> On 09/21/2017 09:41 PM, Andres Freund wrote:
> > On 2017-09-21 09:30:31 -0400, Tom Lane wrote:
> >> Andrew Dunstan <andrew.dunstan@2ndquadrant.com> writes:
> >>> The speed of memset is hardly going to be the dominating factor in a
> >>> 'CREATE DATABASE' command, so we could certainly afford to change to
> >>> plain memset calls here.
> >> Another thought is that it may be time for our decennial debate about
> >> whether MemSet is worth the electrons it's printed on. I continue to
> >> think that any modern compiler+libc ought to do an equivalent or better
> >> optimization given just a plain memset(). If that seems to be true
> >> for recent MSVC, we could consider putting an #if into c.h to define
> >> MemSet as just memset for MSVC. Maybe later that could be extended
> >> to other compilers.
> > +many. glibc's memset is nearly an order of magnitude faster than MemSet
> > on my machine for medium+ length copies, and still 3-4 four times ~100
> > bytes. And both gcc and clang optimize way the memcpy entirely when the
> > length is a fixed short length.
> Perhaps we need some sort of small benchmark program that we can run on
> a representative set of platforms?
I personally don't think that's needed here, given how incredibly widely
used memset is in our codebase.
> I presume if you can make that assertion you already have something
> along those lines?
Not really. I just replaced memset with MemSetAligned in a bunch of
places in the code and looked at profiles.
Greetings,
Andres Freund
--
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers
