Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!sdd.hp.com!samsung!uunet!bywater!arnor!prener!prener
From: prener@watson.ibm.com (Dan Prener)
Newsgroups: comp.arch
Subject: Re: IEEE arithmetic (Goldberg paper)
Message-ID: <1991Jun22.003029.16748@watson.ibm.com>
Date: 22 Jun 91 00:30:29 GMT
References: <9106190124.AA27410@ucbvax.Berkeley.EDU>
Sender: news@watson.ibm.com (NNTP News Poster)
Reply-To: prener@prener.watson.ibm.com (Dan Prener)
Organization: IBM T.J. Watson Research Center
Lines: 74
Nntp-Posting-Host: prener

In article <9106190124.AA27410@ucbvax.Berkeley.EDU>, jbs@WATSON.IBM.COM writes:
|> 
|>          I said:
|> >I think that if a (legal) program behaves differently when compiled with
|> >and without optimization that then the compiler is broken.  Perhaps you
|> >disagree.
|>          Malcolm said:
|>       B = 5.0
|>       IF ( ( B .GT. 3.0 ) .OR. MYTEST(A) )
|>           PRINT some diagnotic messages
|>       LOGICAL FUNCTION MYTEST(X)
|>       REAL X
|>       MYTEST = some boolean work
|>       PRINT a few diagnostics on the screen
|>       RETURN
|>       END
|>          I modified this to:
|>       LOGICAL*4 MYTEST
|>       B = 5.0
|>       IF ( ( B .GT. 3.0 ) .OR. MYTEST(B) )THEN
|>        WRITE(6,*)' MAIN'
|>       ENDIF
|>       END
|>       LOGICAL FUNCTION MYTEST(X)
|>       REAL X
|>       WRITE(6,*)' MYTEST'
|>       MYTEST = .TRUE.
|>       RETURN
|>       END
|>          I checked the behavior with the IBM VS Fortran and Xlf compi-
|> lers.  In both cases the behavior was not changed by optimization.  The
|> behavior differs between the two compilers but that is a different is-
|> sue.  So this example will not convince me that I'm wrong.  (Actually
|> I later softened my original statement to say it is undesirable for
|> optimization to affect a program's behavior.)
|>                        James B. Shearer

As immodest as it might be for me to agree with my colleague, I must
agree with James Shearer.  Indeed, the spirit of FORTRAN is, overwhelmingly,
that a program whose behavior is changed by optimization is not a legal
program.  In particular, I quote from the FORTRAN '77 standard

   It is not necessary for a processor to evaluate all of the operands
   of an expression if the value of the expression can be determined
   otherwise.  This principle is most often applicable to logical
   expressions, but it applies to all expressions.  For example, in
   evaluating the logical expression    X  .GT.  Y  .OR.  L(Z)
   where X, Y, and Z are real, and L is a logical function, the
   function reference L(Z) need not be evaluated if X is greater than
   Y.

to show that the compiler may do it either way.  And I cite another
passage from the same standard

   For example the statements

          A(I) = F(I)
          Y = G(X) + X

   are prohibited if the reference to F defines I or the reference to G
   defines X.

which suggests that programs that admit more than one correct interpretation
are illegal.  

I would say that, since the program(s) offered above are, according to
the language standard, non-deterministic, then they are illegal.

Note that optimization has nothing to do with this.  According to the
standard, a compiler could correctly produce a program which would call
the function MYTEST or not depending upon the time of day at which it
was run.
-- 
                                   Dan Prener (prener @ watson.ibm.com)