Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!uunet!samsung!munnari.oz.au!uhccux!tholen
From: tholen@uhccux.uhcc.Hawaii.Edu (David Tholen)
Newsgroups: comp.lang.fortran
Subject: Re: Array Notation
Message-ID: <10818@uhccux.uhcc.Hawaii.Edu>
Date: 3 Jan 91 22:34:18 GMT
Distribution: comp.lang.fortran
Organization: University of Hawaii
Lines: 34

Presley Smith writes:

> Array notation provides a simplified way to specify array operations
> in many cases, but may, in fact, decrease the performance of the program.

Microsoft FORTRAN version 5.0 also supports array notation, and on page 41
of the reference manual is the following note:

   In processing array expressions, the compiler may generate a less
   efficient sequence of machine instructions than it would if the
   arrays were processed in a conventional DO loop.  If execution
   speed is critical, it may be more efficient to handle arrays
   element-by-element.

So the problem isn't limited to the Cray compiler.  What I'd like to know are
the kind of situations in which array notation slows things down.  Something
as simple as
      C = A + B
where all three are conforming arrays, can be done in two ways:  with
sequential memory accesses or with nonsequential memory accesses.  The former
should be faster than the latter, especially if nonsequential memory access
results in a lot of paging to disk!  With DO loop syntax, it's easy to handle
the array subscripts in the wrong order, thereby causing nonsequential memory
access, and I would hope that array syntax would eliminate that common
programming error and execute as fast as the DO loop version with the
subscripts specified in the right order.  Maybe it's the more complex 
expressions that can be programmed more efficiently with DO loops, but can
somebody give some examples?

I've avoided using array notation simply because I don't know under what
circumstances execution will be slowed, and given that it isn't standard
yet, it represents a compiler "enhancement" that I don't want to use.  You
would think the compiler writers would provide a little more detail on
when to use it and when NOT to use it, but so far...