Functional Programming and Scientific Computing

这项研究的目标是设计、实现和实际应用一种支持科学代码开发的函数式编程语言。函数式语言和科学计算似乎出奇地不匹配。问题不是这些语言的表现力很好，而是它们的效率很差。最近对函数式语言的研究表明，如果在整体语言设计中注意，赋值实际上可能与指代透明度一致。本研究使用一种这样的语言，命令式Lambda演算(ILC)来开发随机过程建模领域的程序。虽然在顺序处理机上效率低下一直是函数式语言的一个缺点，但它们在并行处理机上的效率一直是一个强大的吸引力。本研究的一个重要目标是获得高效的ILC并行实现。这项工作成功的可能性很高，因为ILC中赋值的存在避免了到命令式的转换，让编译器只处理(众所周知的)高阶函数和懒惰求值本身的问题。为了充分利用现代并行体系结构，特别是非统一内存访问机，需要通过对ILC的并行语言扩展来提供对数据放置和任务调度的控制。

The goal of this research is the design, implementation, and realistic application of a functional programming language supporting the development of scientific codes. Functional languages and scientific computing appear to be singularly mismatched. The problem is not with the expressiveness of these languages, which is excellent, but their efficiency, which is poor. Recent research in functional languages suggests that assignment may in fact be consistent with referential transparency, if care is taken in the overall language design. This research uses one such language, the Imperative Lambda Calculus (ILC), to develop programs in the domain of random process modelling. Although inefficiency on sequential processors has long been a drawback of functional languages, their efficiency on parallel processors has been a strong attraction. An important goal of this research is to obtain an efficient parallel implementation of ILC. The potential for success in this endeavor is high because the presence of assignment in ILC obviates the transformation to imperative form, leaving the compiler to handle only the (well understood) problems of higher-order functions and lazy evaluation themselves. To fully exploit modern parallel architectures, especially non-uniform memory access machines, control over the placement of data and scheduling of tasks needs to be provided by way of parallel language extensions to ILC.