使用Halide框架設計和優化OpenVX應用

Abstract

在本研究中，我們研究如何使用特定領域程式語言 – Halide來建構框架以便快速設計和優化OpenVX連通圖。Halide是一款高階影像處理語言，其提供開發者在寫程式時將演算法和排程分離，使開發環境變得更友善，Halide也已被證明是一種用於寫高效能影像處理程式的有效系統。我們利用OpenVX和Halide建構框架以執行影像處理，因為Halide具備OpenVX所缺乏的原語排程，故我們使用Halide來執行OpenVX kernels，此方法使開發者能增加更多開發性並達到更好的效能。我們使用五個實驗來測試，所得到的結果顯示使用Halide搭配OpenVX能顯著提升影像處理及卷積神經網路的效能。

In this study, we investigate how to use a Domain-Specific Language – Halide to build a framework for fast prototyping and optimization of OpenVX graphs. Halide is a new high-level image processing pipeline language. It offers developers to separate the program into algorithms and schedule. This makes developers program friendly. The Halide image processing language has also proven to be an effective system for authoring high-performance image processing code. We built a framework with OpenVX and Halide to implement the image processing system. Since OpenVX is a lack of scheduling primitives, but Halide does. We implemented Halide into OpenVX graphs. This method can significantly improve the performance of image processing and convolutional neural networks.