Message-ID:	<378F6F22.36FA@ns.sympatico.ca>
From:	Klaas <klaas AT ns DOT sympatico DOT ca>
Organization:	N/A
X-Mailer:	Mozilla 3.04 (Win95; I)
MIME-Version:	1.0
Newsgroups:	comp.os.msdos.djgpp
Subject:	Re: 386 SX versus DX - 'int' datatransfer
References:	<MPG DOT 11f97084d7231b29989686 AT news DOT cyberway DOT com DOT sg>
Lines:	25
Date:	Fri, 16 Jul 1999 14:42:58 -0300
NNTP-Posting-Host:	142.177.41.102
X-Trace:	sapphire.mtt.net 932147335 142.177.41.102 (Fri, 16 Jul 1999 14:48:55 ADT)
NNTP-Posting-Date:	Fri, 16 Jul 1999 14:48:55 ADT
To:	djgpp AT delorie DOT com
DJ-Gateway:	from newsgroup comp.os.msdos.djgpp
Reply-To:	djgpp AT delorie DOT com

Rob Kramer wrote:
> 
> Hi all,
> 
> One of my applications (involving a GIF-decoder) I recently tried on a
> 386 SX machine. The decoder is terribly slow in that case. The original
> pre-DJGPP obsolete 16-bit version of the application is way faster. (I'm
> not sure whether the 32-bit version is faster than the 16-bit version on
> a 386 DX)
> 
> The 16-bit version uses shorts as working variables in the decoder, the
> 32-bit version uses ints. Could the big difference between running on a
> SX and a DX be caused by the fact that 32-bit transfers are not a nice
> thing on a SX databus?
> 
> Then would it help if I change back to shorts, or will this harm the
> performance of the decoder on 32 bits databus machines. Note that the
> decoder doesn't actually need ints, but I thought ints were friendlier to
> a Pentium architecture. That's what it normally runs on, but I have to
> support 386SX too :(

Do you use any floating point math? The SX would have to emulate, which
wuold be considerably slower.

-Mike

- Raw text -