In this research we propose a highly predictable, low overhead andyet dynamic, memory allocation strategy for embedded systems withscratch-pad memory. A scratch-pad is a fast compiler-managedSRAM memory that replaces the hardware-managed cache. It ismotivated by its better real-time guarantees vs cache and by itssignificantly lower overheads in energy consumption, area andoverall runtime, even with a simple allocation scheme. Scratch-pad allocation methods primarilyare of two types. First, software-caching schemes emulate theworkings of a hardware cache in software. Instructions are insertedbefore each load/store to check the software-maintained cache tags.Such methods incur large overheads in runtime, code size, energyconsumption and SRAM space for tags and deliver poor real-timeguarantees, just like hardware caches. A second category ofalgorithms partitions variables at compile-time into the two banks.However, a drawback of such static allocation schemes is that theydo not account for dynamic program behavior. We propose a dynamic allocation methodology for global and stackdata and program code that (i) accounts for changing programrequirements at runtime (ii) has no software-caching tags (iii)requires no run-time checks (iv) has extremely low overheads, and(v) yields 100% predictable memory access times. In this methoddata that is about to be accessed frequently is copied into thescratch-pad using compiler-inserted code at fixed and infrequentpoints in the program. Earlier data is evicted if necessary. Whencompared to an existing static allocation scheme, results show thatour scheme reduces runtime by up to 39.8% and energy by up to31.3% on average for our benchmarks, depending on the SRAM sizeused. The actual gain depends on the SRAM size, but our resultsshow that close to the maximum benefit in run-time and energy isachieved for a substantial range of small SRAM sizes commonly foundin embedded systems. Our comparison with a direct mapped cache showsthat our method performs roughly as well as a cached architecture in runtime and energy while delivering better real-time benefits.
展开▼
