Parallel GPU job

#BSUB -o output
#BSUB -e error
#BSUB -R "span[ptile=12]"
#BSUB -n 24
#BSUB -q gpu

# Set up environment
. /etc/profile.d/modules.sh
module load openmpi/1.4.3/gcc
module load cuda/3.2/gcc

job=simpleMPI

NP=0

for TOKEN in $LSB_HOSTS
do
    (( NP++ ))
done

# MCA options
export OMPI_MCA_mtl=psm
export OMPI_MCA_btl=^openib,tcp

# Run job
cd $LS_SUBCWD

$MPIROOT/bin/mpirun -x LD_LIBRARY_PATH -np $NP $job