Message-ID:	<3276C71E.3EFE@gbrmpa.gov.au>
Date:	Wed, 30 Oct 1996 11:10:22 +0800
From:	Leath Muller <leathm AT gbrmpa DOT gov DOT au>
Reply-To:	leathm AT gbrmpa DOT gov DOT au
Organization:	Great Barrier Reef Marine Park Authority
MIME-Version:	1.0
To:	"William D. Kirby" <wdkirby AT ix DOT netcom DOT com>
CC:	djgpp AT delorie DOT com
Subject:	Re: Speed Optimization is getting worse with V2.01
References:	<199610291204 DOT EAA29378 AT dfw-ix4 DOT ix DOT netcom DOT com>

> I do not like to use Borland C++ except for the good IDE.

Its a good IDE, I have to admit... :)

> For 10000 iterations of a 1024 real sample FFT (Fat Fourier Transform) using
> a 486/33 the average run time per iteration has been as follows for DJGPP
> v1.0     18.3ms
> v2.0     20.7ms
> v2.01    22.0ms under DOS with -m486, 20.1ms with no -m486 option
 
> Using Borland 4.5 with PowerPack for 32 bit DPMI
> the run time per iteration is 19.4 under DOS

Just a quick question: how did you time this? Its just that just about
everything I ever coded on Borland ran a _lot_ slower than DJGPP 2.0
(which I use now...) Do you have the source code so I can test the
differences
on my Pentium? Although I _REALLY_ _DONT_ want to reinstall Borland...
:)
 
> I have tired about every 486 and speed optimization option available, but with
> no improvements in run times. It appears the -m486 option hurts the speed,
> and I have seen this reported in other postings.

I have to admit, I don't use the 486 switch, because all my CPU
intensive
stuff doesn't need it, and/or is coded in asm...what optimization are
you
using for the Borland code? I used to optimize for pentium code, and it
still ran slower (visibly...) than 386 code made by DJGPP...And what
optimizations are you using on DJGPP? O2, O3??? I have found the -O 
optimizing switch is a _lot_ more effective than the -m486 switch...
 
> It appears the recent improvements in DJGPP degrading the performance of
> the executable in terms of speed.

I never used 1.x (well, I did, but it was a lot of hassle at the time
as I knew near on nothing about the PC... :) and now I use 2.0. If 2.1
is in fact that much slower, I probably wont install it...but what I
would
_really_ like is some code to show this.
 
> Bill
> William D. Kirby              Voice: (703) 273-0005
> Consultant                   Email: wdkirby AT ix DOT netcom DOT com
> 3527 Cornell Road
> Fairfax, VA 22030-1813

- Raw text -