面向稀疏计算的领域特定语言设计与实现  

Domain Specific Language for Sparse Computing

作　　者：胡燕婷 郑启龙[1,2] HU Yanting;ZHENG Qilong(School of Computer Science,University of Science and Technology of China,Hefei 230026,China;National High Performance Computing Center,University of Science and Technology of China,Hefei 230026,China)

机构地区：[1]中国科学技术大学计算机科学与技术学院,合肥230026 [2]中国科学技术大学国家高性能计算中心,合肥230026

出　　处：《小型微型计算机系统》2024年第5期1241-1248,共8页Journal of Chinese Computer Systems

基　　金：国家核高基重大专项项目(2012ZX01034-001-001)资助。

摘　　要：在稀疏计算领域应用模型的开发过程中,低层次的编程接口需要用户对底层硬件和算法模型同时具有全面、深入的了解.这种要求造成了稀疏计算领域的高门槛特性,不利于非专业人员的开发工作.为了降低稀疏计算领域的学习门槛,简化稀疏计算编程,提高编程的效率,本文提出了一种基于领域特定语言的解决方案,命名为Sparse Computing Programming Solutions:SCPS.SCPS设计和实现了专门的语法规则体系Sparse Computation by Domain Language:SCDL,并配备了相应的语法检查器.其次,SCPS会根据用户对数据和任务的划分进行代码优化,通过代码自动生成器将用户编写的应用模型转换为相关硬件的计算代码.为了提供良好的跨平台特性,代码生成器最终将生成相应模型的OpenCL代码.大规模的实验结果表明,使用SCDL语言构建的代码量仅为对应OpenCL代码的1/15,有效地提高了稀疏计算编程的效率.同时,实验证明了与串行代码相比,自动生成的代码能够达到优化的效果.In the development process of application models in sparse computing,low-level programming interfaces require users to have a comprehensive and in-depth understanding of both the underlying hardware and algorithm models.This requirement leads to the high threshold characteristic in the sparse computing field,which is not conducive to the development work of non-professionals.In order to reduce the learning threshold in the sparse computing field,simplify sparse computing programming,and improve the efficiency of programming,this paper proposes a domain-specific language-based solution,named Sparse Computing Programming Solutions:SCPS.SCPS has designed and implemented a special syntax rule system,Sparse Computation by Domain Language:SCDL,and is equipped with a corresponding syntax checker.In addition,SCPS will optimize the code according to the user's division of data and tasks,and convert the application model written by the user into the calculation code of related hardware through the automatic code generator.In order to provide good cross-platform features,the code generator will eventually generate OpenCL code for the corresponding model.Large-scale experimental results show that the amount of code built by SCDL language is only one-fifteenth of the corresponding OpenCL code,which effectively improves the efficiency of sparse computing programming.At the same time,the experiment proves that the automatically generated code can achieve the optimization effect compared with the serial code.

关 键 词：领域特定语言 稀疏计算 语法检查 代码自动生成 OPENCL 

分 类 号：TP312[自动化与计算机技术—计算机软件与理论]