Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!sdd.hp.com!zaphod.mps.ohio-state.edu!mips!daver!bungi.com!news
From: ian@sibyl.eleceng.ua.oz.au
Newsgroups: comp.sys.nsc.32k
Subject: Re: strtol.c
Message-ID: <9103221637.24185@munnari.oz.au>
Date: 21 Mar 91 20:23:47 GMT
References: <<9103191744.AA06048@hplwbc.hpl.hp.com>>
Sender: news@daver.bungi.com
Lines: 27
Approved: news@daver.bungi.com

Bruce Culbertson writes:
 > >Has anyone found away to compile /usr/src/lib/ansi/strtol.c ??
 > >ie.:
 > >   as: line 124, value too large for displacement
 > >
 > 
 > GCC has a predicate which answers the question: can a given expression
 > be calculated using an some addressing mode of the CPU?  When the
 > answer is yes, GCC emits a single instruction; when it is no, GCC
 > emits instructions to simplify the expression and then calls the
 > predicate again.

Actually, this is not the problem (it was where I looked first
though). As far as I can tell, this predicate works fine. The problem
is in the code for the add3si (?)  pseudo instruction.  I have
supplied a patch to fix it and it ended up in the 1.38 distribution.
Unfortunately, I had the idea right but had a case of brain fade when
I actually typed it in. Basically, the ns32k can't do 3 operand
addition in a single instruction except by using the addr instruction
in which case one of the arguments has to be representable as a
displacement and must be in the range [-2^29, 2^29 - 1], if not it
should use two instructions. (Which it didn't do). The brain fade was
to check for the wrong range [-2^30, 2^30 - 1]. I told RMS about my
mistake and it may have been fixed for version 1.39 (which I haven't
had time to examine).

Ian