A Method for Compiling Domain-Specific Languages 1 into SYCL through Abstract Syntax Trees

XJTU 2024 CSUC Submission6 Authors

31 Mar 2024 (modified: 03 Apr 2024)XJTU 2024 CSUC SubmissionEveryoneRevisionsBibTeXCC BY 4.0
Keywords: SYCL, Domain-Specific Languages, Abstract Syntax Trees
TL;DR: The paper proposes an approach that translates domain-specific languages into abstract syntax trees and subsequently compiles them into SYCL.
Abstract: The escalating demand for computing resources, particularly in the realm of ar- 6 tificial intelligence (AI), necessitates efficient utilization of heterogeneous parallel 7 systems. This study focuses on compiling domain-specific languages, specifically 8 data-centric computation models, into SYCL for heterogeneous many-core systems. 9 SYCL, based on C++17, offers a unified programming model for various hardware 10 accelerators, promoting code reusability across different architectures. Leveraging 11 SYCL’s cross-hardware compatibility and performance optimization capabilities, 12 this project aims to enhance programming efficiency and performance on diverse 13 hardware backends. Through the translation of domain-specific languages into SYCL 14 (DPC++), developers can harness the simplicity and usability of domain-specific 15 languages while achieving high-performance parallel computing. This approach ad- 16 dresses the challenges of complex programming interfaces and poor program port- 17 ability across heterogeneous systems. By enabling domain-specific languages to run 18 in parallel on heterogeneous systems, this research contributes to advancing the de- 19 velopment of heterogeneous computing systems and providing programmers with 20 more flexible and efficient programming tools. The significance of this work lies 21 in its potential to facilitate broader application scenarios and higher execution effi- 22 ciency, ultimately promoting the widespread adoption of domain-specific languages 23 and driving innovation in parallel computing.
Submission Number: 6