[Wien] error in "symmetso"

fecher fecher at mail.uni-mainz.de
Thu Dec 16 17:05:07 CET 2004


**** adress error ****
is not a Wien2k message, but a message from the crashed programm that is based 
on the fortran compiler. That means, not the source code of Wien2k has a 
problem but the program made from the source code by the compiler and linker.

I am fighting with just the same problem (as I use Fe very often as a test 
case) since more than one month, my solution: use the Portland Group Compiler 
instead of the Intel Compiler, this will work for sure, indeed somewhat 
slower because there is no optimized lapack (one may make an own atlas if 
needed). 

It seems that this is an error typically produced by the intel compiler 8.0 or 
higher. I have the experience that sometimes the variables are not passed 
correctly to the subroutines and the program tries to access an array element 
that is not in the dimensions. Mostly element 0, as the integer expected to 
be passed to the subroutine was just lost and set to Zero. This causes either 
some adress error or some segmentation fault, more or less the way the 
compiler or linker likes to handle it at runtime and independent what really 
happened. It should be mentioned that this fault appears even without any 
optimization and in debug mode. Such errors are presently hard to find as the 
ifort 8.0 and higher does not support all runtime checks that have been 
possible with older versions on the one hand and that the intel debugger 
idb81 just crashes. These errors are also independent of any settings for the 
limits of memory or stack. Proposed workarrounds like limitting the stack to 
a fixed large size instead of unlimitted do not work either.
Indeed sometimes one may find something like that or the compile and link 
options help, but this appears rather arbitrarily.

If you like to play arround, use a different Computer, or use a different 
Version of Linux, or both, or use different compile options or different link 
options, or try whatever else. Sometimes it helps, but sometimes not. I have 
some programs that run well if linked statically, even this is supposed not 
to work and produces stupid warnings that some dynamic libraries are needed 
at runtime, indeed ldd tells that the program does not need any dynamic 
library.

Using Intel ifort 8.1 with MKL 7.0.1: The Fe case with and whithout SO, you 
tell about, is running on my laptop with 256MB RAM and a Pentium III mobile 
and Suse Linux 9.1. It does not run on Pentium IV with Suse Linux 9.1 and 2GB 
of RAM (and I tried nearly every possible and impossible combination of 
compiler and linker options). The Fe case runs on Pentium IV 1GB RAM and SUSE 
Linux 9.2, but then it dies for a different structure either with ***adress 
error***, or *** segmentation fault ***, however, not always in the same 
subroutine of Wien2k but depending on the system I like to calculate.
I compiled with 7.x and MKL 6.x, and everything was running fine.

I experienced the same problems with other Fortran programms, producing either 
junk or memory faults, but only with ifort 8.0 or higher. All programs were 
running fine if using a different fortran compiler starting even from Gnu f77 
(possible only if there is no f90 code) or with compilers under Windows like 
the former Digital, now Compaque or HP Fortran 6.6c.

I think there seem to be 2 Problems, first there may be some incompatibilities 
between different Versions of the libraries coming with Linux, ifort, and 
mkl. Second ifort and or mkl or both have presently some serious memory 
managment problems.

Good Luck

Ciao Gerhard

Am Donnerstag, 16. Dezember 2004 15:20 schrieb 王福合:
> Hi, everyone,
>
> When I calculated the magnetic material bcc-Fe with spin-orbit coupling and
> spin-polarized, I had a problem. First,I calculated the structure with
> spin-polarized but without spin-orbit coupling. Then,I wanted to run the
> script "symmetso",but error information showed on the screen just like
> this: [root.....]# x symmetso
>  ** Address Error **
> End of diagnostics
> [root.....]#
>
> The file "*.struct_so" is created,but it is empty.
>
> I have never met the information and don't know what cause it. I hope
> somebody can help me.
>
> Thanks in advance!
>
> Fu-He Wang
>
>
> _______________________________________________
> Wien mailing list
> Wien at zeus.theochem.tuwien.ac.at
> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien





More information about the Wien mailing list