Mailing-List: contact cygwin-developers-help AT sourceware DOT cygnus DOT com; run by ezmlm
Sender: cygwin-developers-owner AT sources DOT redhat DOT com
Message-ID: <3AF700FE.6D01480E@yahoo.com>
Date: Mon, 07 May 2001 16:09:34 -0400
From: Earnie Boyd <earnie_boyd AT yahoo DOT com>
Reply-To: Cygwin Developers <cygwin-developers AT cygwin DOT com>
MIME-Version: 1.0
To: cygwin-developers AT cygwin DOT com
Subject: Re: 1.3.1 *is* slower
References: <20010427160641 DOT A11039 AT redhat DOT com> <3AF6E177 DOT 42A39A64 AT yahoo DOT com> <20010507154351 DOT D29090 AT redhat DOT com>
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit

Christopher Faylor wrote:
> 
> On Mon, May 07, 2001 at 01:55:03PM -0400, Earnie Boyd wrote:
> >Christopher Faylor wrote:
> >>
> >> It is true that 1.3.1 is slower than 1.1.8.  I benchmarked it on my NT 2K
> >> dual processor 300MHZ system.
> >>
> >> A config.status --recheck in my gdb build directory takes a minute 1 second
> >> with 1.3.1, 52 seconds with 1.1.8, and 48 seconds with 1.1.5.
> >>
> >> Not a very encouraging trend.
> >>
> >> Anybody have any ideas?  This is usually a sign of an extra disk open somewhere
> >> but I don't see anything obvious.
> >>
> >
> >Hey Chris, what do your timings give you now?  Just curious if my
> >strrchr patch improved it.
> 
> As I mentioned, I'm working on an asm version of strrchr to complement the asm version
> of strchr.  The timings show something like a 5 - 6% improvement with the asm version.
> I haven't tried with the newlib version.
> 
> I'm very glad that you found this bottleneck.
> 

I've actually known about it for some months.  I couldn't do anything
about it due to working environment restrictions.  Now that has changed
and you'll probably see more patches from me.

Earnie.

_________________________________________________________
Do You Yahoo!?
Get your free @yahoo.com address at http://mail.yahoo.com