Using PGDBG with GEMCLIM 3.3.2 / 3.3.2.1

Some general information about the PGDBG, the Portland Group debugger:
http://www.pgroup.com/doc/pgitools.pdf

PGDBG can be a veeeery helpful tool for debugging.
This debugger is installed on marvin and can be used to debug GEMCLIM 3.3.2 and other Fortran codes.

So far I did not find out how to use PGDBG for debugging of codes using more than one core. If you do, please let me know!
Therefore...

When using PGDBG for debugging GEMCLIM 3.3.2 make sure you model is running on only 1 core:

configexp.dot.cfg:
BACKEND_OMP=1;

gemclim_settings.nml:
Ptopo_npex      = 1          , Ptopo_npey      = 1          ,
Ptopo_nblocx    = 1          , Ptopo_nblocy    = 1          ,
Ptopo_smtdyn    = 1          , Ptopo_smtphy    = 1          ,

First you have to "prepare" your executable(s) for the debugger:

- Set the model environment as usual

- Make sure you have a copy of the following object file in your directory 'malibLinux_x86-64_pgi611':
   ~winger/gem/v_3.3.2_clim/Abs/rpn_comm_stubs/rpn_comm_stubs.o
(These are stubs so that MPI will not be used.)
Make sure you do not forget to remove it before running with MPI again!

- Compile all routines you want to debug in debugging mode by adding "OPTIL=0 FCOMPF=-debug":
   make routine.o OPTIL=0 FCOMPF=-debug
or
   make objloc OPTIL=0 FCOMPF=-debug

To compile a lot of routines you can extract a comdeck (*.cdk) which is included in a lot of routines, for example 'options.cdk'.
Then do the 'make objloc ...').

- Create the absolutes, this time adding (OPTIL=0 DEBUG=-debug):
   make gemclimdm_nompi OPTIL=0 DEBUG=-debug
   make gemclimntr_nompi OPTIL=0 DEBUG=-debug

Now you should be able to start the model

Go into the directory with your config files. (Make sure UM_EXEC_ovbin points to your new absolutes.)

There launch the model with:
   Um_lance . -BACKEND_nosubmit
This command will create the jobs for the entry, ${UM_EXEC_exp}_E, and the model, ${UM_EXEC_exp}_M but not submit them.

To run the entry execute the following 3 lines from the entry job, ${UM_EXEC_exp}_E and export one variable:
1) Go into the execution directory (replace ${UM_EXEC_exp} by the real name):
   cd /home/${USER}/MODEL_EXEC_RUN/headnode/${UM_EXEC_exp}
2) Execute the config file:
   . RUNENT_upload/configexp.cfg
3) Run the entry
You need to use the debug version of the script used to run the entry. Therefore:
   export RUNENT=Um_runent.debug.ksh
The run the entry with for example:
   . r.call.dot ${RUNENT:-Um_runent.ksh} \
        -inrep /local/sata2/winger/gemclim/lam/nestinfo/teta_3321_debug_200701 \
        -anal /local/sata1/dueymes/data/ECMWF/anal_climct39_2007010800_ECMWF_200701080000_new \
        -climato /local/sata1/data/climatology/clim_gemdm320_1080x540 \
        -geophy /local/sata1/dueymes/data/geophys_arctic_0.25deg_364x308 \
        -climat 1

(Your '-inrep ... -anal ... -climato ... -geophy ...' might be different.)

The debugger will open automatically.
In the upper part of the command window type 'run', or click the 'Run' button.
Once the entry is finished, quit the debugger with 'q'.

Now you can run the model in a similar way you ran the entry.
Execute the following 3 lines from the model job, ${UM_EXEC_exp}_M and export one variable: :
1) Go back into the execution directory (replace ${UM_EXEC_exp} by the real name):
   cd /home/${USER}/MODEL_EXEC_RUN/headnode/${UM_EXEC_exp}
2) Execute the config file:
   . RUNMOD_upload/configexp.cfg
3) Run the model
   export RUNMOD=Um_runmod.debug.ksh
The run the model with for example:
   . r.call.dot ${RUNMOD:-Um_runmod.ksh} -lam_init ${UM_EXEC_lam_init} -lam_bcs ${UM_EXEC_lam_bcs} \
         -geophy_m ${UM_EXEC_geophy_m} -xchgdir ${UM_EXEC_xchgdir} \
         -barrier ${UM_EXEC_barrier} -timing ${UM_EXEC_timing} \
         -manoutp ${UM_EXEC_manoutp}                               \
         -climat    1    -outrep   /local/fiber1/winger/gemclim/EXECDIR

(Your '-outrep ... ' should be different.)

The debugger will again open automatically. And you do there whatever you want.

Rerun the model
Once you did the setup above you do not have to redo it again. In case you want to use the debugger again (or on another day or in a different window). Just execute the following 3 lines:
1) Go back into the execution directory (replace ${UM_EXEC_exp} by the real name):
   cd /home/${USER}/MODEL_EXEC_RUN/headnode/${UM_EXEC_exp}
2) Execute the config file:
   . RUNMOD_upload/configexp.cfg
3) Go into the working directory !!!:
   cd RUNMOD/work
4) Start the debugger:
   pgdbg -dbx ../bin/ATM_MOD.Abs &

Have fun ;-) !

Author: Katja Winger
Last update: August 2011