I have the same problem on IBM at the CINES. If you don't use the option WriteMullikenPop, your job will work with n(>1) processes. May be there is a problem (on IBM) within the parallel version of mulliken.F ?
Lionel GUEZ wrote:
Hello. I installed Siesta 1.3f1p on an IBM Power 4 machine. I compiled the parallel version of Siesta. Compilation was ok, without any error message. Then I tested Siesta with the input files: Examples/Vps/O.psf Examples/Vps/H.psf Examples/H2O/h20.fdf from the Siesta distribution. I ran the program with 1, 2 and 4 processes. I compared the results with those in: Examples/H2O/Out/h2o.out All three executions were ok. Then I tested Siesta (same binary executable) with the input files: Examples/Fe/Fe.fdf Examples/Vps/Fe.psf from the Siesta distribution. In "Fe.fdf", I commented out the line: DM.UseSaveDM true # to use continuation files I ran the program with 1 process. I compared the results with those in: Examples/Fe/Out/Fe.out This execution was ok. However, with 2 processes (same binary executable, same input files), the program does not work. It hangs on a call to MPI_reduce. Here is the call stack where the program hangs: * mpi_reduce_t, module mpi__r8_v, Interfaces.f90, line 2962 * mulliken, mulliken.F, line 341 * siesta, siesta.F, line 1798 Has anyone tried these particular "Fe" input files from the Siesta distribution, running with 2 processes? Sincerely, Lionel GUEZ Idris support team www.idris.fr
