A JIT Compiler for Neural Network Inference

Abstract

This paper describes a C++ library that compiles neural network models at runtime into machine code that performs inference. This approach in general promises to achieve the best performance possible since it is able to integrate statically known properties of the network directly into the code. In our experiments on the NAO V6 platform, it outperforms existing implementations significantly on small networks, while being inferior on large networks. The library was already part of the B-Human code release 2018 [12], but has been extended since and is now available as a standalone version that can be integrated into any C++14 code base [18].