Skip to content

Source code of the ICPP '22 paper: "TileSpMSpV: A Tiled Algorithm for Sparse Matrix-Sparse Vector Multiplication on GPUs" by Haonan Ji, Huimin Song, Shibo Lu, Zhou Jin, Guangming Tan, and Weifeng Liu.

Notifications You must be signed in to change notification settings

SuperScientificSoftwareLaboratory/TileSpMSpV

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

5 Commits
 
 
 
 

Repository files navigation

TileSpMSpV

TileSpMSpV is an open source code that utilizes TileSpMSpV to accelerate the BFS algorithm.


Paper information

Haonan Ji, Huimin Song, Shibo Lu, Zhou Jin, Guangming Tan and Weifeng Liu, "TileSpMSpV: A Tiled Algorithm for Sparse Matrix-Sparse Vector Multiplication on GPUs," Proceedings of the 51st International Conference on Parallel Processing (ICPP), 2022, pp. 1-11, DOI: https://doi.org/10.1145/3545008.3545028.

Contact us

If you have any questions about running the code, please contact Haonan Ji.

E-mail: [email protected]

Introduction

Sparse matrix-sparse vector multiplication (SpMSpV) is an important primitive for graph algorithms and machine learning applications. The sparsity of the input and output vectors makes its floating point efficiency in general lower than sparse matrix-vector multiplication (SpMV) and sparse matrix-matrix multiplication (SpGEMM). TileSpMSpV provides a tiled algorithm for BFS applying TileSpMSpV on GPUs including three kernels called Push-CSC, Push-CSR and Pull-CSC.

Installation

To better reproduce experiment results, we suggest an NVIDIA GPU with compute capability 8.6. TileBFS evaluation requires the CUDA GPU driver, the nvcc CUDA compiler, and the cuSPARSE library, all of them are included with the CUDA Toolkit. The artifacts have been tested on Ubuntu 18.04/20.04, and are expected to run correctly under other Linux distributions.

Execution of TileSpMSpV

Our test programs currently support input files encoded using the matrix market format. All matrix market datasets used in this evaluation are publicly available from the SuiteSparse Matrix Collection.

  1. The command 'make' generates an executable file 'test'.

make

  1. Run code on matrix data. Running the program requires three parameters: the number of threads, matrix path, and BFS source point.

./test 16 test.mtx 0

Release version

Apr 13,2023 Version Alpha

About

Source code of the ICPP '22 paper: "TileSpMSpV: A Tiled Algorithm for Sparse Matrix-Sparse Vector Multiplication on GPUs" by Haonan Ji, Huimin Song, Shibo Lu, Zhou Jin, Guangming Tan, and Weifeng Liu.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published