CS 420 Final Project

Final project for CS 420: OpenMP and MPI implementation of Cholesky factorization

To compile the code

bash build_and_run.sh

To run the code

cd build/
./Main -h # Show available options

# Example:
# Run OpenMP implementation of a newly generated matrix of
# dimension 32 with 2 threads
./Main -o -g -d 32 -p 2

# Run MPI implementation of a existing matrix of size 64
# on 4 MPI processes
mpirun -np 4 ./Main -m -d 64

Testing correctness with SciPy

• Required packages: NumPy, SciPy
• Dimension (dim) needs to match the corresponding file
• Unfactored SPD matrix file and corresponding result output must be readily available.
• To run the test, do

  python test.py

  with the correct matrix dimension

Cluster script generation

./scripts/submit_batch
python generate_submit_scripts.py

This will generate a series a subdirectories and submit scripts in sbatch format (weak scaling is currently disabled). Parameters can be customized by editing the generate_submit_scripts.py file.

Cluster script submission

Before submission, the code must be compiled as the submission script does not automatically build the project. We recommend uncommenting all file output in main.cpp and utils/cholesky.cpp to avoid outputing redundant large matrices.

./scripts/submit_batch # to show a list of available commands
./scripts/submit_batch -ma # mpi strong scaling
./scripts/submit_batch -oa # omp strong scaling