Muhammad Rashid, Thierry Goubier, Bernard Pottier (2007) A High Level Generic Application Analysis Methodology For Early Design Space Exploration In: Workshop on Design and Architecture of Signal and Image Processing (DASIP) Abstract: The software implementations of sophisticated multimedia applications/algorithms are often huge and it is virtually impossible to analyze these applications without generic automated tools and appropriate methodologies. Architectural implementation choices for these applications based on merely
designer experience without objective measures can lead to costly re-design loops. Application analysis at algorithmic level can produce a variety of useful information providing valuable design
space exploration indications.
We present a generic application analysis approach based on instrumentation based profiling for early design space exploration. First we transform the source specification of application implemented in a high level language (Smalltalk in this paper) into an internal trace tree representation by dynamic analysis. The trace tree of the source specification is then characterized/explored at the algorithmic level. The results of characterization provide guidelines to the designer to select target architecture(s) for the application. These guidelines include
memory, control and processing orientations as well as the inherited spatial parallelism in the specification. The aim is to improve application architecture matching by bridging the gap between application specification and target architecture. As a case study, we have taken MPEG-2 decoder implemented in Smalltalk. Experimental results show the applicability of the the proposed methodology for early design space exploration.
Notes:
