SI2 SSI: Collaborative Research: Sustained Innovations for Linear Algebra Software (SILAS)

随着串行处理器主导的计算机架构时代即将结束，处理器设计中几个前所未有的变化的融合已经产生了广泛的共识，即计算科学和工程的许多基本软件基础设施已经完全过时了。数学库历来一直处于软件的先锋地位，必须快速适应这种设计革命，因为它们是常见的低级软件主力，可以为许多不同类型的应用程序执行所有最基本的数学计算。线性代数软件持续创新 (SILAS) 项目更新了计算科学和工程历史上使用最广泛的两个数值库 — LAPACK 和 ScaLAPACK（缩写为 Sca/LAPACK） — 增强和强化了它们，以应对处理器架构和系统设计中正在进行的革命。 SILAS 创建了一个分层的软件组件包，能够在平台部署金字塔的各个级别上运行，从桌面到世界上最大的超级计算机。它实现了三个互补的目标： 1) 只要有可能，SILAS 通过许多计算科学家熟悉的 Sca/LAPACK 编程接口提供对最新算法、数值实现和性能的无缝访问； 2) 在必要时，SILAS 通过新的接口扩展提供先进的算法、数值实现和性能能力； 3) SILAS 提供了一个精心设计的渠道，通过该渠道，这些领域研究前沿的新发现可以尽快传递给所有依赖高性能线性代数的应用社区。 SILAS 中包含的改进和创新来自多种来源。它们代表了 PI 自己的算法和软件研究议程的结果（包括设计和经过良好测试的原型），该议程针对的是多核、混合和超大规模系统架构。它们是与用户、供应商以及大型 NSF 和 DOE 超级计算设施的管理人员广泛且持续互动的结果。它们源于与计算机科学和工程其他领域的跨学科接触，预测新架构和编程模型的需求和机遇。最后，它们来自研究社区对开发和提供现有 Sca/LAPACK 代码的增强版本的热情参与。SILAS 的主要影响是 Sca/LAPACK 库对计算科学许多分支的重要性的直接影响。 Sca/LAPACK 库是密集线性代数的社区标准，已被大量用户、计算中心和 HPC 供应商采用和/或支持。学习使用它们是许多领域和许多学术机构的计算科学家或工程师教育的基本部分。 Sca/LAPACK 历史上广泛使用的应用领域包括（以及许多其他示例）飞机机翼设计、雷达截面研究、船舶和其他离岸结构周围的流动、固体在液体中的扩散、降噪以及光通过小颗粒的扩散。此外，与 SILAS 合作增强和改造这些库以适应下一代平台的应用合作伙伴名单扩展了这一传统名单，包括量子化学、自适应网格细化方案、计算材料科学、地球物理流、随机模拟和“大数据”数据库研究。没有其他数字图书馆可以与社区进行如此广泛的整合。因此，我们有充分的理由相信，用最先进的方法和算法增强这些库，并使它们适应新兴平台（达到极限规模），将对研究和教育界、政府实验室和私营企业产生相应的巨大影响。

As the era of computer architectures dominated by serial processors comes to a close, the convergence of several unprecedented changes in processor design has produced a broad consensus that much of the essential software infrastructure of computational science and engineering is utterly obsolete. Math libraries have historically been in the vanguard of software that must be quickly adapted to such design revolutions because they are the common, low-level software workhorses that do all the most basic mathematical calculations for many different types of applications. The Sustained Innovation for Linear Algebra Software (SILAS) project updates two of the most widely used numerical libraries in the history of Computational Science and Engineering---LAPACK and ScaLAPACK, (abbreviated Sca/LAPACK)---enhancing and hardening them for this ongoing revolution in processor architecture and system design. SILAS creates a layered package of software components, capable of running at every level of the platform deployment pyramid, from the desktop to the largest supercomputers in the world. It achieves three complementary objectives: 1) Wherever possible, SILAS delivers seamless access to the most up-to-date algorithms, numerical implementations, and performance, by way of Sca/LAPACK programming interfaces that are familiar to many computational scientists; 2) Wherever necessary, SILAS makes advanced algorithms, numerical implementations and performance capabilities available through new interface extensions; and 3) SILAS provides a well engineered conduit through which new discoveries at the frontiers of research in these areas can be channeled as quickly as possible to all the application communities that depend on high performance linear algebra. The improvements and innovations included in SILAS derive from a variety of sources. They represent the results (including designs and well tested prototypes) of the PIs' own algorithmic and software research agenda, which has targeted multicore, hybrid and extreme scale system architectures. They are an outcome of extensive and on-going interactions with users, vendors, and the management of large NSF and DOE supercomputing facilities. They flow from cross-disciplinary engagement with other areas of computer science and engineering, anticipating the demands and opportunities of new architectures and programming models. And finally, they come from the enthusiastic participation of the research community in developing and offering enhanced versions of existing Sca/LAPACK codes.The primary impact of SILAS is a direct function of the importance of the Sca/LAPACK libraries to many branches of computational science. The Sca/LAPACK libraries are the community standard for dense linear algebra and have been adopted and/or supported by a large community of users, computing centers, and HPC vendors. Learning to use them is a basic part of the education of a computational scientist or engineer in many fields and at many academic institutions. Application domains where Sca/LAPACK have historically been heavily used include (among a host of other examples) airplane wing design, radar cross-section studies, flow around ships and other off-shore constructions, diffusion of solid bodies in a liquid, noise reduction, and diffusion of light through small particles. Moreover, the list of application partners working with SILAS to enhance and transform these libraries for next generation platforms expands this traditional list to include quantum chemistry, adaptive mesh refinement schemes, computational materials science, geophysical flows, stochastic simulation and database research for "big data". No other numerical library can claim this breadth of integration with the community. Thus, there is every reason to believe that enhancing these libraries with state of the art methods and algorithms and adapting them for new and emerging platforms (reaching up to extreme scale), will have a correspondingly large impact on the research and education community, government laboratories, and private industry.