The SMAUG C++ API

Introduction

SMAUG is a deep learning framework that enables end-to-end simulation of DL models on custom SoCs with a variety of hardware accelerators. SMAUG is designed to enable DNN researchers to rapidly evaluate different accelerator and SoC designs and perform hardware-software co-design. Simulation is powered by the gem5-Aladdin SoC simulator, allowing users to easily write new hardware accelerators and integrate them into SMAUG for testing and exploration.

SMAUG provides stable Python and C++ APIs, allowing users to work at varying levels of abstraction. This site will describe the C++ APIs and provide tutorials on how to work with them. It is divided into the following sections:

• Overall execution flow
  Overview of how SMAUG consumes and processes a model.
• Building a custom operator
  How to build your own operator and integrate it with the Python API.
• Tiling optimizers in SMAUG
  How to write a custom tiling optimizer.
• SMAUG in simulation
  Differences in behavior when simulating or running natively on the host machine.