push $_TD
on the stack, but we dont any more)

Benchmarks:
---
create table t (a int);
create or replace function func(int) returns int as $$ return $_[0];
$$ language plperl;
create or replace function trig() returns trigger as $$
$_TD->{'new'}{'a'}++; return 'MODIFY'; $$ language plperl;

-- pre patch, simple function call, 3 runs
SELECT sum(func(1)) from generate_series(1, 10000000);
Time: 30908.675 ms
Time: 30916.995 ms
Time: 31173.122 ms

-- post patch
SELECT sum(func(1)) from generate_series(1, 10000000);
Time: 26460.987 ms
Time: 26465.480 ms
Time: 25958.016 ms

-- pre patch, trigger
insert into t (a) select generate_series(1, 1000000);
Time: 18186.390 ms
Time: 21291.721 ms
Time: 20782.238 ms

-- post
insert into t (a) select generate_series(1, 1000000);
Time: 19136.633 ms
Time: 21140.095 ms
Time: 22062.578 ms

Attachment

pgsql-hackers by date:

Previous
From: Tom Lane
Date:
Subject: Re: Add support for logging the current role
Next
From: Noah Misch
Date:
Subject: Re: ALTER TYPE 0: Introduction; test cases