Mark,
In the specific case of the namestrcpy () function, it will copy a
maximum of 64 bytes, but the length of the source string is unknown. I
would have to think that memcpy () would certainly win if you knew the
source and destination sizes etc. Perhaps there are some places like
that in the code that don't use memcpy () currently?
David
-----Original Message-----
From: mark@mark.mielke.cc [mailto:mark@mark.mielke.cc]
Sent: Wednesday, September 27, 2006 4:27 PM
To: Strong, David
Cc: pgsql-hackers@postgresql.org
Subject: Re: [HACKERS] Faster StrNCpy
On Wed, Sep 27, 2006 at 07:08:05AM -0700, Strong, David wrote:
> We sometimes see TupleDescInitEntry () taking high CPU times via
> OProfile. This does include, amongst a lot of other code, a call to
> namestrcpy () which in turn calls StrNCpy (). Perhaps this is not a
> good candidate right now as a name string is only 64 bytes.
Just wondering - are any of these cases where a memcpy() would work
just as well? Or are you not sure that the source string is at least
64 bytes in length?
memcpy(&target, &source, sizeof(target)); target[sizeof(target)-1] = '\0';
I imagine any extra checking causes processor stalls, or at least for
the branch prediction to fill up? Straight copies might allow for
maximum parallelism? If it's only 64 bytes, on processors such as
Pentium or Athlon, that's 2 or 4 cache lines, and writes are always
performed as cache lines.
I haven't seen the code that you and Tom are looking at to tell
whether it is safe to do this or not.
Cheers,
mark
--
mark@mielke.cc / markm@ncf.ca / markm@nortel.com
__________________________
. . _ ._ . . .__ . . ._. .__ . . . .__ | Neighbourhood
Coder
|\/| |_| |_| |/ |_ |\/| | |_ | |/ |_ |
| | | | | \ | \ |__ . | | .|. |__ |__ | \ |__ | Ottawa, Ontario,
Canada
One ring to rule them all, one ring to find them, one ring to bring
them all and in the darkness bind them...
http://mark.mielke.cc/
