Message-Id: <199804221431.JAA01744@mail1.kcnet.com>
Comments: Authenticated sender is <edgee@cyberpass.net>
From: "Ed G." <edgee@cyberpass.net>
To: pete@dunnington.u-net.com (Pete Turnbull)
Date: Wed, 22 Apr 1998 09:31:05 -0400
MIME-Version: 1.0
Content-type: text/plain; charset=US-ASCII
Content-transfer-encoding: 7BIT
Subject: Re: Floating Point-How Important
Reply-to: edgee@cyberpass.net
CC: pups@minnie.cs.adfa.oz.au
Sender: owner-pups@minnie.cs.adfa.oz.au
Precedence: bulk

> Sounds reasonable to me, but rather than worry about the sign of the data, I'd
> just decode the instruction far enough to see how much data followed, and skip
> over it.

Your suggestion, of course, is emminently reasonable.  Sometimes it's 
easier to argue about an old approach than think up a new one.

I'm looking at page 2 of Bob's pdp11_cpu.c source which has a nice 
summary of the various instruction formats for the PDP-11.  The 
summary includes octal ranges for double, single operand instructions etc.

How about this approach to determine how many words to skip when 
scanning for floating point op codes:

Assuming Bob's summary is complete, my program could classify
instructions as either double, single or no operand based on the
range they fall in. No operand cases could be discarded.  Single and
double operands would have to be further decoded to see if they use
any of the PC addressing modes (immediate, absolute, relative, and
relative deferred) and the appropriate number of words skipped.

Would this work?  Have I forgotten anything important?

Ed