Compile SCALE in parallel mode

[Clara Rojas]

Hi! I'm trying to recompile the SCALE binaries in Linux. I did the build intructions as are in the manual, but I got some errors that I can't solve... Any idea would be very useful!!

The steps I followed are:

1. Installation of all programs (Fortran, C++ compiler, Lapack libraries, Cmake, and OpenMPI)

2. Setup configuration

              Parallel SCALE (with MPI support) 

              cp script/configure_scale_mpi.sh build/gcc 

              chmod u+x build/gcc/configure_scale_mpi.sh 

              /configure_scale_mpi.sh ../.. 

3. When I run 'make install' from build/gcc, I get this error message:

      [  0%] Built target teuchoscomm

make[2]: *** No rule to make target '/opt/vendors/gcc-4.8.3/shared/lib/liblapack.so', needed by 'Trilinos/packages/teuchos/numer             ics/src/libteuchosnumerics.so.6.2'.  Stop.

CMakeFiles/Makefile2:1603: recipe for target 'Trilinos/packages/teuchos/numerics/src/CMakeFiles/teuchosnumerics.dir/all' failed

make[1]: *** [Trilinos/packages/teuchos/numerics/src/CMakeFiles/teuchosnumerics.dir/all] Error 2

Makefile:126: recipe for target 'all' failed

make: *** [all] Error 2

I though that it's looking for the wrong directory where the libraries are installed: /opt/vendors/gcc-4.8.3/shared/lib/liblapack.so. But I have no idea where exactly these libraries are installed or placed. Maybe I did a wrong configuration at some step. Any idea about the source of error? 

Thank you so much!!

[Rob Lefebvre]

Hi Clara,

It appears you are using the deployed configure_scale_mpi.sh script as-is. This lists resource locations not likely on your machine, which is what appears to be the situation for LAPACK. 

If you have LAPACK on your machine you will want to update the script's LAPACK variable value to point to your system's install of LAPACK. If you do not have LAPACK, it is relatively straightforward to download, compile, and install. 

I recommend LAPACK 3.5.0, available from http://www.netlib.org/lapack/#_lapack_version_3_5_0. 

Best regards,

Rob Lefebvre

SCALE Code System

[Clara Rojas]

Hello, Rob. Thank you for your reply. I installed Lapack 3.5.0 and I managed to avoid the previous error. However, there is a new error related to lapack libs again, as you can see in the attached file. Is there a specific way to compile and install lapack once I downloaded? As I'm installing it, it seems that the libs .so are not being generating correctly.

Thank you again!

Regards,

Clara.

[Rob Lefebvre]

Hi Clara,

This error makes it look like the LAPACK location is still not being communicated to the build. 

Did you update the script's LAPACK variable value to point to the new installation location? Also, you will want to check that the location configured is the location that exists.

LAPACK=/opt/vendors/gcc-4.8.3/shared/lib/

with .so located as such

   -D TPL_LAPACK_LIBRARIES:STRING=${LAPACK}/liblapack.so \
   -D TPL_BLAS_LIBRARIES:STRING=${LAPACK}/libblas.so \

When you perform an 
ls -l /opt/vendors/gcc-4.8.3/shared/lib/liblapack.so

 does the file exist? 

If you conduct a verbose compile you will see the complete location of the expected library

make VERBOSE=1

You can check that the expected location is where you actually installed LAPACK.

[Clara Rojas]

Hello Rob! Finally, I could change the installation path of LAPACK and everything seems to work well. 

However, I found another error when I try to run the t6-depl sequence. As you can see in the attatched file, I get an 256 error code. I checked if another user had the same error, and I saw that the reason could be the MPI version I'm using. I installed the 1.8.5 versión of MPI, so I don't know why I'm still getting this error message. 

Any idea about how I could fix it? Thank you in advance!

[Rob Lefebvre]

Hi Clara,
Can you include your t6-depl input for me to test with?

What happens if you run it without mpiexec?