GROW: A Row-Stationary Sparse-Dense GEMM Accelerator for Memory-Efficient Graph Convolutional Neural Networks

TL;DR

GROW is a novel GCN accelerator that employs Gustavson's algorithm for sparse-dense GEMM, optimizing data movement and achieving higher energy efficiency for graph convolutional neural networks.

Contribution

It introduces a row-stationary sparse-dense GEMM architecture specifically designed for GCNs, improving energy efficiency over existing accelerators.

Findings

Significant energy-efficiency improvements over state-of-the-art GCN accelerators.

Effective software/hardware co-design balancing locality and parallelism.

Utilization of Gustavson's algorithm for optimized sparse-dense matrix multiplication.

Abstract

Graph convolutional neural networks (GCNs) have emerged as a key technology in various application domains where the input data is relational. A unique property of GCNs is that its two primary execution stages, aggregation and combination, exhibit drastically different dataflows. Consequently, prior GCN accelerators tackle this research space by casting the aggregation and combination stages as a series of sparse-dense matrix multiplication. However, prior work frequently suffers from inefficient data movements, leaving significant performance left on the table. We present GROW, a GCN accelerator based on Gustavson's algorithm to architect a row-wise product based sparse-dense GEMM accelerator. GROW co-designs the software/hardware that strikes a balance in locality and parallelism for GCNs, achieving significant energy-efficiency improvements vs. state-of-the-art GCN accelerators.