CAREER: Type-Directed Compilation

编译时类型信息被认为对函数式语言（如标准ML）的高效编译非常有价值，但是对于如何在真实的编译器中支持类型导向编译以及通过各种基于类型的优化可以获得多少性能增益却知之甚少。本研究的目标是设计和识别重要的面向类型的编译技术，这些技术可以显著提高机器学习程序的性能。这些新的编译技术有望带来以下好结果：(1)ML程序中的数据对象（包括递归数据结构）将能够使用与C程序中一样高效的未装箱数据表示；(2)通过跨模块类型专门化和内联扩展，有效实现多态函数和参数化模块；(3)传播到编译器后端的类型信息将用于辅助程序分析，如闭包分析和指令调度。CAREER奖的教育部分有两个目标：构建一个新的编译器实验室工具包（基于上述基于类型的ML编译器），并将最先进的编译器技术纳入本科和研究生课程。

Compile-time type information is believed to be very valuable in efficient compilation of functional languages such as Standard ML. But little is known on how to support type-directed compilation in real compilers and how much performance gain one can get from various type-based optimizations. The goal of this research is to design and identify important type-directed compilation techniques that can significantly improve the performance of ML programs. These new compilation techniques are expected to lead to the following nice results: (1) data objects (including recursive data structures) in ML programs will be able to use unboxed data representations that are as efficient as those in C programs; (2) polymorphic functions and parametrized modules will be implemented efficiently through cross-module type specialization and inlining expansion; (3) type information propagated into the compiler back end will be used to assist program analysis such as closure analysis and instruction scheduling. The educational component of this CAREER award seeks two objectives: the construction of a new compiler lab kit (based on the above type-based ML compiler) and the incorporation of state-of-the-art compiler technology into the undergraduate and graduate curriculum.