Overseas Travel Grant for Dr L Freeman

计算机架构中的多核革命对HPC软件开发的未来具有深远的影响。二十多年来，高性能架构的用户，无论是单处理器还是同构并行计算机，都能够依赖于科学软件库，如LAPACK和ScaLAPACK，以提供一种高级编程方法，其中库软件封装了实现可扩展和可移植高性能所需的大部分编程工作。在这项提案中，我们计划启动旨在开发自适应运行时策略的研究，使未来的HPC数值线性代数软件能够利用新兴的算法和编程技术，用于新兴的多/众核架构。首先，我们将专注于单侧因子分解。

The multi-core revolution in computer architecture has profound implications for the future of HPC software development. For more than two decades, users of high performance architectures, both uniprocessors and homogeneous parallel computers, have been able to rely on scientific software libraries, such as LAPACK and ScaLAPACK, to provide a high-level programming approach in which the library software encapsulates much of the programming effort that is required to achieve scalable and portable high performance. In this proposal, we plan to initiate research that aims to develop self-adapting run-time strategies that will enable future HPC Numerical Linear Algebra software to take advantage of emerging algorithmic and programming techniques for the emerging multi/many-core architectures. Initially we will focus on one-sided factorisations.