This repository provides a prototype for the Axb paper and explores the effect of sparse storage formats on different sparse kernels. For details please check our supplement report.
- CMake
- C++ compiler (tested for GCC)
To install Axb, you should first clone the repository and then build it as shown below:
git clone https://github.com/leila68/Axb.git
cd Axb
mkdir build
cd build
cmake -DCMAKE_BUILD_TYPE=Release ..
make
Axb provides two different demo that each tests sparse matrix-vector multiplication (SpMV) and sparse lower triangular matrix solver (SpTRSV) for general sparse matrices in matrix market format and random banded matrices.
To test the two different kernels for matrix market formats, you should put the list of matrix paths, line by line, in a separate file and then emit the following:
Axb list_file_path.txt > out1.csv
An example input file containing matrix file paths is shown here.
To test different formats on a banded matrix, you should provide two
integer values, the matrix dimension dim and the number of bands
dl. The following command shows Axb evaluates banded matrices:
Axb dim dl > out2.csv
In either demo, Axb generates CSV files. Some example graphs based on a set of matrices are shown in section 3 of this the doc.
title={Sparsity-Aware Storage Format Selection},
author={Cheshmi, Kazem and Cheshmi, Leila and Dehnavi, Maryam Mehri},
booktitle={2018 International Conference on High Performance Computing \& Simulation (HPCS)},
pages={1034--1037},
year={2018},
organization={IEEE}
}