Lebwohl Lasher Accelerated Computing Project

SCIFM0004 Accelerated Computing Project: The Lebwohl-Lasher Model of Liquid Crystals.

This project experiments with different acceleration methods (NumPy Vectorisation, Numba, Cython, openMP, MPI) and various method combinations to speed up a Python program that simulates the 2D Lebwohl-Lasher model.

1. Navigating the Repository:

Each method/ version of the program has its own folder containing:

• The modified version of the script (e.g. numba_no_vectorisation.py)
• Set up and run scripts (for building and running cython versions)
• 'Outputs': Folder containing output files (times and parameter values)
• 'Tests & Validation Checks': Folder containing plots and statisitcs used to verify that each modified version still works correctly and produces the expected results. These plots and stats are produced by running the script 'Testing_strategy_Script.py' using the command:

python Testing_Strategy_Script.py <original_code_output_file.txt> <new_version_output_file.txt>

• Profiling file detailing which functions are doing the most work in the program (e.g. prof_V1_NUmba.txt)

The 'Scaling Tests' folder investigates how each method/version scales with the size of the model and number of threads (for parallelised versions). It contains:

• 'Scaling_plots.ipynb': Jupyter Notebook plotting the data to produce scaling graphs.
• Data files (csv) required for the Jupyter Notebook
• Output files for all runs

'execution_times_log.xlsx': This is a log of all the changes made for each verison and their execution times for various configurations.

2. How to Run the Programs:

All programs for this project were ran using HPC on Blue Crystal 4 through a submission of a slurm file. These are provided in the folder 'Slurm Files (HPC)'

Command Line Usage

Original, NumPy Vectorised, and Numba (including parallel) versions:

$ python <filename> <ITERATIONS> <LATTICE SIZE> <TEMPERATURE> <PLOTFLAG>

Cython (non parallel) versions, first compile the program by running the setup script followed by the run script:

$ python <setup script filename> build_ext -fi

$ python <run script filename> <ITERATIONS> <LATTICE SIZE> <TEMPERATURE> <PLOTFLAG>

Parallel Cython runs the setup file in the same way but inlcudes an additional argument for the run script. The number of threads were also specified in the slurm file:

$ python <parallel run script> <ITERATIONS> <LATTICE SIZE> <TEMPERATURE> <PLOTFLAG> <NUM THREADS>

Parallel Numba versions were run once before timings were recorded to allow cache to store memory. The number of threads were specified in the slurm file.

MPI (mpi4py) versions:

$ mpiexec -n <NUM PROCESSES/WORKERS> python <mpi script filename> <ITERATIONS> <LATTICE SIZE> <TEMPERATURE> <PLOTFLAG>