Rosko: Row Skipping Outer Products for Sparse Matrix Multiplication Kernels

TL;DR

Rosko introduces a novel row skipping outer product technique that significantly reduces computation and memory access in sparse matrix multiplication for neural networks, outperforming existing solutions.

Contribution

The paper presents Rosko, a new sparse matrix multiplication kernel that efficiently skips entire rows, reducing computation and memory use without auto-tuning.

Findings

Achieves up to 6.5x runtime reduction on CPUs

Effectively handles sparsities from 65% to 99.8%

Outperforms existing auto-tuning and vendor-optimized libraries

Abstract

We propose Rosko -- row skipping outer products -- for deriving sparse matrix multiplication (SpMM) kernels in reducing computation and memory access requirements of deep neural networks (DNNs). Rosko allows skipping of entire row computations during program execution with low sparsity-management overheads. We analytically derive sparse CPU kernels that adapt to given hardware characteristics to effectively utilize processor cores and minimize data movement without the need for auto-tuning or search space exploration. Rosko can be integrated with other outer product scheduling methods, allowing them to leverage row skipping by using Rosko's packing format to skip unnecessary computation. Rosko kernels outperform existing auto-tuning and search-based solutions as well as state-of-the-art vendor-optimized libraries on real hardware across a variety of neural network workloads. For matrices with sparsities ranging from 65% to 99.8% typically found in machine learning, Rosko kernels achieve up to a 6.5x runtime reduction on Intel and ARM CPUs.