CAREER: Scalable Compiler Optimizations for New Software

本研究计划的目标是为新兴软件应用开发编译器优化技术。与目前的实践状况相比，新的软件应用程序将用类型安全语言（例如Java）编写，将包含数百万行代码，通常由分布在多台机器上的组件构建，并将运行在具有多个控制线程的处理器上，但将花费数百个周期来访问主存储器。现代软件将需要现代优化，这些优化将需要在三个方面进行扩展：随着程序的大小，随着分布式组件的数量，以及随着内存延迟的增长。本研究建议通过利用全新形式的推测、分析、需求驱动分析、运行时检查和硬件支持来实现可伸缩性。为了开发这些技术，将进行三项具体努力：消除数组边界检查、需求驱动的指针分析和基于指针的数据结构的预取。

The goal of this research program is to develop compiler optimization technology for emerging software applications. Compared to the current state-of-the-practice, the new software applications will be written in type-safe languages (e.g., Java), will contain millions of lines of code, will often be built from components distributed across multiple machines, and will run on processors that will have multiple threads of control, but will take hundreds of cycles to access the main memory. Modern software will require modern optimizations that will need to be scalable in three ways: with the size of the program, with the number of distributed components, and with the growing memory latency. This research proposes to achieve scalability by exploiting radically novel forms of speculation, profiling, demand-driven analysis, run-time checks, and hardware support. To develop these technologies, three specific efforts will be undertaken: elimination of array-bounds checks, demand-driven pointer analysis, and prefetching for pointer-based data structures.