CCF-SHF: CSR: Small: Compilation for Multi-core Processors with Limited Local Memories

多核处理器提供了在不增加功耗的情况下继续提高峰值性能的唯一途径。然而，在现代多核体系结构中，不仅要表达应用程序中的所有并行性，而且要通过有效的应用程序管理来利用可用的并行性，这都面临着严峻的挑战。硬件中缺乏内存虚拟化的趋势使这些挑战更加复杂，这在未来的处理器（如IBM Cell）和英特尔实验性的80核处理器中可以观察到。硬件中不支持内存管理，因为缓存一致性协议不能扩展到100或1000个内核，而且硬件中的内存管理消耗大量的能量。软件中的内存管理可以利用应用程序和数据特性来减少开销，但是它增加了应用程序程序员的负担。该项目旨在开发工具和技术来自动管理多核处理器每个核心中存在的有限本地存储器。除了高效执行之外，该项目的主要目标是减轻应用程序程序员精心制作应用程序的负担，将其划分并映射到核心上，以确保其正确执行和可移植性。

Multi-core processors provide the only way to continue improving peak performance without much increase in the power consumption. However, there are serious challenges not only in expressing all the parallelism in the application, but also in exploiting the available parallelism by efficient application management on modern multi-core architectures. These challenges are only compounded by the trend of the absence of memory virtualization in the hardware, that is observed in futuristic processors, like the IBM Cell, and the Intel experimental 80-core processor. Memory management cannot be supported in hardware, because cache coherency protocols do not scale to 100s or 1000s of cores, and also because memory management in hardware consumes significant energy. Memory management in software can exploit application and data characteristics to reduce the overhead, however, it increases the burden of the application programmer. This project aims to develop tools and techniques to automatically manage the limited local memories present in each of the cores of a multi-core processor. In addition to power-efficient execution, the main objective of the project is to relieve the application programmer of the burden of carefully crafting the application, dividing and mapping it onto the cores to ensure its correct execution and portability.