Application-Specific Memory Management for Embedded Systems

We propose a methodology to improve the performance of embedded processors running data-intensive applications by allowing embedded software to manage on-chip memory on an application-specific or task-specific basis. We provide this management ability with a novel hardware mechanism, column caching. Column caching provides software with the ability to dynamically partition the cache. Data can be placed within a specified set of cache “columns” to avoid conflicts with other cached items. By mapping a column-sized region of memory to its own column, column caching can also provide the same functionality as a dedicated scratchpad memory including predictability for time-critical parts of a real-time application. Column caching enables the ability to dynamically change the ratio between scratchpad size and cache size for each application, or each task within an application. Thus, software has much finer software control of on-chip memory. We present column caching and techniques to automatically layout program data structures in a column cache; these techniques achieve significant improvements in performance because on-chip memory is utilized effectively.