HighP5: Programming using Partitioned Parallel Processing Spaces

作     者：Yanhaona, Muhammad Nur Grimshaw, Andrew Mickey, Shahriar Hasan 

作者机构：Brac University Bangladesh University of Virginia United States 

出 版 物：《Journal of the Brazilian Computer Society》 (J. Braz. Comput. Soc.)

年 卷 期：2024年第30卷第1期

页      面：653-687页

核心收录：

学科分类：0809[工学-电子科学与技术（可授工学、理学学位）] 08[工学] 0835[工学-软件工程] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

基　　金：The authors would like to thank Rich Knepper and Craig Stewart of Indiana University  K.S.M. Tozammel Hossain of University of North Texas  and Golam Rabiul Alam of Brac University for provid-ing access to their machines  compute cluster  and supercomputer for various experiments. This work is partially funded by the XSEDE: Extreme Science and Engineering Discovery Environment project of National Science Foundation  USA. The fund was granted to Andrew Grimshaw for his investigation on GFFS: a Global Federated File System that con-nects compute and storage resources across university campuses and supercomputing centers for fostering research collaboration 

主　　题：Parallel programming 

摘      要：HighP5 is a new high-level parallel programming language designed to help software developers to achieve three objectives simultaneously: programmer productivity, program portability, and superior program performance. HighP5 enables this by fostering a new programming paradigm that we call hardware-cognizant parallel programming. The paradigm uses a uniform hardware abstraction and a declarative programming syntax to allow programmers to write hardware feature-sensitive efficient programs without delving into the detail of those feature implementations. This paper is the first comprehensive description of HighP5’s design rationale, language grammar, and core features. It also discusses the runtime behavior of HighP5 programs. In addition, the paper presents preliminary results on program performance from HighP5 compilers on three different architectural plat-forms: shared-memory multiprocessors, distributed memory multi-computers, and hybrid GPU/multi-computers. © 2024, Brazilian Computing Society. All rights reserved.