Date: Tue, 21 May 1996 08:00:20 +0200 (IST)
From: Eli Zaretskii <eliz AT is DOT elta DOT co DOT il>
To: Austin Appleby <applea AT weiss DOT che DOT utexas DOT edu>
Cc: djgpp AT delorie DOT com
Subject: Re: Making the nearptr hacks use longs.
In-Reply-To: <4nqe0f$som@geraldo.cc.utexas.edu>
Message-Id: <Pine.SUN.3.91.960521075640.13486B@is>
Mime-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII


On 20 May 1996, Austin Appleby wrote:

> it'll be even speedier... Basically, the idea is to set videoptr to a 
> long pointer instead of a char pointer, so that the thing will optimize 
> down to some resemblance of a movsd op instead of a movsb, thus a 4x 
> improvement. But for the life of me I don't know how to get the thing to 
> work...

`dosmemputl' (and `movedata') already use 32-bit movsd, so I don't see 
how you can get any real speed-up here.  The only situation where nearptr 
is faster is when you operate on single pixels or small groups of 
pixels.  If you need to move large buffers, `movedata' is as fast as 
nearptr, but much more safe (as it doesn't disable memory protection), 
and also works on any DPMI platform.