Developing a BLAS library for the AMD AI Engine

TL;DR

This paper introduces AIEBLAS, an open-source BLAS library optimized for AMD AI Engine, enabling easier and more flexible use of spatial architectures for scientific computing beyond machine learning.

Contribution

The paper presents a reusable, customizable BLAS implementation tailored for AMD AI Engine, facilitating broader scientific application development on spatial architectures.

Findings

Open-source implementation of BLAS for AMD AI Engine

Designed for easy reuse and customization

Supports scientific computing beyond ML workloads

Abstract

Spatial (dataflow) computer architectures can mitigate the control and performance overhead of classical von Neumann architectures such as traditional CPUs. Driven by the popularity of Machine Learning (ML) workloads, spatial devices are being marketed as ML inference accelerators. Despite providing a rich software ecosystem for ML practitioners, their adoption in other scientific domains is hindered by the steep learning curve and lack of reusable software, which makes them inaccessible to non-experts. We present our ongoing project AIEBLAS, an open-source, expandable implementation of Basic Linear Algebra Routines (BLAS) for the AMD AI Engine. Numerical routines are designed to be easily reusable, customized, and composed in dataflow programs, leveraging the characteristics of the targeted device without requiring the user to deeply understand the underlying hardware and programming model.