panjete / para-lu

Parallel Implementations of LU decpompositions, leveraging (i) pthreads and (ii) openMP

LU Decomposition

Implementation of parallel algorithms for Gaussian elimination to factor a dense N x N matrix into an upper-triangular and a lower-triangular matrix, sequentially, with incremental improvements by parallelisation using pthread and openmplibraries.

Running instructions

The detailed code functionalities and descriptions have been reported in the report.pdf. Further, we have commented and explained the modules and helper functions to our best attempts for easy readability.

We submit a Makefile, which compiles all the relevant files using the command make. The execution instructions are mentioned below.

File Structure

Sequential Files

Housed in sequential/.

seq_LU.cpp - is where a clean, commented, sequential version of the LU Decomposer is kept to serve as a reference and a baseline for comparison. Also prints exec_time and L2,1 norm of the residual array to aid in performance analysis.

To run the executable, simply execute ./seq.out <matrixDimension> 1. Here, the "1" is a vestigial remnant of the numthreads to be developed in the parallel codes.

OpenMP

Housed in openmp/.

openmp_LU.cpp - is where the OpenMP aided parallel LU is developed.

And run using ./omp.out <matrixDimension> <numThreads>

omp_LU_transpose.cpp - is an alternative version where we experimented with the Column Major Order Matrices instead.

pthreads

Housed in pthreads/.

pthread_LU.cpp and pthreads_LU_alt.cpp are the two variations of pthread implementations. The details can be found in the report.

And run using ./pth.out <matrixDimension> <numThreads>

Helpers

/plots directory houses the code for generating the plots for the times, efficiencies and speedups obtained. It is also were these generated plots are saved.

Team

Goonjan Saha
Gurarmaan S. Panjeta
Viraj Agashe