Horton

Re: Performance improvements for src/port/snprintf.c

Tom Lane <tgl@sss.pgh.pa.us>

View thread View on postgr.es
From: Tom Lane <tgl@sss.pgh.pa.us>
To: Andres Freund <andres@anarazel.de>
Cc: pgsql-hackers@lists.postgresql.org, Thomas Munro <thomas.munro@enterprisedb.com>, Andrew Gierth <andrew@tao11.riddles.org.uk>, Alexander Kuzmenkov <a.kuzmenkov@postgrespro.ru>
Date: 2018-10-03T16:22:13Z
Lists: pgsql-hackers 
Andres Freund <andres@anarazel.de> writes:
> On 2018-10-03 12:07:32 -0400, Tom Lane wrote:
>> [ scratches head ... ]  How would that work?  Seems like it necessarily
>> adds a strlen() call to whatever we'd be doing otherwise.  palloc isn't
>> going to be any faster just from asking it for slightly fewer bytes.
>> I think there might be something wrong with your test scenario ...
>> or there's more noise in the numbers than you thought.

> I guess the difference is that we're more likely to find reusable chunks
> in aset.c and/or need fewer OS allocations.  As the memory is going to
> be touched again very shortly afterwards, the cache effects probably are
> neglegible.

> The strlen definitely shows up in profiles, it just seems to save at
> least as much as it costs.

> Doesn't strike me as THAT odd?

What it strikes me as is excessively dependent on one particular test
scenario.  I don't mind optimizations that are tradeoffs between
well-understood costs, but this smells like handwaving that's going to
lose as much or more often than winning, once it hits the real world.

			regards, tom lane

Commits

  1. Improve snprintf.c's handling of NaN, Infinity, and minus zero.

  2. Rationalize snprintf.c's handling of "ll" formats.

  3. Provide fast path in snprintf.c for conversion specs that are just "%s".

  4. Make assorted performance improvements in snprintf.c.

  5. Set snprintf.c's maximum number of NL arguments to be 31.

  6. Always use our own versions of *printf().