[SOLVED] Compilation woes of saturne...

Hi Forum,

this may seem stupid enough...

I already compiled successfully Code Saturne in another system, a Debian lenny 5.0 Stable...

I installed another PC in the lab, but somehow, I am failing to compile CS again... and I am sure that I am repeating all the steps i took to successfully compile CS before... :

# tar -xzvf fvm-0140.tgz&&tar -xzvf bft-111.tgz&&tar -xzvf ecs-200-beta2.tgz&&tar -xzvf mei-100.tgz&&tar -xzvf ncs-200-beta2.tgz&&clear&&ls -la

drwxr-xr-x 7 alex alex 4096 2009-09-01 00:34 .
drwxr-xr-x 24 alex alex 4096 2009-08-25 20:03 ..
drwxr-xr-x 7 8380 3299 4096 2009-09-01 00:35 bft-1.1.1
-rw-r--r-- 1 alex alex 444194 2009-08-13 01:02 bft-111.tgz
drwxr-xr-x 9 8380 3299 4096 2009-07-28 14:11 ecs-2.0.0-beta2
-rw-r--r-- 1 alex alex 1226998 2009-08-13 01:02 ecs-200-beta2.tgz
drwxr-xr-x 7 8380 3299 4096 2009-09-01 00:38 fvm-0.14.0
-rw-r--r-- 1 alex alex 777515 2009-08-13 01:03 fvm-0140.tgz
-rw-r--r-- 1 alex alex 30340 2009-08-16 00:51 index.php
-rw-r--r-- 1 alex alex 2444 2009-08-17 03:02 march
drwxr-xr-x 9 8380 3299 4096 2009-09-01 00:41 mei-1.0.0
-rw-r--r-- 1 alex alex 423567 2009-08-13 01:03 mei-100.tgz
drwxr-xr-x 16 8380 3299 4096 2009-09-01 00:45 ncs-2.0.0-beta2
-rw-r--r-- 1 alex alex 13719174 2009-08-13 01:13 ncs-200-beta2.tgz

# cd bft-1.1.1
# ./configure --prefix=/usr/local/saturne
# make&&make install
# cd ../fvm-0.14.0
# ./configure --prefix=/usr/local/saturne --with-mpi=/usr/bin --with-mpi-include=/usr/include/mpi --with-mpi-lib=/usr/lib --with-bft=/usr/local/saturne
# make&&make install
# cd ../mei-1.0.0
# ./configure --prefix=/usr/local/saturne --with-bft=/usr/local/saturne --disable-python-bindings
# cd ../ncs-2.0.0-beta2
# ./configure --prefix=/usr/local/saturne --with-mpi=/usr/bin --with-mpi-include=/usr/include/mpi --with-mpi-lib=/usr/lib --with-fvm=/usr/local/saturne --with-bft=/usr/local/saturne --with-mei=/usr/local/saturne --with-syrthes=/usr/local/syrthes/syrthes3.4.2

So far so good :

Configuration options:
use debugging code: false
MPI (Message Passing Interface) support: yes
OpenMP support: no
BLAS (Basic Linear Algebra Subprograms) support: yes
Libxml2 (XML Reader) support: yes
MEI (Mathematical Expressions Interpreter) support: yes
SYRTHES 3 coupling support: yes
IP socket support (for SYRTHES 3 or CFD_Proxy): yes
Dynamic loader support (for YACS): yes

The package has been configured. Type:
make
make install

To generate and install the Code_Saturne kernel

.................................


# make


...............................

And the annoying result, regardless my explicit instruction to compile with MPI support, is this... :


