Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.2 9/18/84; site dataioDataio.UUCP
Path: utzoo!watmath!clyde!burl!ulysses!mhuxr!mhuxt!houxm!vax135!cornell!uw-beaver!uw-june!entropy!dataio!bright
From: bright@dataioDataio.UUCP (Walter Bright)
Newsgroups: net.lang.c
Subject: Re: Lisp to C conversion
Message-ID: <876@dataioDataio.UUCP>
Date: Thu, 12-Dec-85 13:04:31 EST
Article-I.D.: dataioDa.876
Posted: Thu Dec 12 13:04:31 1985
Date-Received: Fri, 13-Dec-85 20:38:11 EST
Reply-To: bright@dataio.UUCP (Walter Bright
Organization: Data I/O Corp., Redmond WA
Lines: 25

In article <124000003@ima.UUCP> johnl@ima.UUCP writes:
>It's not hard for a
>compiler to tell that some piece of a program has no inputs, and once it's
>figured that out, it can compute the result of that piece at compile time
>and substitute its result.

Programs that have no inputs tend to fall into two groups:

	o Benchmarks.

	o Programs that are to be run only once per recompilation.

In either case, it doesn't seem worthwhile to compute the program
output at compile-time.

As for parts of programs that have no inputs, or the inputs are always
known, this is a major focus of global flow optimization techniques.
Unfortunately, I have yet to see one powerful enough that worked
across loops (I'd love to see one that could optimize the notorious
sieve benchmark down to printf("1899 primes\n");).

I've also seen some work on attempting to determine when run-time
array bounds checking could be safely eliminated. Of course, the
corollary is that out-of-bounds detection could be accomplished
at compile-time... but I'm dreaming on.