Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.2 9/18/84; site gitpyr.UUCP
Path: utzoo!watmath!clyde!bonnie!akgua!gatech!gitpyr!robert
From: robert@gitpyr.UUCP (Robert Viduya)
Newsgroups: net.micro.att
Subject: Re: What's wrong with this program (3b2)?
Message-ID: <493@gitpyr.UUCP>
Date: Tue, 25-Jun-85 03:08:00 EDT
Article-I.D.: gitpyr.493
Posted: Tue Jun 25 03:08:00 1985
Date-Received: Wed, 26-Jun-85 06:25:10 EDT
References: <104@gwsd.UUCP> <476@gitpyr.UUCP> <339@cuae2.UUCP> <199@hadron.UUCP>
Organization: Georgia Tech, Atlanta
Lines: 36

> 
> I think a major point is, if the 3B2  n e e d s  the '-f' flag, it
> should automagically be included at compile time.  We did this for
> PDP-11's without floating point:  there was a compile-time option
> (when compiling the C compiler!) to determine whether -f was a
> compiler option or requirement.  (We == the UNIX community;
> reference == V5/V6 ff -- pre-COFF days.)
> 
> 	Joe Yao		hadron!jsdy@seismo.{ARPA,UUCP}

I disagree with this, since using the -f increases the size of the program,
*regardless* of whether you use floating point numbers in the program or
not.  I was doing some futzing around earlier this evening with the 3B2
and discovered that a (non-floating point) program compiled with -f can
sometimes be up to 10K larger than it would be compiled without (I used ls(1)
to get my numbers, not size(1)).

From what I understand, floating point simulation is done by trapping the
"illegal instructions" as the program executes and calling a subroutine
to handle the instruction.  The assembler does not generate calls to the
floating point subroutines.  This means that the loader cannot tell if the
floating point library is needed or not (unless it examines each instruction,
looking for a floating point one) and because of that it will always load
in the floating point library if the -f option was given.

I've heard rumors that AT&T is coming out with, or already sells, a new
3B2 with the WE32100 (did I get that number right?) processor which *does*
support hardware floating point.

			robert
-- 
Robert Viduya							01111000
Georgia Institute of Technology

...!{akgua,allegra,amd,hplabs,ihnp4,masscomp,ut-ngp}!gatech!gitpyr!robert
...!{rlgvax,sb1,uf-cgrl,unmvax,ut-sally}!gatech!gitpyr!robert