Re: NEC-LIST: Re: Solaris compilation

Mike-

I played around with the nec2dx.f source code and the f77 and g77
compilers this afternoon and think I've found a solution to
your reported problem. Looks like you were chasing a red-herring.....

What I did:

1) Compiled source code under f77 with a variety of different options.
   Result: resulting nec2dx executable all ran your test.nec input
   deck just fine.

2) Compiled source code under g77 with a variety of different options.
   Result: resulting nec2dx executable all ran your test.nec input
   deck with the error you reported.

3) Took a big step backwards and looked over everything. Noticed that the
   test.nec deck didn't have an EN card at the end. Added an EN card.
   Result: Source code compiled under g77 (with a number of different
   options) produce an executable that rand the modified test.nec deck
   just fine. Tried the modified test.nec deck with an executable compiled
   under f77. This ran just fine as well.

So it seems that the f77 compiled executable ran the nec deck without
the final EN card OK, but the g77 compiled executable required the final EN
card to execute properly.

(BTW, all the above done on a Sun Ultra-10 running Solaris 8, g77 version 3.2.1,
f77 version 4.2)

-Ray Anderson WB6TPU
Staff Signal Integrity Engineer
VLSI Engineering
Sun Microsystems Inc.

>
>That works too, but i still get an error at the end of my nec output
>that i dont get when i use the dos version. Anyone know whats wrong
>or have any ideas?
>
>***** CARD ERROR - INVALID NUMBER AT INTEGER POSITION 1
> ***** TEXT --> =°?$ÛÉ LINEAR ` o?Ë` o?Ë` o?Ë`@V`@V`
>
>http://spawn.signull.com/~mikes/nec2/solaris/test.out
>
>
>-Mike Spenard
>
>
>
>> mike,
>>
>> try...
>>
>> g77 -O4 -Wno-globals nec2dx.f
>>
>> when i last tried compiling as you ahve it below the compiler finished
>> but generated warnings. the man/info pages for the gnu tools are
>> helpful when you have warnings to work with... you might also try
>> using -O5 or 6 instead jsut to see if it speeds things up (g77 should
>> support this...). i havent done any numerical validation of the gcc
>> executables. i assume others have done such, but maybe not...
>>
>>
>> chris
>>
>>
>> to get it working on solaris. The fortran code.
>> I have the C port and i cant get it to compile with gcc. So im
>> sticking
>> with the fortran code for now.
>> Ive had a few problems.. i dont know fortran, im fluent with C
>> which is why i wanted a C port..anyway..
>>
>> I dont have f77, but g77. The compile instructions
>> say to do "f77 -fast -O4 -parallel nec2dx.f"
>> g77 doesnt support -fast and -parrallel arguments, are
>> there equivalent arguments for g77?
>>
>> It does compile without those arguments, but when i run it it
>> processes
>> my .nec file, but at the very end i get..
>>
>> ***** CARD ERROR - INVALID NUMBER AT INTEGER POSITION 1
>> ***** TEXT --> =°^X$ÛÉ LINEAR `
>>
>> my output file can be found here
>> http://spawn.signull.com/~mikes/nec2/solaris/test.out
>>
>> and the rest here..
>> http://spawn.signull.com/~mikes/nec2/solaris/
>>
>>
>> -Mike
>>
>>
>> > It's been (roughly) ported to C, but I have no idea of the quality
>> or
>> > fidelity. I would imagine a better strategy would be to find the
>> FORTRAN
>> > version compiled for Linux. There are a bunch of other variants it
>> has
>> been
>> > compiled for on the various archive sites.
>> >
>> > As far as Beowulf and clusters with NEC, it's been done quite a lot.
>> Your
>> > real question is whether you want to do the matrix math on the
>> cluster
>> (i.e.
>> > one problem, divided up among processors), in which case you need to
>> track
>> > down one of the modified versions that uses PVM or MPI matrix math
>> > libraries. Or, the alternate strategy, running a lot of parallel
>> copies
>> > for, e.g. genetic algorithms, systematic sensitivity analysis, etc.,
>> in
>> > which case it's an embarrassingly parallel problem and some scripts
>> will
>> do
>> > very nicely, along with some sort of parser to scatter the input
>> files and
>> > gather the output files.
>> >
>> > ----- Original Message -----
>> > From: "Mike Spenard" <mikes_at_signull.com>
>> > To: <nec-list_at_gweep.ca>
>> > Sent: Friday, December 27, 2002 6:57 PM
>> > Subject: NEC-LIST: Unix NEC or C port?
>> >
>> >
>> > > Has NEC been ported to a Unix variant or to C?
>> > > Being stuck in dos mode is a drag, i would like
>> > > to be able to run NEC on a beowolf cluster to
>> > > do some more sophisticated modeling.
>> > >
>> > > Mike Spenard
>> > > --
>> > > The NEC-List mailing list <nec-list_at_gweep.ca>
>> > > http://www.gweep.ca/mailman/listinfo.cgi/nec-list
>> > --
>> > The NEC-List mailing list <nec-list_at_gweep.ca>
>> > http://www.gweep.ca/mailman/listinfo.cgi/nec-list
>> >
>>
>>
>--
>The NEC-List mailing list <nec-list_at_gweep.ca>
>http://www.gweep.ca/mailman/listinfo.cgi/nec-list

-- 
The NEC-List mailing list <nec-list_at_gweep.ca>
http://www.gweep.ca/mailman/listinfo.cgi/nec-list