MM5 V3.7 on pgf90 6.0-8. Segmentation fault

Hi,

I’m currently running successfully MM5/pgf90/MPICH on a debian.

However, I’m not able to run the OpenMP version (not mpi) of the code, getting segmentation faults at runtime. I’ve tried to compile using the configure.user provided in http://www.pgroup.com/resources/mm5/configure_60.linux32, and setting “unlimit” before the of execution of mm5.exe

I set “setenv OMP_NUM_THREADS 2” just before execution as the machine I’m trying to run this has 2 processors.

I don’t think it is a hard memory limitation because the computer has 4Gb of memory + swap.

Any idea why Segmentation is fault?

Thanks,

Víctor.

Hi Victor,

If you are compiling in 64-bit mode, then you need to add the define flag “-DDEC_ALPHA” to the compilation. If your in 32-bit mode, I’m not sure. You can try setting the OpenMP stack using the environment variable “MPSTKZ” to a large value.

Example:

export MPSTKZ=8M

If this doesn’t work, can you make available data set which illustrates the error and describe how you created your executable so we can recreate the error here. Note that we don’t offically support Debian so I’ll need to use SuSE.

Thanks,
Mat

Dear Mat,

Thanks for your reply.

We are running on a 64-bits AMD Opteron (Procesor 246) but the OS and all the software we run is 32 bits.
We did already try your suggestions but it keeps “Segmentation fault”'ing.

1> uname -a
Linux machine 2.4.32 #6 SMP mar may 30 12:52:52 CEST 2006 i686 GNU/Linux
(it is a debian 32 bits Stable)

2> cat /proc/version
Linux version 2.4.32 (root@machine) (gcc version 3.3.5 (Debian 1:3.3.5-13)) #6 SMP mar may 30 12:52:52 CEST 2006


The process we follow until we get to the error is:


3> Change to MM5 dir (you can download it anonymously from ftp://eady.uib.es/pub/victor/pgroup/mm5dir.tar.gz ) and:
make uninstall ( ftp://eady.uib.es/pub/victor/pgroup/make_uninstall.output )
make ( ftp://eady.uib.es/pub/victor/pgroup/make.output )
unlimit
setenv MPSTKZ 32M
cd Run
./mm5.exe ( ftp://eady.uib.es/pub/victor/pgroup/mm5.print.out )


Let me know if you need other info from our system,

Victor.

Hi Victor,

Thank you for posting these files. I download them last week and did get the seg fault using the binary you built. However due to time constraints, I’ve had to pass this on to another engineer. It may take us some time to understand the issue, but are investigating.

Can you try re-compiling using our latest version (6.2-3) found HERE?

Thanks,
Mat

Hi Victor,

We think that the runtime error is a stack overflow but for some reason, setting the stack size to unlimited isn’t enough. When we explicitly set the stack size to 128M in a cshell environment, your executable ran to completion.

Can you please try setting “limit stacksize 128M” and re-running?

Thanks,
Mat