RE: NEC-LIST: Lahey LF90 vs DEC PowerStation

From: Alexandre Kampouris <ak_at_email.domain.hidden>
Date: Mon, 02 Feb 1998 21:08:56 -0500

Hello,

I hadn't had the opportunity to compare different compilers, but I'm
quite happy with the Digital Visual Fortran V5.0.

Please note that "PowerStation" was a Microsoft product, which was
discontinued early last year. MicroSoft and Digital have an agreement
where the former abandons all Fortran work, and the latter integrates
its own compiler into MicroSoft's Developer Studio. This standard
environment lets you stick with the same programmer's toolchest for
all Microsoft languages, and also DEC Fortran. There is still a lot of
confusion betwen the trademarks.

May I offer you to send you a compiled stand-alone "command line mode"
binary so that you could perform your own comparisons?

I think it would be quite straightforward for you to run your
benchmarks on it, unless you have your own special code
modifications. I'd be interested to know myself how DF and Lahey
compares.

I presently run NEC2D with MAXSEG=1300, MAXMAT=600. (64 MB run time
memory system allocation). The binary is 564kB long, and half of that
when zipped. I haven't tried any special code manual
optimization. Although I used heavily symmetry in my last job, I
checked the problem once without that feature to make sure everything
was OK, which resulted in about 1200 segments. I haven't recorded the
run time, but it fan in a few minutes.

One advantage I see with DF5.0 is that The MathWorks supports it for
Matlab MEX/DLL file development. I plan to integrate NEC2 directly as
a computation engine available directly within Matlab. Presently, I
generate geometry card files with Matlab, then spawn NEC2, but
extracting output data using ad hoc programs is tedious. Data
extraction notwithstanding, it is really nice to work entirely from
within Matlab. But integrating NEC2 would involve a good code clean
up, especially to implement dynamic problem memory allocation using
the new Fortran 90 constructions, and also solving memory ownership
and management issues between the Matlab front-end and NEC2 back-end.

I had a few gripes with DF5.0, mostly regarding the documentation,
which was quite thin in the version originally shipped. This is now
largely solved.

I'm a tad worried by the acquisition of Digital by Compaq, as I don't
know yet what this means for long term software developements.

Those were my 2 pfennigs.

Alexandre Kampouris
Montreal
Received on Tue Feb 03 1998 - 09:42:53 EST

This archive was generated by hypermail 2.2.0 : Sat Oct 02 2010 - 00:10:38 EDT