Simple execution engine for testing SDP data flow with GPUs

Requires MPI (any). Requires pthreads Requires GPUDegrid Requires c++11 compatible compiler (i.e. gcc 4.8, PGI 15)

Build This now requires code from GPUDegrid. In the build directory invoke %> git clone https://github.com/SKA-ScienceDataProcessor/GPUDegrid Then, make sure mpic++ is in your path and invoke 'make'

Run Set MV2_SUPPORT_DPM=1 in your run script Use mpirun with 3 ranks

%> mpirun -N 3 -n 3 ./EEngine

Trouble-shooting

• "no rule to make GPUDegrid/degrid_gpu.cu required by degrid_gpu.o" You most likely didnt clone GPUDegrid. See above under "Build"
• MPID_Open_Port no implemented Make sure to set MV2_SUPPORT_DPM=1 on each node

Cluster specific utilities NVIDIA PSG cluster: source setup_psg.sh to load appropriate modules, set environment variables qsub sub.sh to submit a job on 3 nodes

Wilkes: source setup_wilkes.sh sbatch wilkes_submit.tesla