Conclusions and Future Work
Power and cost efficient design, while meeting performance constraints on a (partly) programmable platform, has become a very important issue for a broad class of embedded applications. Many of the applications for low-cost, low-power design turn out to be data-dominated. Experiments have shown that for such data-dominated applications, a very large part of the power consumption is due to the data storage and data transfer; also on programmable platforms. Also the area cost (memory footprint) is fora large part dominated by the memories. However, until recently, outside our group (together with the universities directly co-opeation with us) and U.C.lrvine, little research has been done on systematically optimizing the storage and transfer of background data. Since the late 80’s our research group is convinced that the memory organisation should be optimized as a first step in the system design/compiler methodology for this type of applications. This has been formalized in the DTSE methodology which is presented here for the first time with its overall trajectory for platforms with partly predefined memory architectures.
KeywordsStorage Management Loop Transformation Programmable Platform Exploration Step Transform Source Code
Unable to display preview. Download preview PDF.