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

Dear Fellow CEMists,

Thanks John, Jonathan and Jerry for the prompt responses. I had more or
less come to the same conclusion myself. The error occurs in seven places
where the matrix CM is declared.

The PC with 4GB RAM has an Intel P4x64 processor but it is currently running
Windows 2000 Professional (for compatibility with all the other machines).
NEC4 is compiled with CVF incorporating the Intel Math Kernel Library
routines for LU decomposition to speed up the execution time. We been using
it successfully, up to the full 2GB, (i.e. ~11,100 segs DP or 15,700 SP) on
a P3 with 2GB RAM for some time, execution times for a 1.2GHz processor
being about 3 hours.

I could install Windows XPx64 and IVF 9.1 (which requires MS Visual Studio
2002 on). According to Intel's web site, IVF will support Intel 64 bit
processors provided the "x64 Compiler and Tools" component is installed. I
am not sure about the MKL. Intel says that v8.1 is highly optimised for
(among others) P4 processors, but it doesn't say anything about 64 bit. I
have since found the following statement under 'Known Issues' on the Intel
knowledge base: The maximum size of an array that can be created by Intel(r)
IA-32 compilers is 2 GB. (ref:
http://www.intel.com/support/performancetools/fortran/windows/sb/cs-007828.h
tm) I cannot find a definitive statement that IVF running on 64 bit PC and
OS will overcome this limitation, so I shall ask the question of them and
report back. I am not sure if Jerry's suggestion of passing the CM array by
subroutine calls rather than a common statement will be necessary.

Having a physics background (with a good grounding in Fortran), I am a
CEMist with more emphasis on the E than the C, so sometimes my eyes glaze
over when looking at the Intel and MS web sites!

Regards,

Paul Carlier

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