NEC-LIST: Re: Compiling NEC for a Large Number of Segments

Hello, all. I have been following this discussion for a couple of
days now, and just couldn't help not putting in my experience on this
issue. From my experience, the 32-bit Windows platform (probably)
does not allow any process to run using more than 2 GB of memory,
therefore, you probably cannot compile it as such. In my previous
attempts to run a simulation process on a 32-bit Windows platform, I
was unable to run anything requiring more than 1.8 GB. (nevertheless,
I had spoken to someone knowledgeable enough on this, who told me
that some Linux platforms allow you to overcome this limitation on a
32-bit PC).

Therefore, I had moved my compilation (not a NEC code, but another
Fortran MoM code) onto the Solaris Unix platform, which is a 64-bit
system. Compiling it as a 64-bit compilation was a little tricky,
since by default, my compiler generally compiles it as a 32-bit code
(maybe my compiler is somewhat obsolete by today's standard). I had
to use a processor specific option to indicate that the processor is
a 64-bit type, and that the architecture is also a 64-bit type, plus
some form of compilation optimization. I tested running the code, and
it passed with flying colors.

In my opinion, it is generally better, and relatively inexpensive
these days, to move onto a full 64-bit system than to try figuring
out the 32-bit system.

Regards,
~Kin

At 18:24 8/9/2006, you wrote:
>Mr. Carter and all;
>
>I have had for some time worked on getting the maximum number of
>segments out of NEC-4 built on a Unix-like system. The best that I have
>gotten is 30000 segments. Beyond that, I get compiler exceptions of the
>same general sort as you got. Earlier discussions on the board alluded
>to the need to change certain parts of the Fortran code to accomidate
>larger segment definition. This setup was compiled and runs on MacOS X
>10.4.6 on a klunky old 32-bit, 768MHz, QuickSilver G4 - slow, but it got
>results.
>
>All attempts to run very large models even for very good commercial
>NEC-4 packages resulted in failures of Windows memory management. The
>IT manager here is setting us up with a Linux PC (probably Xandros
>64-bit when it is released). This will allow compiling NEC-4 Unix
>source using g77 which is a f2c-like translater (I think) that then
>calls the gcc compiler.
>
>In case you want to try the Linux route, the reference that enabled me
>to compile was posted on this list by John B. Wood 03/04/04:
>
> > Hello. I ran a thread on this subject a while back. NEC 4.1 can be
> > somewhat optimized to run on a Linux box by compiling with g77 as
> > follows:
> >
> > g77 nec41.f -O3 -fno-automatic
> >
> > where nec41.f is the NEC 4.1 f77 source code file. The default
> > executable filename will be nec41.o. The above gave me results
> > consistent with those I had obained on an old HPUX workstation. You
> > might also try
> >
> > g77 -O3 -fno-emulate-complex -funroll-loops nec41.f
> >
> > This version of the compile was published in the Nov. '98 ACES
> > newletter but I never could get the executable to run reliably on the
> > laptop I was using (Panasonic Toughbook CF-27). Sincerely,
> >
>
>I have used both of the above proceedures to compile NEC-4 on Mac OS X.
>and have gotten good results for double-precision models of about 16000
>wire segments. As I remember, someone mentioned a complication with
>very large models using wire segmants that connect to patches.
>
>Hope this helps,
>
>Darryl Holder
>
>--
>The NEC-List mailing list
>NEC-List_at_robomod.net
>http://www.robomod.net/mailman/listinfo/nec-list

-- 
The NEC-List mailing list
NEC-List_at_robomod.net
http://www.robomod.net/mailman/listinfo/nec-list