libtool: link: gcc -funroll-loops -O2 -Wuninitialized -std=c99 -funsigned-char -pedantic -W -Wall -Wshadow -Wpointer-arith -Wcast-qual -Wcast-align -Wwrite-strings -Wstrict-prototypes -Wmissing-prototypes -Wmissing-declarations -Wnested-externs -Wunused -Wfloat-equal -rdynamic -O -o .libs/cs_solver ./.libs/libsaturne.so -L/usr/local/saturne/lib /usr/local/saturne/lib/libmei.so -L/usr/lib /usr/local/saturne/lib/libfvm.so /usr/local/saturne/lib/libfvm_coupl.so /usr/lib/libmedC.so /usr/lib/libhdf5.so -lpthread /usr/local/saturne/lib/libbft.so -lz -lmpi /usr/lib/libxml2.so -lcblas -latlas -L/usr/lib/gcc/x86_64-linux-gnu/4.3.2 -L/usr/lib/gcc/x86_64-linux-gnu/4.3.2/../../../../lib -L/lib/../lib -L/usr/lib/../lib -L/usr/lib/gcc/x86_64-linux-gnu/4.3.2/../../.. -lgfortranbegin -lgfortran -lm -ldl -Wl,-rpath -Wl,/usr/local/saturne/lib
./.libs/libsaturne.so: undefined reference to `MPI_Type_create_struct'
./.libs/libsaturne.so: undefined reference to `MPI_Get_address'
collect2: ld returned 1 exit status
make[3]: *** [cs_solver] Error 1
make[3]: Leaving directory `/home/alex/Desktop/Newdat/CDSat/ncs-2.0.0-beta2/src/apps'
make[2]: *** [all-recursive] Error 1
make[2]: Leaving directory `/home/alex/Desktop/Newdat/CDSat/ncs-2.0.0-beta2/src'
make[1]: *** [all-recursive] Error 1
make[1]: Leaving directory `/home/alex/Desktop/Newdat/CDSat/ncs-2.0.0-beta2'
make: *** [all] Error 2


..........................................


...I feel quite stupid about this...


BTW, some useful ( ? ) info :

Syrthes was compiled with INTEL, and I do not have libhdf5-openmpi libraries nor headers, instead I have libhdf5-serial libs and headers...


Is this an issue... ?

Could somebody please post the patch in order to avoid having to compile with MPI support...?

... man, I feel clueless here....

Thanks in advance


Alex<br /><br />Post edited by: Alex, at: 2009/09/02 00:55

Hi Claus

I understand that i must use MPI to build CS-2.0.0... in fact I did it :

with fvm ,

# cd ../fvm-0.14.0
# ./configure --prefix=/usr/local/saturne --with-mpi=/usr/bin --with-mpi-include=/usr/include/mpi --with-mpi-lib=/usr/lib --with-bft=/usr/local/saturne

and with ncs,

# cd ../ncs-2.0.0-beta2
# ./configure --prefix=/usr/local/saturne --with-mpi=/usr/bin --with-mpi-include=/usr/include/mpi --with-mpi-lib=/usr/lib --with-fvm=/usr/local/saturne --with-bft=/usr/local/saturne --with-mei=/usr/local/saturne --with-syrthes=/usr/local/syrthes/syrthes3.4.2

what I do not understand is that what i did now, resulted successful about a month ago, in another Debian 5.0 install that i made... and now it is not working in a fresh install in the very same OS/arch in another machine...

There is something else that is eluding me here...

Anyway, I will try with the most recent files of CS, and post my results here...

I think ( and hope ) that the ./configure process is independent of the exact sequence of the options...

( ... ie : stating

./configure --prefix=/usr/local/saturne --with-mpi=/usr/bin --with-mpi-include=/usr/include/mpi --with-mpi-lib=/usr/lib --with-fvm=/usr/local/saturne --with-bft=/usr/local/saturne --with-mei=/usr/local/saturne --with-syrthes=/usr/local/syrthes/syrthes3.4.2

amounts to the same as stating

./configure --prefix=/usr/local/saturne --with-fvm=/usr/local/saturne --with-bft=/usr/local/saturne --with-mei=/usr/local/saturne --with-mpi=/usr/bin --with-mpi-include=/usr/include/mpi --with-mpi-lib=/usr/lib --with-syrthes=/usr/local/syrthes/syrthes3.4.2...

right... ?? )

I will also try installing the MPI version of HDF instead of the Serial... maybe this matters... :-|


Thanks for your answer

BRGDS

Alex

Post edited by: Alex, at: 2009/09/02 00:53

Post edited by: Alex, at: 2009/09/02 00:54<br /><br />Post edited by: Alex, at: 2009/09/02 00:54

Hi Forum

I solved the issue :

The problem is that i was not specifying correctly the include and lib folder of mpi.

In my case they were set to /usr/lib/openmpi/include and /usr/lib/openmpi/lib after i ran

update-alternatives --display mpi

followed by

update-alternatives --config mpi

to set my mpi to OpenMPI instead of mpich


I SOLVED IT....


Developers... :

It would be nice if one would not have to compile all the packages of Code saturne with MPI support, ... lots of potential users are not resourceful enough to fix this kind of nitty-witty details, and are precluded from using this wonderful piece of software... this is my opinion...


############ A CFD PRACTITIONER ISN'T NECESSARILY A GEEK #########

:-D

BRGDS


Alex<br /><br />Post edited by: Alex, at: 2009/09/02 01:07