CSR: Medium: Easy PRAM-Based High-Performance Parallel Programming with Immediate Concurrent Execution (ICE)

并行编程正处于危机之中：除了最坚定的（和财力雄厚的）软件供应商之外，它的困难使所有人望而却步，而除了常规程序之外，加速通常令人失望。仅仅通过在算法、编程语言、编译器、硬件、电源或应用程序方面的增量思想来增加并行计算的大量知识基础，尽管它们可能很有趣，但不太可能改变这一现实。我们的论点是，如果没有语言和体系结构的协同设计，就无法克服简单编程和高效并行化不规则程序的双重挑战。因此，我们正在开发一种新的易于编程的语言，称为ICE，作为由XMT、PRAM算法模型和ICE组成的生态系统的一部分，共同实现这一双重目标。UMD在过去十年中开发的XMT架构能够在不规则程序中利用细粒度并行。ICE基于成功的PRAM算法模型，为并行程序设计提供了丰富的理论基础，并为数百个问题提供了并行算法，其广泛的影响包括：（i）更简单的并行程序设计模型，导致更简洁和直观的代码;（ii）将程序员从并行程序设计中的许多复杂问题中解放出来，导致更高的生产力和对并行程序设计的接受;（iii）ICE和PRAM的接近意味着程序员将立即拥有丰富的？图书馆？并行算法而不必编写自己的;（iv）一种语言，特别适合于一种架构，这种架构在难以并行化的不规则应用程序上表现出前所未有的加速;（v）一种非常活跃的外延程序，已经看到XMT在许多研究和教育机构中使用。

Parallel programming is in crisis: its difficulty dissuades all but the most determined (and deep-pocketed)software vendors, and speedups are often disappointing for all but regular programs. Merely augmentingthe substantial knowledge-base of parallel computing with incremental ideas in algorithms, programminglanguages, compilers, hardware, power or applications, interesting as they may be, is unlikely to changethis reality.This project advances a powerful idea for drastically improving ease-of-programming within the context ofa holistic many-core research architecture called XMT. Our contention is that without the co-design oflanguage and architecture, one cannot conquer the twin challenges of easy programming and efficientparallelization of irregular programs. Therefore we are developing a new easy-to-program languagecalled ICE as part of ecosystem consisting of XMT, the PRAM algorithmic model, and ICE, that togetherdeliver on this twin goal. The XMT architecture, developed at UMD over the last decade, is capable ofexploiting fine-grained parallelism in irregular programs. ICE is based on the successful PRAM algorithmmodel, which provides a rich theory for parallel programming, and has led to published parallel algorithmsfor hundreds of problems.The broader impact of this project includes (i) much easier parallel programming model leading to moresuccinct and intuitive code; (ii) freeing the programmer of many complex issues in parallel programming,leading to greater productivity and acceptance of parallel programming; (iii) the nearness of ICE andPRAM means that programmers would instantly have a rich ?library? of parallel algorithms without havingto write their own; (iv) a language especially suitable to an architecture that has demonstratedunprecedented speedups on hard-to-parallelize irregular applications; and (v) a very active outreachprogram that has already seen XMT being used at many research and educational institutions.