Compiling on the Bluegene machine at the UR

The CRC's wikipage on the bg/p system is found ​here.

The bluegene P (bg/p) architecture is unique for several reasons. In particular, there exist two ways to compile an executable:

• xlf90: compile for code execution on the login nodes
• bgxlf90: compile for code execution on the cluster

The cluster (or worker nodes, if you will) do not support linux kernels, meaning that it is not possible to run interactive jobs like you can on BlueHive. Hence, debugging must either be done with by compiling with xlf90 and running on the head node, or going through a cycle of compiling with bgxlf90, submitting a job, and looking at the resulting core files, if any.

As of 06/2010, bg/p does not have the module system as does BlueHive. However, the environment is such that the Makefile.bluegene.inc should be sufficient to correctly compile astrobear on the system.

Note that, unlike BlueHive, the Fortran 77 compiler must be explicitly defined in the Makefile.bluegene.inc. As mentioned above, if a version of the code is to be compiled and run on the login node, the compiler commands in this file must be changed accordingly (most simply by removing the "bg" which prepends the commands).