Machine Learning based SpMV Optimization

Overview

Replication package for sparse matrix-vector multiplication

/AutoML: Source code for training and testing ML models using AutoML tool

/spmv: Source code of our approach (SpMV algorithm)

Requirements

• OS: Ubuntu 18.04
• CUDA: >= 10.2
• CUSP: 0.5.1

Usage

1. Makefile Setup

Setting CUDA_PATH to local cuda path

CUDA_PATH = /usr/loca/cuda

Setting CUSP_INCLUDES to local cusplibrary path

CUSP_INCLUDES = path2cusplibrary

2. Compile:

make

3. Run:

./spmv matrixfile