I wanted to do some calculations with ScaLAPACK. To introduce myself to the API, I started with the "Hello World" example from Intel MKL Reference Guide.

While running this code with a square number of processes, everything is going fine, but when using 10 processes, the last process, which doesn't get aligned in the grid, does not call MPI_FINALIZE while calling BLACS_EXIT. I tried to investigate the problem and could create a workaround by checking the finalization state with MPI_FINALIZED and calling MPI_FINALIZE if necessary.

But: When I use the tools provided from Ubuntu 12.04 (gfortran 4.6.3, OpenMPI 1.4.3, BLACS 1.1) this problem does *NOT* appear. As a result, I would consider this a bug in Intel MKL.

I was using Intel Studio 2013.0.079 and OpenMPI 1.6.2 on SUSE Linux Enterprise Server 11.

Best regards!

1 contribution / 0 nouveau(x)
Reportez-vous à notre Notice d'optimisation pour plus d'informations sur les choix et l'optimisation des performances dans les produits logiciels Intel.