From:	jesse AT lenny DOT dseg DOT ti DOT com (Jesse Bennett)
Newsgroups:	comp.os.msdos.djgpp
Subject:	Re: Netlib code [was Re: flops...]
Date:	13 Feb 1997 04:12:48 GMT
Organization:	Texas Instruments
Lines:	39
Message-ID:	<5du4c0$kor$1@superb.csc.ti.com>
References:	<Pine DOT LNX DOT 3 DOT 91 DOT 970208044630 DOT 1946B-100000 AT aditya DOT unigoa DOT ernet DOT in>
	<5dh4gi$ek9$1 AT superb DOT csc DOT ti DOT com>
	<1997Feb12 DOT 130129 DOT 27922 AT indyvax DOT iupui DOT edu>
Reply-To:	jbennett AT ti DOT com (Jesse Bennett)
NNTP-Posting-Host:	lenny.dseg.ti.com
Mime-Version:	1.0
To:	djgpp AT delorie DOT com
DJ-Gateway:	from newsgroup comp.os.msdos.djgpp

In article <1997Feb12 DOT 130129 DOT 27922 AT indyvax DOT iupui DOT edu>,
	mwood AT indyvax DOT iupui DOT edu (Mark H. Wood) writes:
> I have to ask:  why not simply code the relevant parts of your program in
> Fortran?

Because in my application all of the code is relevant.  The truth of
the matter is that I have a huge investment in C code that I don't
want to rewrite.  But, I am considering doing just that due to the
lack of portable, high quality and high performance C libraries for
linear algebra.  I am currently calling Fortran compiled LAPACK/BLAS
library functions from my C code and the performance is lackluster, to
say the least.  The problem is not with the performance of the Fortran
code but with the memory bandwidth overhead associated with converting
the C row-major matrices to the Fortran column-major order prior to
calling each Fortran function and then back to row-major order before
returning to the C code.  For small-medium scale problems this is
probably no big deal, but I have programs that consume 10-15 *days* of
CPU time on a 500Mhz DEC Alpha system.  My initial experiments
indicate that this time would be reduced by at least 50% in a native
Fortran implementation.  So I will probably do as you suggest, rewrite
the relevant portions of my code in Fortran.  In other words - all of
it.

The sad thing (to me) is that well written C can perform at least as
well (and often better) than equivalent Fortran code in numerical
analysis applications.  Unfortunately, as it stands today, most
serious scientific code will continue to be written in Fortran77
because of the huge body of well written and highly optimized code
that is already available for these applications, as exemplified by
the netlib collection.  Much of this code represents 30+ years of
development so it is very mature and stable.  Most (if not all) of the
Fortran BLAS routines were written in the 70's, for example.

You just had to ask, didn't you. :^)

Oh well... I feel better now.  Hmm... I wonder where I put that
Fortran text?

--Jesse

- Raw text -