sferez / SSPP_Sparse_Matrix_CUDA

Small Scale Parallel Programming, Sparse Matrix multiplication with CUDA

Small Scale Parallel Computing

This study examined the performance of OpenMP and CUDA parallelization for Sparse matrix-vector multiplication (SpMV) on a hybrid CPU and GPU platform.

It implemented both programming models on a set of sparse matrices with varying sizes and densities.

Report

The report is available at Medium: Article

Structure

The following code folders are included in this project:

• 'input' contains the input matrices in matrix market format
• 'output' contains the sparsity pattern of some output matrices
• 'src/CUDA' contains the CUDA code parallelization
• 'src/OMP' contains the OpenMP code parallelization
• 'src' contains the custom class and python script to run the code

Requirements

• CMake
• CUDA
• OpenMP
• Python 3 (optional)

Build

Each folder contains a CMakeLists.txt file to build the code. Each parallelization is built separately for testing purpose.

Python script have been provided to run all the configurations and generate the results.

Authors

• @sferez