CAREER: Semantics, Abstractions, and Tools for a Pragmatic Verified LLVM Compiler

职业：经过验证的实用 LLVM 编译器的语义、抽象和工具

• 批准号: 1453086
• 负责人: Santosh Nagarakatte
• 金额: $ 54.5万
• 依托单位: Rutgers University New Brunswick
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2015
• 资助国家: 美国
• 起止时间: 2015-01-15 至 2020-12-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1453086&HistoricalAwards=false
• 关键词: CAREER; Semantics; Abstractions; Tools; Pragmatic

Title: : CAREER: Semantics, Abstractions, and Tools for a Pragmatic Verified LLVM CompilerCompilers are crucial components of the computing ecosystem. They can transform source programs written in multiple languages (e.g, C/C++/Java) into multiple target architectures (e.g, x86/ARM). Compiler bugs, however, break properties enforced at the level of source programs and can lead to unintended application behavior and disasters in safety-critical domains. This project aims to develop pragmatic and lightweight formal techniques for making mainstream compilers robust, in a manner that can be easily adopted by compiler developers. The intellectual merit of this project is the development of real-world compiler verification environments drawing influence from multiple areas like programming languages, architecture, and concurrency. The project's broader significance and importance are: (1) improving the quality of a large number of software projects by improving the robustness of the compiler, (2) inculcating formal reasoning for building large systems (compilers in particular) that are correct by construction, and (3) educating high school students, undergraduates, and graduate students for developing software with lightweight formal methods.This project aims to build new techniques and tools that can check the correctness of optimizations with the help of the developer. The project achieves this by (1) designing mathematically-precise semantics for various components of the mainstream LLVM compiler, (2) designing domain specific languages for writing and specifying LLVM optimizations, which not only check optimizations for correctness but also generate efficient C++ implementations, (3) designing techniques to build precise translation validators, which compare the source and the target code with assistance from the compiler developer. Technology developed by this research will not only improve the reliability of compilers but also the reliability of other large software systems which depend on correct compilation.

标题：：CAREER：Semantics，Abstractions，and Tools for a Pragmatic Verified LLVM语义、抽象和工具LLVM是计算生态系统的重要组成部分。它们可以将用多种语言（例如C/C++/Java）编写的源程序转换为多种目标体系结构（例如x86/ARM）。然而，更复杂的错误会破坏在源程序级别强制执行的属性，并可能导致意外的应用程序行为和安全关键域中的灾难。这个项目的目的是开发实用和轻量级的形式化技术，使主流编译器健壮，编译器开发人员可以很容易地采用的方式。这个项目的智力价值是开发真实世界的编译器验证环境，从编程语言，体系结构和并发性等多个领域汲取影响。该项目更广泛的意义和重要性是：（1）通过提高编译器的健壮性来提高大量软件项目的质量，（2）灌输构建大型系统的形式化推理（特别是编译器）是正确的建设，和（3）教育高中生，本科生，这个项目的目的是建立新的技术和工具，可以在开发人员的帮助下检查优化的正确性。该项目通过（1）为主流LLVM编译器的各种组件设计精确的语义，（2）设计用于编写和指定LLVM优化的领域特定语言，不仅检查优化的正确性，还生成高效的C++实现，（3）设计构建精确翻译验证器的技术，它在编译器开发人员的帮助下比较源代码和目标代码。本研究所开发的技术不仅可以提高编译器的可靠性，而且可以提高其它依赖于正确编译的大型软件系统的可靠性。

项目成果

Approximating trigonometric functions for posits using the CORDIC method

使用 CORDIC 方法近似三角函数

• DOI: 10.1145/3387902.3392632
• 发表时间: 2020
• 期刊: CF '20: Proceedings of the 17th ACM International Conference on Computing Frontiers
• 作者: Lim, Jay P;Shachnai, Matan;Nagarakatte, Santosh
• 通讯作者: Nagarakatte, Santosh

Debugging and detecting numerical errors in computation with posits

• DOI: 10.1145/3385412.3386004
• 发表时间: 2020-06
• 期刊: Proceedings of the 41st ACM SIGPLAN Conference on Programming Language Design and Implementation
• 作者: Sangeeta Chowdhary;Jay P. Lim;Santosh Nagarakatte