Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.2 9/18/84; site looking.UUCP
Path: utzoo!watmath!looking!brad
From: brad@looking.UUCP (Brad Templeton)
Newsgroups: net.micro
Subject: Re: Re: 386 Family Products (8086 sucks
Message-ID: <466@looking.UUCP>
Date: Wed, 11-Dec-85 04:20:42 EST
Article-I.D.: looking.466
Posted: Wed Dec 11 04:20:42 1985
Date-Received: Thu, 12-Dec-85 04:54:23 EST
References: <133@daisy.UUCP> <97800013@ima.UUCP> <464@looking.UUCP> <879@ecsvax.UUCP>
Reply-To: brad@looking.UUCP (Brad Templeton)
Organization: Looking Glass Software Ltd. Waterloo, Ontario
Lines: 40
Summary: 

In article <879@ecsvax.UUCP> hes@ecsvax.UUCP (Henry Schaffer) writes:
>
>   I would say that many programs use single objects larger than 64k.  (How big
>a double precision floating point matrix fits in 64k?)  Most programs are
>(or should be) sufficiently modular that 64k code segments are not much of a
>penalty, but in scientific/statistical/simulation computations it is quite
>common to have data objects >64k.  (On an 8088/8086 allowing the transparent-
>to-the-user use of data objects>64k appears to have an unavoidable and 
>significant performance degradation.)
>   There may be some language bias here.  Traditionally C has not seen heavy
>use in the number-crunching community and so large data objects may not have
>been common.  The number-crunching community has primarily used FORTRAN, and
>that's where you might see things like:
>REAL*8  XPY(2500), XPX(31375)
>(which is a quote from one of my programs - where we sometimes have to
>increase the dimensions to handle a larger than usual data set!)
>
>--henry schaffer

Yes, numerical applications do generate such arrays.  Although I suspect
you would be hard presed to find single *vectors* longer than 64K, so I
think you might be able to get around such troubles if you wanted a small
amount of effort.

But all this aside, is the 8086 the processor to run these heavy floating
point applications?  Certainly not without the 8087, and if you're that
keen on speed but insist on 8086 you should have gone to 286, where it is
actually possible to have segment 2 64K beyond segment 1, allowing full
sized objects with little work by the compiler.  I don't know if anybody
has done this, though.

I don't think most of the bitching is from numerical analysts, somehow.
Besides, several compilers now offer arrays larger than 64K, and the
Microsoft C compiler even allows individual objects to be huge in a program
that is not otherwise bogged down with the complex pointer arithmetic this
involves.


-- 
Brad Templeton, Looking Glass Software Ltd. - Waterloo, Ontario 519/884-7473