Designing adaptable processors by extending static compilation techniques with dynamic architectural reconfiguration.

专门为给定应用程序设计处理器通常涉及两个不同的阶段。第一阶段确定应用程序的计算密集型部分，以确定处理器的构建块。在下一个阶段，通过编译实现应用程序的功能。由于每个应用程序都有自己不同的计算需求和性能需求，因此根据所执行的应用程序的类型，这两个阶段的变化是可以预期的。因此，根据应用程序特定需求的适应性在设计周期的这两个阶段都是必不可少的。这种两阶段适应性在当前的方法中通常是不存在的。处理器体系结构通常具有固定的组合并使用静态编译技术。我们提出了一种由可定制的编译流引导的动态适应性架构。整个过程是应用程序驱动的，其中静态编译技术通过动态体系结构重新配置得到扩展。运行时反馈机制引入了对体系结构和编译过程的修改。

Designing a processor specifically for a given application typically involves two distinct phases. The first phase identifies the computationally intensive parts of the application to determine the building blocks of the processor. In the next phase, the application’s functionality is realised through compilation. As each application has its own diverse computational demands and performance requirements, variations in both these phases are to be expected depending on the type of application executed. Consequently, adaptability according to application-specific requirements is essential in both these phases of the design cycle. This two-phase adaptability is typically absent in present day methodologies. More often the processor architecture has a fixed composition and uses static compilation techniques. We propose a dynamically adaptable architecture that is steered by a customisable compilation flow. The entire process is application-driven, where static compilation techniques are extended with dynamic architectural reconfiguration. A run-time feedback mechanism introduces modifications to the architecture and the compilation process.