DAGMC Compilation

[makeclean]

Not possible on CHTC submit node due to version of gfortran (<< 4.2) that Mendeleev used to compile the periodic table.

Compile on own machine using standard compilation, i.e.

make clean build CONFIG='dagmc gfortran' FC=gfortran CC=gcc MOAB_DIR=moabdir

Linked against moab that was linked against hdf5 only. When this builds objects but fails to link, use the following link statement

gfortran *.o ../dagmc/*.o -o mcnp5 -static -lstdc++ -L/opt/condor_mcnp/moab/lib/
-static -lMOAB -static -ldagmc -L/opt/dagmc/hdf5/lib -static -lhdf5 -lstdc++

Should eventually find more elegant method.

[makeclean]

However, due to very old kernel version on CHTC, can't use executable compiled on my systems. When we try to to run this on CHTC get the error,

FATAL: Kernel too old
Segmentation Fault

Will report to CHTC

[makeclean]

Currently buillding gcc-4.4 and its dependent libraries on CHTC in order to bypass problem created by age of linux distro.

[makeclean]

GCC-4.4 did not work. Installed GCC-4.5.1 instead. After wrestling for a while, realised that the DAGMC routines are built with whichever c++ compiler that MOAB was build against. I had not updated MOAB build since I installed the newer GCC toolkit and was thus linking using older version of g++, which could not understand the fortran module name mangling of functions.

Modifying the toolchain such that all were build against gcc-4.5.1, we get a successfully static build of DAGMC,

 /home/davisa/compilation/gcc-4.5.1/bin/gfortran dagmc/*.o src/*.o -o mcnp5 -static -lstdc++ 
-L/home/davisa/dagmc/moab/lib -static -lMOAB -L/home/davisa/dagmc/hdf5/lib -static -lhdf5 -static -lz 
-L/home/davisa/compilation/gcc-4.5.1/lib64/ -static -lstdc++ -L/home/davisa/dagmc/moab/lib -static -ldagmc

To avoid future problems like this, would it be possible for @zwelchWI to investigate the possibility of modifying Linux.gcf and makefiles appropriately such that we set CXX at compile time rather than the MOAB default.

[zwelchWI]

I will look into it asap and let you know

[makeclean]

There is no rush, just something to think about.

[makeclean]

DAG-MCNP5 is compiled successfully to a static executable and has run as expected. Had several surface and cell tallies along with a tet mesh score. Using a submit script as below

 #############################################
 ## submit description file for a parallel program
 #############################################

universe = vanilla
executable = /home/davisa/dagmc/mcnp5src/bin/mcnp5
arguments = c i=divertor_sim_cont g=divertor_simgeom.h5m r=runtpe o=divertor_sim.o 
output = std.out
error = std.err
log = std.log
copy_to_spool           = false
should_transfer_files   = yes
when_to_transfer_output = on_exit
transfer_input_files = runtpe,divertor_sim_cont,divertor_simgeom.h5m,divertor_simmesh.h5m
+AccountingGroup = "EngrPhysics_Wilson"
queue

Model used was a model of ITER divertor being used for lost particle studies, faceted to 10^-3 tolerance on my local machine and then transferred files that were needed to run.

[makeclean]

Im going to mark this as closed unless there are any comments?

[makeclean]

More converged flux data

[gonuke]

Can we put all the input files and submission scripts necessary to recreate this into the file repo?

[makeclean]

Certainly can, put in as part of commit 91eab9c