From: | Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> |
---|---|
To: | Andres Freund <andres(at)anarazel(dot)de> |
Cc: | Robert Haas <robertmhaas(at)gmail(dot)com>, Jeevan Ladhe <jeevan(dot)ladhe(at)enterprisedb(dot)com>, PostgreSQL Developers <pgsql-hackers(at)postgresql(dot)org> |
Subject: | Re: Binary search in fmgr_isbuiltin() is a bottleneck. |
Date: | 2017-09-28 22:52:28 |
Message-ID: | 15039.1506639148@sss.pgh.pa.us |
Views: | Raw Message | Whole Thread | Download mbox | Resend email |
Thread: | |
Lists: | pgsql-hackers |
Andres Freund <andres(at)anarazel(dot)de> writes:
> I might be worse than you... But anyway, here's a patch doing
> so. Looking at profiles, it turned out that having the integer limits as
> extern variables in a different TU isn't a great idea.
Uh, what? Access to fmgr_nbuiltins shouldn't be part of any critical path
anymore after this change.
> So I moved what
> used to be fmgrtab.c to fmgrtab.h, and included it directly in fmgr.c.
I'm kind of -0.5 on that. I believe part of the argument for having
things set up as they were was to allow external code to access the
fmgr_builtins table (as my speed-test hack earlier today did).
While I'm not sure that anything really is using that API, I do not
believe we'd gain any performance by removing it, so why do so?
We can leave the table and the fmgr_nbuiltins variable completely as-is,
and just add an index table, which fmgr.c could be aware is of size
exactly "FirstBootstrapObjectId" entries.
> Is this roughly what you were thinking of?
I think you need the "no entry" values to be -1; 0 is a valid index
into the fmgr table.
regards, tom lane
From | Date | Subject | |
---|---|---|---|
Next Message | Andres Freund | 2017-09-28 22:56:36 | Re: Binary search in fmgr_isbuiltin() is a bottleneck. |
Previous Message | Andres Freund | 2017-09-28 22:47:20 | Re: Minor codegen silliness in ExecInterpExpr() |