Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!tut.cis.ohio-state.edu!cs.utexas.edu!hellgate.utah.edu!cdr.utah.edu!moore
From: moore%cdr.utah.edu@cs.utah.edu (Tim Moore)
Newsgroups: gnu.gcc
Subject: Re: short jumps
Keywords: port
Message-ID: <1990Feb2.094716.12223@hellgate.utah.edu>
Date: 2 Feb 90 16:47:16 GMT
References: <1049@tuvie>
Organization: University of Utah CS Dept
Lines: 37

Newsgroups: gnu.gcc
Subject: Re: short jumps
Summary: 
Expires: 
References: <1049@tuvie>
Sender: 
Followup-To: 
Distribution: 
Organization: University of Utah CS Dept
Keywords: port

In article <1049@tuvie> inst182@tuvie (Inst.f.Techn.Informatik) writes:
>I want to make a port of gcc to a new processor that has short relative
>jumps for up to +-128 instructions. Now my problem is the following:

>This jump has completely different semantics than a far jump. How can I
>check if I can use the short jump instead of a far jump. I cannot delay
>the decision till assembly time, so I *MUST* have gcc emit the right 
>instruction. *ANY* hints will be appreciated.

It's not an easy problem. You can't make this decision before assembly time 
unless your compiler knows the length of each instruction and how that length
depends on the instruction being "near" or "far."

A good paper that describes this problem and an algorithm to solve it is
"Assembling Code for Machines with Span-Dependent Instructions",
Thomas G. Szymanski, CACM, 21(4), April 1978. I implemented this
algorithm for Utah Common Lisp's assembler, and I believe that gas
uses a similar algorithm.

>					thanx in advance,
>						mike


Tim Moore                     moore@cs.utah.edu {bellcore,hplabs}!utah-cs!moore
"Ah, youth. Ah, statute of limitations."
		-John Waters