Embedded Program Specialization for Multiple Criteria Trade-offs

Abstract

Improvement of characteristics of embedded system programs is important, because of limited resources available for applications in mobile devices. Decreasing power consumption is especially important because of limited battery life and slow growth of battery capacities. Here we analyze a problem of program specialization with multiple criteria (execution time, power, memory, accuracy) in mind. We propose a framework of the methodology for assessing multiple criteria in the problem domain and describe high-level models for evaluating embedded program characteristics using Feature Diagrams. To improve the characteristics of embedded software algorithms, we use function approximation and data specialization (look-up tables). In a case study we analyze the characteristics and trade-offs of the implementations of cosine function. Ill. 7, bibl. 23 (in English; summaries in English, Russian and Lithuanian).