A State Transformation based Partitioning Technique using Dataflow Extraction for Software Binaries

Hardware-Software Partitioning and decompilation is a key issue in the Codesign of embedded systems. Partitioning in binary level helps in independent usage of software languages for the compilers. In this paper, the critical kernel of the software binary is relocated to the hardware and this is identified using instruction level profiling. The partitioned software binary is represented with initial and final state by a set of register value pairs. In the software binary the initial state to final state transformation is derived by equating the final state in terms of algebraic place holders, and then synthesized into hardware. A generalized decompiler is also designed to generate equivalent HDL for software binary block. The proposed method is applied to standard benchmarks and show significant speedup with lesser hardware resources. A source level partitioning is carried out for the scheduled elliptic wave filter and buffer size is estimated in binary and source level approach. Key words