Design of Programming Languages and Computing Performance

在过去的十年中，编程语言的大量涌现为编程语言特性的实验创造了丰富的环境。编程语言社区已经了解了程序员的喜好，以及对某些计算领域有吸引力的特定功能。社区还发现，一些对程序员有吸引力的功能会限制代码生成器和优化器生成高效代码的能力，从而导致显著的性能成本。然而，很少有研究试图将语言特性与编译器优化技术和硬件设计联系起来。 这一研究计划的长期愿景是促进编程语言的开发，并帮助设计和实施支持软件堆栈，以促进计算解决方案的快速原型设计，同时保持良好的执行性能。对这一目标的贡献包括：(I)表征计算机程序行为的方法和研究；(Ii)测量硬件对同步和通信的支持以及硬件加速器设计的实际限制的工具；(Iii)将特定语言特征映射到阻碍性能的困难或低效代码生成的责任分析。在短期内，研究计划将开发一种方法来表征由数据输入的变化引起的程序行为变化；它将开发一种具体的方法来确定事务性存储器的硬件支持的实际实施中的限制；它还将有助于设计一种新的编程语言，该语言旨在提供有助于快速原型解决方案的语言特征，但在包括语言特征时非常明智，以避免可能阻碍代码生成器的工作并导致性能降低的特征。

The vast proliferation of programming languages in the past decade has created a rich environment for experimentation with programming language features. The programming language community has learned much about programmers' preferences and about specific features that are attractive to certain domains of computing. The community has also discovered that some features that are attractive to programmers do incurr a significant performance cost by limiting the ability of code generators and optimizers to produce efficient code. However, very few studies have attempted to connect language features with compiler optimization techniques and hardware design. The long-term vision for this research program is to contribute to the development of programming languages and also to the design and implementation of the supporting software stack that facilitates rapid prototyping of computing solutions while sustaining good execution performance. The contributions toward this aim include: (i) methodology and studies to characterize the behaviour of computer programs; (ii) tools to measure the actual limitations both of hardware support for synchronization and communication and of hardware accelerator designs; (iii) a blame analysis to map specific language features to difficult, or inefficient, code generation that hinders performance. In the short term the research program will develop a methodology to characterize program-behaviour changes that are caused by changes to data inputs; it will develop a specific methodology to determine the limitations in actual implementations of hardware support for transactional memories; and it will also contribute for the design of a new programming language that aims to deliver language features that facilitate rapid prototyping of solutions but that is very judicious in the inclusion of language features to avoid features that could hinder the job of a code generator and lead to lower performance.