Newsgroups: comp.lang.fortran
Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!zaphod.mps.ohio-state.edu!sdd.hp.com!news.cs.indiana.edu!ariel.unm.edu!cie.uoregon.edu!scavo
From: scavo@cie.uoregon.edu (Tom Scavo)
Subject: matrix multiplication
Reply-To: scavo@cie.uoregon.edu (Tom Scavo)
Organization: Campus Information Exchange, University of Oregon
Keywords: parameters, pass-by-value, pass-by-reference
Message-ID: <1991May04.170203.22222@ariel.unm.edu>
Date: Sat, 04 May 91 17:02:03 GMT
Lines: 32

As you all know, a classic programming exercise is the matrix 
multiplication problem

     A   =   B   C  .
    mxn     mxp pxn

A robust algorithm will also handle the special case

     X   =   Y   X                                           (1)
    mxn     mxm mxn

where the result is to replace one of the matrices being 
multiplied.  Such a calculation will break a naive algorithm, 
however.  One solution is to pass  B  and  C  by *value* (with 
loss of efficiency in both time and space) but of course Fortran's 
insistence that all parameters be passed by reference precludes 
such an approach.  On the other hand, pass-by-value can be 
simulated using a local array (for the lhs) declared within the 
matrix multiplication routine.  Unfortunately, the dimensions of 
this local array put an operational limit on the size of the input 
matrices.

I already have an inadequate routine that uses such a local array.  
How would I write a *general* matrix multiplication routine that 
handles tricky computations such as those in (1), and accomodates 
matrices of arbitrary size?  Also, does anyone know of a Fortran 
compiler that implements pass-by-value-result, a kind of delayed 
pass-by-reference where the actual parameters are updated at the 
point of return?  Or is this non-standard?

Tom Scavo
scavo@cie.uoregon.edu