NetNews Usenet Archive 1993 #1/NN_1993_1.iso/spool/comp/lang/forth/3753

Text File | 1993-01-04 | 1.8 KB | 41 lines

Newsgroups: comp.lang.forth
Path: sparky!uunet!email!mips.complang.tuwien.ac.at!anton
From: [e-mail] (Anton Ertl)
Subject: Re: Hardware ONLY issues
Message-ID: <[e-mail]>
Sender: [e-mail]
Nntp-Posting-Host: mips.complang.tuwien.ac.at
Organization: Institut fuer Computersprachen, Technische Universitaet Wien
References:  <4186.UUL1.3#[e-mail]>
Date: Mon, 4 Jan 1993 14:45:12 GMT
Lines: 28

In article <4186.UUL1.3#[e-mail]>, [e-mail] (ForthNet articles from GEnie) writes:
|> Category 9,  Topic 2
|> Message 126       Tue Dec 22, 1992
|> G.LEFAVE [Gene]              at 11:23 EST

|> One question.  Why would keeping the address of NEXT in a register on a 68040
|> be considered unrealistic?  Since the 68040 has more registers then Forth can
|> realistically use, I would think that was desirable.  It would also make the
|> 68040, 486 comparison even more interesting.

In the benchmarks all occurences of NEXT are inlined. In the
subroutine threading benchmark the address of the word "Next" is taken.
That work just performs a NEXT and could also be called "Noop".

I a real Forth program we do not just use two different words, but
hundresds. While the Forth system could hold the addresses of the,
say, three most frequently used Forth words in registers, this would
result in about 20% of calls through registers. So, the typical case
is calling with immediate address, and doing this unoptimization makes
the benchmark more realistic.

BTW, I was surprised to find out that the unoptimized version is
slightly faster (on the 68040).

- anton
--
M. Anton Ertl                    Some things have to be seen to be believed
[e-mail] Most things have to be believed to be seen