Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!usc!apple!portal!cup.portal.com!Jake-S
From: Jake-S@cup.portal.com (Jake G Schwartz)
Newsgroups: comp.sys.handhelds
Subject: Re: Notes on Internals
Message-ID: <42437@cup.portal.com>
Date: 18 May 91 01:51:29 GMT
References: <282dbbf8:3077comp.sys.handhelds@hpcvbbs.UUCP>
Organization: The Portal System (TM)
Lines: 25

Scotty Thompson comments on experiences converting user code to internal
RPL:

> Again, was it worth it?  NO.  I see no super-noticable speed increases
> when using the internal routines or just plain vanilla RPL (what do you
> think I prefer now?).  Maybe 5-10% increase.

> It would appear to me that M/L is the only way to really speed-up
> programs, that internal routines should only be called when a function is
> available internally not otherwise accessible or for simple header
> routines, and that I will never do that again...

That's funny, because when I converted some of my stuff, not only was the
resulting code HALF as large, it ran FIVE TIMES as fast. I manipulated many
numerical integers as System Binaries as opposed to reals, and utilized
the True/False "externals" in place of the 1/0 reals as well. In addition,
single-stepping the stuff was okay when using Rick Grevelle's HACKIT 
material, since you can convert any token in stack level one immediately
to it's ASCII equivalent with ->ASCI and back again with ASCI->. All the
speed increase seems to be gained when using those internal data types.

And machine code would probably have speeded it up perhaps another five
times, but manual stack manipulation is a bit too much for me yet.

Jake Schwartz