Xenos: Dataflow-Centric Optimization to Accelerate Model Inference on Edge Devices

TL;DR

Xenos introduces a dataflow-centric optimization framework for edge device model inference, significantly improving performance over prior operator-centric methods and extending to distributed multi-device setups.

Contribution

It proposes a novel dataflow-centric optimization approach with operator linking and DSP-aware splitting, outperforming existing frameworks like TVM and enabling distributed inference.

Findings

Reduces inference time by up to 84.9% with vertical optimization.

Achieves up to 96.2% speedup through horizontal dataflow optimization.

Outperforms TVM by 3.22x to 17.92x in inference speed.

Abstract

Edge computing has been emerging as a popular scenario for model inference. However, the inference performance on edge devices (e.g., Multi-Core DSP, FGPA, etc.) suffers from inefficiency due to the lack of highly optimized inference frameworks. Previous model inference frameworks are mainly developed in an operator-centric way, which provides insufficient acceleration to edge-based inference. Besides, the operator-centric framework incurs significant costs for continuous development and maintenance. In this paper, we propose Xenos, which can automatically conduct dataflow-centric optimization of the computation graph and accelerate inference in two dimensions. Vertically, Xenos develops operator linking technique to improve data locality by restructuring the inter-operator dataflow. Horizontally, Xenos develops DSP-aware operator split technique to enable higher parallelism across multiple DSP units. Our evaluation proves the effectiveness of vertical and horizontal dataflow optimization, which reduce the inference time by 21.2\%--84.9\% and 17.9\%--96.2\% , respectively. Besides, Xenos also outperforms the widely-used TVM by 3.22$\times$--17.92$\times$. Moreover, we extend Xenos to a distributed solution, which we call d-Xenos. d-Xenos employs multiple edge devices to jointly conduct the inference task and achieves a speedup of 3.68x--3.78x compared with the single device.