Pipeline is a fundamental pattern to parallelize a series of stage tasks on input data in loops. Mainstream pipeline programming frameworks count on data abstractions to perform pipeline scheduling. Although this design is convenient for data-centric parallel applications, it is not efficient for algorithms that only exploit task parallelism in pipeline. To address these limitations, we leverage modern C++ to design a new task-parallel pipeline programming framework called Pipeflow. Pipeflow separates data abstractions and task scheduling, enabling an efficient implementation of task-parallel pipeline scheduling using lightweight atomic operations.
In this talk, I will cover four major components: 1. What is Pipeflow? 2. Why do we need a C++ task-parallel pipeline framework? 3. Designs, implementations, and deployments of the proposed Pipeflow programming model and scheduling algorithm. 4. Real use cases of Pipeflow and its performance advantages in industrial applications.
By the end of the presentation, the audience will know how to leverage Pipeflow to implement task-parallel pipeline algorithms in the applications.
View Slides
Mr. Cheng-Hsiang Chiu is now a Ph.D. student advised by Dr. Tsung-Wei Huang at University of Utah. Mr. Chiu's current research focuses on high performance, parallel and heterogeneous computing, mainly on developing libraries for programmers to write parallel applications easily. Mr. Chiu's active project is Taskflow.