Approximate computing is an emerging computing paradigm that aims to achieve energy-efficient digital systems with acceptable errors. Although several approximation techniques at different levels of system designs have been proposed, the question "which part of the circuit design or program can be approximated?" has not been well answered yet This work aims to answer this question by developing a software framework to systematically analyze the target application code and guide the designers about the part of the program to which approximation can or cannot be applied. Experimental results show that the proposed framework can successfully extract the significance ranking of variables/inputs to be approximated more quickly than existing statistical works.
展开▼