PDC 2023 summer school project

The task of the project is to parallelize the code in energy_storms.c with OpenMP, HIP and MPI. This work is done by Zimin, Ziyi and Qian.

How to compile

The parallelized version are energy_storms_hip.hip, energy_storms_mpi.c, and energy_storms_omp.c.

Compling all files through the Makefile:

make clean
make all

This will generate energy_storm_seq, energy_storm_omp, energy_storm_hip and energy_storm_mpi.

How to run

When running on Dardel, we can first use the env.sh script to setup the enviroment. Then, we need to allocate the resource and then run the program. For example, for MPI, we can first allocate the node by:

./mpi_salloc.sh

Then, to run the MPI version:

srun -n 32 ./energy_storms_mpi 30000 test_files/test_02_a30k_p20k_w1

Check omp_salloc.sh, hip_salloc.sh and mpi_salloc.sh to see how to allocate the node for OpenMP, HIP and MPI. Example of usage can be found in omp_time.sh, hip_time.sh and mpi_time.sh.

Repository structure

Apart from the files that are mentioned in above. We also have:

• result repository that contain the files generated from omp_time.sh, hip_time.sh and mpi_time.sh. They contain the running time of OpenMP, HIP and MPI on different layer size and number of particles.
• notebook/graphs.ipynb is the jupyter notebook that generates the plot showing how all program versions' running time scales with the layer size and number of particles.
• energy_storms_mpi_new.c and energy_storms_omp_new.c are our alternative solutions for MPI and OpenMP.