ReOPN: Retargetable Statistical Optimisation and Parallelisation for Network Processors

网络处理器是经过优化以尽可能高的速度支持网络协议的专用设备。对性能、成本和能源消耗的严格约束导致了具有特定于域的指令集扩展和特定于应用的处理元件的异类多核体系结构的发展。然而，这种适应对这些设备的可编程性产生了强烈的影响，并使软件开发成为一项既耗时又昂贵的任务。该项目研究网络处理器的可重定向和自适应编译器指导的优化和并行化的新的统计方法，以克服由于网络协议和应用的迅速增加的复杂性而引起的迫在眉睫的软件开发危机。这项研究将使高级编程语言的高效实现成为可能，并在对成本和性能高度敏感的嵌入式计算领域提高程序员的生产率。

Network processors are application specific devices optimised to support network protocols at the highest possible speed. Tight constraints on performance, cost, and energy consumption have led to the development of heterogeneous multi-core architectures with domain specific instruction set extensions and application specific processing elements. However, this adaptation has a strong impact on the programmability of these devices and makes software development a time-consuming and expensive task. This project investigates novel statistical approaches to retargetable and adaptive compiler directed optimisation and parallelisation for network processors to conquer a looming software development crisis caused by the rapidly increasing complexity of network protocols and applications. This research will enable the efficient implementation of high-level programming languages and increase programmer productivity in a highly cost and performance sensitive embedded computing domain.