g++ 3.3 I/O Performance

Torsten Reuss t.reuss@gmx.net
Tue Dec 16 20:32:00 GMT 2003


On Mon, 2003-12-15 at 22:19, lrtaylor@micron.com wrote:
> Which version of GCC are you using (that is, 3.3.0, 3.3.1 or 3.3.2)?  From what I understand, there was a bug report for that and it was supposed to have been fixed in 3.3.2, or is slated to be fixed soon.  It had to do with the I/O implementation calling a slow system function (something like snprintf or something).  Perhaps someone knows better than I, but I know it's definitely not fixed in 3.3.1 or earlier.
> 
> Cheers,
> Lyle

I tested today on a different machine (running debian sarge) the 2.95.4
against 3.3.2 performance. Unfortunately the results were the same, or
let's say at least quite similar.

Without I/O the g++-3.3.2 compiled program ran 10% faster
With lot's of writing to cout, 3.3.2 ran 50% slower

That's a better result than the 80% slower I'm gettin on my other
machine but that could be also because its different hardware
(PentiumIII against Athlon-XP). Anyways, I find 50% drop in performance
hard to swallow.

Does anybody know if this issue is discussed for gcc development?

Regards,
				Torsten




> -----Original Message-----
> From: gcc-help-owner@gcc.gnu.org [mailto:gcc-help-owner@gcc.gnu.org]On
> Behalf Of Torsten Reuss
> Sent: Monday, December 15, 2003 2:08 PM
> To: gcc-help@gcc.gnu.org
> Subject: g++ 3.3 I/O Performance
> 
> 
> Hi,
> 
> I'm wondering if anybody can help me on why g++-3.3 is about 80% slower
> than g++-2.95.4 for an I/O intensive program I'm working on.
> 
> Now, before I go into the details, that's 80% slower with a call to
> ios::sync_with_stdio(false) inserted already. Without that it was 5
> times slower.
> 



More information about the Gcc-help mailing list