Path: utzoo!attcan!utgpu!jarvis.csri.toronto.edu!rutgers!sun-barr!apple!ames!vsi1!altos86!elxsi!beatnix!corbett
From: corbett@beatnix.UUCP (Bob Corbett)
Newsgroups: comp.lang.fortran
Subject: Fortran 8x:  pointers and optimization
Message-ID: <2773@elxsi.UUCP>
Date: 26 Jun 89 22:31:02 GMT
Sender: news@elxsi.UUCP
Reply-To: uunet!elxsi!corbett (Bob Corbett)
Distribution: usa
Organization: ELXSI, San Jose
Lines: 44


     There have been articles posted to this group claiming that pointers
make optimization impossible and that the SET RANGE and IDENTIFY statements
would not affect optimization.  Those claims are simply wrong.

     Adding pointers will make optimization more difficult.  Adding
functionality to a language almost always makes optimization more difficult.
Adding pointers will make alias detection more difficult.  However, there
are good and well-publicized algorithms for performing alias detection in the
presence of pointers.  See, for example, pages 647-653 of the book Compilers:
Principles, Techniques, and Tools by Aho, Sethi, and Ullman.  Optimization of
codes that do not use pointers should be unaffected beyond the need to
detect that pointers are not used.

     Many languages that provide pointers do not allow pointers to reference
named variables.  Such a restriction makes alias detection easier.  I would
be willing to accept that restriction in Fortran 8x.

     The claim that the SET RANGE and IDENTIFY statements do not affect
optimization is bizarre.  I cannot understand how anyone could think that
the IDENTIFY statement does not create an alias when the element named in the
IDENTIFY statement is required to have the ALIAS attribute.  The aliasing
problem created by the SET RANGE statement is subtler.  The SET RANGE
statement makes it more difficult to tell when references to array elements
overlap.

     I suspect that some of those who think SET RANGE and IDENTIFY are more
benign than pointers missed the fact that SET RANGE and IDENTIFY are
executable statements, not declarations.  For example, the code fragment

       IF (P) THEN
	   IDENTIFY (X = A)
       ELSE
	   IDENTIFY (X = B)
       END IF

       X = Z

was allowed.  Thus, adding pointers does not make optimization harder than
the IDENTIFY statement would.  However, pointers offer more functionality
than the IDENTITY statement.

						 Yours very truly,
						 Bob Corbett