OptiSpMV: High-Performance Sparse Matrix-Vector Multiplication

Overview

OptiSpMV is a pioneering project designed to optimize the multiplication of sparse matrices by fat vectors, a crucial operation in scientific and engineering computations. Leveraging the power of High-Performance Computing (HPC), OptiSpMV employs parallel computing strategies through OpenMP and CUDA, aiming to significantly enhance computational efficiency across different sparse matrix formats: Compressed Sparse Row (CSR) and ELLPACK.

Project Structure

OptiSpMV/
├── CSR/
│   ├── matrixMultivectorProductCRS.* # Source code for the serial implementation of the CSR format
│   ├── matrixMultivectorProductCRSCUDA.* # Source code for the CUDA implementation of the CSR format
│   └── matrixMultivectorProductCRSOpenMP.* # Source code for the OpenMP implementation of the CSR format
├── ELLPACK/
│   ├── matrixMultivectorProductELLPACK.* # Source code for the serial implementation of the ELLPACK format
│   ├── matrixMultivectorProductELLPACKCUDA.* # Source code for the CUDA implementation of the ELLPACK format
│   └── matrixMultivectorProductELLPACKOpenMP.* # Source code for the OpenMP implementation of the ELLPACK format
├── scripts/
│   ├── *.sh # Scripts for performance analysis
│   └── *.sub # Submission scripts for HPC systems
├── utils/
│   ├── utils.* # Utility functions for matrix operations
│   └── MatrixDefinitions.h # Definitions of matrix structures
├── runCuda.cpp # Main program for the CUDA implementation
├── runOpenMP.cpp # Main program for the OpenMP implementation
└── results/
    └── images/ # Directory for storing performance analysis images

Getting Started

Prerequisites

• An HPC system with MPI and CUDA installed.
• GCC for compiling OpenMP programs.
• NVIDIA CUDA Toolkit for compiling CUDA programs.

Installation

1. Clone the repository to your local machine or HPC cluster.

   git clone https://github.com/AlexisBalayre/OptiSpMV.git

2. Navigate to the project directory:

   cd OptiSpMV

3. Compile the source code. A makefile is provided for convenience:

   make all

Running the Program

• To run the OpenMP implementation:

  ./runOpenMP

• To run the CUDA implementation:

  ./runCuda

Results