Path: utzoo!attcan!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!yale!mintaka!ogicse!orstcs!tgd@tesla.CS.ORST.EDU
From: tgd@tesla.CS.ORST.EDU (Tom Dietterich)
Newsgroups: comp.lang.lisp
Subject: Re: Floating point in CL?
Summary: Use compiler tracing facilities
Keywords: lucid, franz
Message-ID: <20643@orstcs.CS.ORST.EDU>
Date: 29 Sep 90 21:35:52 GMT
References: <TIM.90Sep27140541@watt.cstr.ed.ac.uk>
Sender: usenet@orstcs.CS.ORST.EDU
Distribution: comp
Organization: Computer Science Department, Oregon State Univ.
Lines: 44

To write good floating point code, I have found it essential to use
the facilities of the lucid and franz compilers to trace the
compilation process.  In lucid, you can do the following in your
source file:

(eval-when (compile) (compiler-options :show-optimizations t))

(defun ....)

(eval-when (compile) (compiler-options :show-optimizations nil))

When you invoke the production compiler, you will get a very nice
listing showing the type inference of the compiler and any expressions
that it could not fail to optimize.

In franz, the thing to do is include

(defun ....
  (declare (:explain :calls :types))
)

which gives a similar (but not quite as pretty) trace.

--Tom

P.S. In lucid, all floats should be declared single-float, and I find
it essential to call (the ..) quite often.  Short floats are handled
incorrectly (at least in lucid3.0 on sparc architectures).

P.P.S.  Many folks simply define their own versions of +, -, etc that
include calls to the at each point:

For example: (courtesy of Scott Fahlman)

(defmacro *sf (&rest args)
  `(the single-float
	(* ,@(mapcar #'(lambda (x) (list 'the 'single-float x)) args))))


-- 
Thomas G. Dietterich
Department of Computer Science
Dearborn Hall, 303
Oregon State University