NetNews Usenet Archive 1993 #3/NN_1993_3.iso/spool/comp/softsys/matlab/113

Text File | 1993-01-22 | 4.0 KB | 81 lines

Newsgroups: comp.soft-sys.matlab
Path: sparky!uunet!spool.mu.edu!yale.edu!ira.uka.de!Germany.EU.net!ubrinf!mathematik.uni-Bremen.de!bengt
From: [e-mail] (Bengt Martensson)
Subject: Re: Generating FORTRAN from MATLAB code ?
Message-ID: <[e-mail]>
Sender: [e-mail] (NEWS Service)
Nntp-Posting-Host: athena
Organization: University of Bremen, Germany
References: <[e-mail]> <raghu.727577016@morgan>
Date: Fri, 22 Jan 1993 11:15:19 GMT
Lines: 68

In article <raghu.727577016@morgan>, [e-mail] (Raghu B) writes:
|> [e-mail] (Werner P. Keilholz) writes:
|>
|> >can anybody tell me if there is some way to automatically generate
|> >FORTRAN code from a program written with MATLAB ?
|>
|> >This question came up in a discussion with a collegue who intends
|> >to develop numerical routines with MATLAB. I tried to talk him into
|> >using MAPPLE instead, because a salesperson once showed me a niffty
|> >little command, called 'FORTRAN', which generates the FORTRAN code
|> >corresponing to a MAPPLE expression automatically. I know neither
|> >MAPPLE nor MATLAB, but I like the idea of directly using the routines
|> >written by my collegue instead of rewriting everything in FORTRAN
|> o>or C (we want to use the result in a product that should be free of
|>
|>     Maple does give FORTRAN codes corresponding to Maple
|> expression. But you wouldn't want to use Maple to convert a Matrix
|> expression into Fortran Codes. You are dead then. If you have
|> say 10 x 10 matrix, Then FORTRAN will generate 100 Statements,
|> to define that matrix. You wouldn't want to do that. Another way
|> to reduce the no. of statements , is to use FORTRAN(xxx,optimize)
|> which might help you to reduce the number of statements. Overall
|> it is not an efficient way to generate FORTRAN codes, from
|> MAPLE. It is just an additional feature in MAPLE but not a
|> worthy one.
|>
|>                     Raghu,B
|>

With optimization, Maple actually generates quite efficient Fortran
(or C) code. As I understand it, there is quite some effort put into
those routines. If the code looks hairy, it is because you are
translating a problem to a low level language such as Fortran (or C).

To the point: If the problem is to convert a Matlab algorithm to a
compiled language (for performance reasons e.g.) I believe that the
best way would be to use a matrix class library in C++. In that way,
you can write matrix operations like in Matlab, while enjoying the
performance advantages of a compiled language. Since you don't have to
translate matrix operations, lack of automatic translation is probably
a minor issue.

One such matrix package was recently posted to comp.sources.misc,
called newmat, written by Robert Davies <[e-mail]>.
There are also commercial packages available (e.g. Rogue Wave).
This is from the posting of newmat:

       Newmat is an experimental  matrix  package in  C++.  It supports
    matrix types: Matrix, UpperTriangularMatrix, LowerTriangularMatrix,
    DiagonalMatrix, SymmetricMatrix, RowVector, ColumnVector, BandMatrix,
    UpperBandMatrix, LowerBandMatrix and SymmetricBandMatrix.   Only one
    element type (float or double) is supported.

       The package includes the operations *, +, -, (defined as operators)
    inverse, transpose, conversion between types, submatrix, determinant,
    Cholesky decomposition, Householder triangularisation, singular value
    decomposition, eigenvalues of a symmetric matrix, sorting, fast Fourier
    transform, printing, an interface with "Numerical Recipes in C" and an
    emulation of exceptions.

       It is intended for matrices in the range 15 x 15 up to the size your
    computer can store in one block.

--
Bengt Mσrtensson                                [phone] (office)
Institute for Dynamical Systems                 [phone] (home)
University of Bremen                            [phone] (fax)
P.O. Box 330 440, D-2800 Bremen 33, Germany     [e-mail]