Path: utzoo!utgpu!utstat!jarvis.csri.toronto.edu!mailrus!tut.cis.ohio-state.edu!ucbvax!decwrl!sun!pitstop!sundc!seismo!uunet!microsoft!w-colinp
From: w-colinp@microsoft.UUCP (Colin Plumb)
Newsgroups: comp.lang.c
Subject: Re: C optimizer
Message-ID: <667@microsoft.UUCP>
Date: 16 Feb 89 00:37:31 GMT
References: <515@larry.UUCP> <9648@smoke.BRL.MIL> <13134@steinmetz.ge.com> <11767@haddock.ima.isc.com>
Reply-To: w-colinp@microsoft.uucp (Colin Plumb)
Organization: very little
Lines: 33

karl@haddock.ima.isc.com (Karl Heuer) wrote:
> In article <13134@steinmetz.ge.com> davidsen@crdos1.UUCP (bill davidsen) writes:
> >One solution would be to provide a keyword ... which would indicate that a
> >procedure always returns the same value for a set of given arguments.
> 
> If the concept is to be useful, it had better mean "no observable side effects
> AT ALL".

> This is automatic if the "magic" is embedded in the standard
> header files, via a keyword.

Um... Karl, what strange disease has come over your brain to make it suddenly
like adding keywords to C? :-)

This is a perfect example of a good use for #pragma.  In <math.h>, I could
just include:

#pragma functional
double sin(double), cos(double);

And the compiler could take that as a hint to use common subexpression
elimination (a pretty common optimisation) on sin and cos.  A good
compiler already knows that things like + and ~ are purely functional,
and need only generalise the common subexpression and dead-code eliminators
to handle more conventional function calls.

This is also an example of the point of having a compiler ignore an
unrecognised #pragma.  Without this #pragma, the code would compute the
same result, just more wastefully.
-- 
	-Colin (uunet!microsoft!w-colinp)

"Don't listen to me.  I never do."