報告題目：Enabling Efficient and Scalable Parallelization for Data-Intensive Computations

報告時間：2024年5月22日15:00-16:00

報告地點：437bwin必贏國際官網大樓B405

報告人：邱俊喬

報告人國籍：中國

報告人單位：香港城市大學

報告人簡介：Dr. Junqiao QIU is an Assistant Professor in the Department of Computer Science at City University of Hong Kong. Prior to joining CityU, he was a tenure-track assistant professor at Michigan Technological University and earned his Ph.D. from the University of California Riverside. His research interests span the areas of compilers and systems, with a focus on enabling efficient parallel computing for data-intensive applications and those with irregular data access patterns. He is a recipient of the ACM SIGPLAN PAC Award, the NSF CRII Award, and the Best Paper Award at ASPLOS 2020.

邱俊喬博士是香港城市大學計算機科學系的助理教授。在加入城市大學之前，他曾在密歇根理工大學擔任助理教授，并在加利福尼亞大學河濱分校獲得博士學位。他的研究興趣涵蓋編譯器和系統領域，重點關注為數據密集型應用和具有不規則數據訪問模式的應用實現高效的并行計算。他曾獲得ACM SIGPLAN PAC獎、NSF CRII獎和ASPLOS 2020最佳論文獎。

報告摘要：Exploiting parallelism is crucial for achieving high-performance data processing on modern processors. However, many data processing routines still run serially due to the sequential nature of their underlying computation models. In this presentation, I will demonstrate how to effectively break inherent data dependencies and enable scalable and efficient data-parallel processing.

I will begin by introducing our previous work on using speculation to auto-parallelize bitstream processing applications. Following this, I will discuss our ongoing projects that push the boundaries of speculative parallelization. These include leveraging non-SIMD vector instructions to accelerate speculative parallelization, integrating speculation into pattern-aware graph mining applications, and enabling efficient concurrent GPU-based inferences.

Finally, I will conclude the talk by sharing my ideas on parallelizing more general applications, aiming to broaden the applicability of these techniques.

邀請人：李清安、袁夢霆