Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!swrinde!zaphod.mps.ohio-state.edu!wuarchive!mit-eddie!uw-beaver!ubc-cs!alberta!ncc!idacom!andrew
From: andrew@idacom.uucp (Andrew Scott)
Newsgroups: comp.lang.forth
Subject: Re: Postfixer FORTH
Message-ID: <1990Aug24.221345.14475@idacom.uucp>
Date: 24 Aug 90 22:13:45 GMT
References: <1599.UUL1.3#5129@willett.pgh.pa.us>
Organization: IDACOM, a division of Hewlett-Packard
Lines: 52

Doug Philips writes:
> Oops.  I forgot that.  I guess the only difficulty might be if you were
> using a subroutine threaded system with an optimizer in it, such as
> Andrew Scott describes in the article "Extensible Optimizing Compiler"
> (Forth Dimensions, Volume XII, Number 2).  I'm not sure what would happen
> to a label on an instruction that "disappeared".  Maybe the label would
> just inhibit optimization "across it"?  Probably just an academic question.

I hope I understand the issue, as I'm jumping into this discussion thread a
bit late, but you *did* refer to my article.  :-)

The question of preventing optimizing across labels is not so academic.  If
you think of a branch destination like a label, the optimizer queue must be
"flushed" before the branch is resolved.

For example, suppose the phrase DUP >R was optimized to a single instruction.
The following code illustrates the problem:

: FOO   ( x \ y -- )
    HORIZ? IF  OVER  ELSE  DUP  THEN     ( x coord for horiz., y for vertical )
    >R ...
;

THEN doesn't really compile any code - it only resolves a branch.  There is no
code between DUP and >R, but they must not be combined.

My solution was to compile a "do nothing" word that has an optimization rule
that "does nothing" (i.e. doesn't actually lay down code):

: NOTHING ;

SEQ:  NOTHING  IS:  ( nothing ) ;

The optimization rule does in fact complete any pending sequences, as NOTHING
never appears in any other rule.

The definition of THEN becomes:

: THEN     ' NOTHING COMPILE-TOKEN  >RESOLVE ; IMMEDIATE

(Our tick is state smart.  Here's a vote for the proposal for COMPILE-TOKEN !)


I force a "compilation" of NOTHING whenever I have to stop optimization.  In
your discussion, the same method could be used whenever internal labels need
to be generated.

I hope this all made some sense...
-- 
Andrew Scott	| mail:		andrew@idacom.uucp
		| - or -	{att, watmath, ubc-cs}!alberta!idacom!andrew
		| - or -	uunet!myrias!aunro!idacom!andrew