Installing on UF HPC

UF HPC has all of the dependencies we need in the module system. CMake cannot automatically find the dependencies using the module system. We need to use a script similar to this to specify all the dependencies in the CMake:

module load cmake cuda python/2.7.3.shared boost 
cmake .. \
  -DCMAKE_C_FLAGS="-O3 -march=native" \
  -DCMAKE_CXX_FLAGS="-O3 -march=native" \
  -DCUDA_TOOLKIT_ROOT_DIR=${HPC_CUDA_DIR} \
  -DBOOST_ROOT=${BOOST_ROOT} \
  -DPYTHON_LIBRARIES=${PYTHONHOME}/lib/libpython2.7.so \
  -DPYTHON_INCLUDE_DIR=${PYTHONHOME}/include/python2.7

First line loads the required modules. It basically sets the environment variables that we are using in the next line. Next line is a CMake command that sets the value of several CMake variables from environment variables.

In order to test or run swarm applications, you would need to submit jobs to `gpu` queue. For example:

#!/bin/bash
#
#PBS -q gpu
#PBS -l nodes=1:ppn=1:gpus=1:fermi
#PBS -l walltime=00:30:00
#PBS -N SwarmTest
#PBS -d /path/to/swarm/build

echo On Host $HOSTNAME
export CUDA_DEVICE=`sed -e 's/.*-gpu//g' < $PBS_GPUFILE`
echo Running on GPU: $CUDA_DEVICE

make test

On the 7th line, change the part after -d to the absolute path to your Swarm-NG build directory, (e.g. the directory from which you run cmake).