Handling Architecture-Application Dynamic Behavior in Set-top Box Applications

This paper focuses on distributed optimization paradigm, based on vertical profiling. The proposal is based on enhance local optimization and peer-to-peer (P2P) source code compilation across application build flow. The earlier procedures incorporate additional steps for pre/post profiling during compilation, scheduling and linking phases. This iterative activity is carried out by methods based on wait-for-graphs or profile monitors. These methods introduce a centralized evaluation at source code (for each transformation and monitored parameters during successive approximation). The proposal introduces the asynchronous evaluation in source to source (sts) process. As a result the code transformation do not trapped in local optimization, rather look for global optimization (for both energy and cycle). Methodology is readily adaptable in an iterative compilation environment where application source code is optimized to satisfy user constraints on code size, energy, execution time and optimal target architecture usage. Experimental results show that our approach enhances parallelism up to 38% in G-721 speech codec, increases architecture correlation up to 17% in video transcodec H-264L, improves power efficiency as much as 43% for resVQ DSP algorithms. The technique incurs low overheads and enhances the application architecture correlation.