From:	sandmann AT clio DOT rice DOT edu (Charles Sandmann)
Message-Id:	<10207201815.AA26206@clio.rice.edu>
Subject:	Re: DJGPP 2.04 performance & size
To:	djgpp-workers AT delorie DOT com
Date:	Sat, 20 Jul 2002 13:15:50 -0500 (CDT)
Cc:	rich AT phekda DOT freeserve DOT co DOT uk
In-Reply-To:	<3D394720.B1DECD07@phekda.freeserve.co.uk> from "Richard Dawe" at Jul 20, 2002 12:18:56 PM
X-Mailer:	ELM [version 2.5 PL2]
Mime-Version:	1.0
Reply-To:	djgpp-workers AT delorie DOT com
Errors-To:	nobody AT delorie DOT com
X-Mailing-List:	djgpp-workers AT delorie DOT com
X-Unsubscribes-To:	listserv AT delorie DOT com

> Could someone else test on Windows 2000 or XP please? You can find my test
> results and executables here:
>     http://www.phekda.freeserve.co.uk/richdawe/djgpp/workers/
> I tested on Windows '98 SE.

(I've edited the "user" and "sys" times displayed, since they seem to be
completely meaningless).

Tests on Windows 2000, 22Gb FAT32 partition (everything in cache - machine 
stays up > 6 months at a time and has 768Mb). 

sh-2.04$ time ls-head -lR /djgpp > /dev/null
real    0m25.760s

sh-2.04$ time ls-203 -lR /djgpp > /dev/null
real    0m14.390s

sh-2.04$ time /djgpp/bin/ls-v203 -lR /djgpp > /dev/null
real    0m11.420s                         (no profiling)

sh-2.04$ time /djgpp/bin/lsv2b4 -lR /djgpp > /dev/null
real    0m4.940s                          (really old image)

> ls from Fileutils 4.1 built against v2_03_1:
>     real    0m50.700s

> ls from Fileutils 4.1 built against HEAD:
>     real    0m59.650s

So you see a 18% slow down vs 2.03; I see a 79% (on other directories such
as /djgpp/src I saw up to 100% slow down). So HEAD seems Win2K unfriendly.

Looking at the profile output files, almost all the time is spent in 
__dpmi_int and mcount (89% for 203 and 91% for head).  But since we don't
have a gprof call in dpmi_int you can't see how many times it's called or
from where.  Modify a copy, add this call and it might give some hints
on what's happening.

- Raw text -