open TELEMAC-MASCARET The mathematically superior suite of solvers

Hi! Well, finally, I could compile Telemac v7p2 (SVN-trunk one month ago) using the following config-file:
---->
[Configurations]
configs : smuc2intelmpipar
[general]
root : /path-to-telemac/opentelemac3/v7p2
version : v7p2
language : 2
modules : update system -dredgesim
options :
mods_all : -I <config>
sfx_zip : .gztar
sfx_lib : .a
sfx_obj : .o
sfx_mod : .mod
sfx_exe :
val_root : <root>/examples
val_rank : all
[smuc2intelmpipar]
options : parallel mpi
par_cmdexec : mpiexec <config>/partel_para < PARTEL.PAR >> <partel.log>
mpi_cmdexec : mpiexec -wdir <wdir> <exename>
mpi_hosts:
cmd_obj_c : mpicc -c <srcName> -o <objName>
cmd_obj : mpif90 -c -O3 -DHAVE_MPI -DHAVE_PARMETIS -convert big_endian -sox -xsse4.2 <mods> <incs> <f95name>
cmd_lib : ar cru <libname> <objs>
cmd_exe : mpif90 -convert big_endian -o <exename> <objs> <libs>
incs_parallel : -I/lrz/sys/intel/impi/5.1.3.181/intel64/include
libs_partel : <root>/optionals/parmetis-4.0.3/lib/libparmetis.a <root>/optionals/parmetis-4.0.3/lib/libmetis.a
libs_all : <root>/optionals/parmetis-4.0.3/lib/libparmetis.a <root>/optionals/parmetis-4.0.3/lib/libmetis.a /lrz/sys/intel/impi/5.1.3.181/intel64/lib/libmpi.so
---->
This was no that easy as it might sound as I had to compile parmetis as well. And here are some degrees of freedom, where I'm not sure that I chose the correct one.
But fine, the usual workflow (partel serial/telemac parallel) seems to work. So I tried to use partel_para instead of partel. Bur here, something went wrong. First of all, partel_para starts parallely. So far okay.
But then, partel_para complains about a missing file 'partel.par' (lower cases). I found this file name hardcoded inside partel_para.F. But the common workflow script (runcode.py) only creates PARTEL.PAR (upper case). But this is not just a question of case-sensitivy, unfortunately. Also the content of PARTEL.PAR does not fit to what partel_para expects. As I couldn't find any documentation about the input of partel_para, I simply tried from the source code expectation (partel_para.F:253):
---->
OPEN(UNIT=ID_INPUT,FILE='partel.par')
READ(ID_INPUT,*) NAMEINP
READ(ID_INPUT,*) NAMECLI
READ(ID_INPUT,*) NPARTS
READ(ID_INPUT,*) PMETHOD
CLOSE(ID_INPUT)
---->
So, I tried (partel.par):
---->
T2DGEO
T2DCLI
28
3
---->
However, this crashed - later (I still have to fix the right position) - with backtrace of the fortrtl.

So,I'm probably doing something really wrong here, as it makes me have the impression that partel_para (partel and parmetis coupling) is still not really supported. But this can hardly be after all! Could somebody tell me, where in my confused mind lies the error? Is my config-file okay? What is missing?

Many thanks for any kind of hint
Cheers, Martin

Hi,

This is my bad i modifief the partel.par and forgot to update partel_para accordingly.
Your solution should have worked that is weird.
I'll try to have a look at that before the holidays.

Hello Yugi

Btw. once in general: Thank you really much for your engagement!

So. After one week off due to obscure disease, I downloaded from the trunk v7p2, and compiled with the same options as in the last post. First of all: dredgesim was missing (cured by copy from previous source -> Ok.). Second, in partitioner_para.F on lines 179 and 338, the function P_ALLGATHERV_I does not contain the comm variable anymore (I suppose that this is the MPI communicator, which was now made "global" such that it does not need to be given as parameter anymore). So, I removed comm, and compiled successfully. Third, the Python-script runcode.py is still not preparing for partel_para -- so "partel.par" is not created inside the working directory. I changed once more according to what I mentioned in the last post. And now it works!

I still have to test it more extensively - scaling properties etc. But at least, when it worked once, I'm at least confident that all input files work properly.

So, once more, thank you very much
Cheers,
Martin

Oje, it goes on! I was too early too happy

Now splitting using parmetis seems to work. But it generates files of the scheme T2DGEO00027-00001, (T2DRES00027-00001 ? probably not by partel_para) and T2DCLI00027-00001. The latter are for the boundary files, I suppose. Anyway, telemac2d is supposed to take files of scheme T2DPAR00027-00001 (might be wrapped by the python script?). At least I now get the following error:
|forrtl: severe (29): file not found, unit 7, file /home/hpc/pr28fa/di49zop/Progs/opentelemac3/Cases/2km/16/t2d.cas_2016-12-20-13h09min35s/T2DPAR00027-00027

Any ideas?
Riddles over riddles! Thanks in any case
Cheers, Martin

Okay. Now I checked. partel_para should have created the T2DPAR files. But did not. So, it was not successful, yet. (Shit!)

Hi Yugi, sorry for being a bit late - not enough sleep I right now comprehended your last post. So, I'll wait patiently for your next update
In any case, merry X-mas and a happy new year!
Cheers,
Martin

Indeed the generation of T2DPAR is a new feature that came with hermes and as partel_par is not hermes-ready hence the error.

I do not know yet when i'll have time to work on that i'll keep you posted.

Hello Yugi, happy new year for you
Okay, so I've to wait a bit longer. Thanks for the info.
Cheers,
Martin

Dear Martin and Yugi

I just came back to test on my computer, I realise that I get the same problem.

Once you done it, could you share with me your configuration ?

Thanks & Best Regards
Huy