DSA Monitoring Framework for HW/SW Partitioning of Application Kernels leveraging VPs

Abstract:
Rapid and efficient design space exploration is the key enabler for effective Domain Specific Architectures (DSAs) development. In this paper, we present a DSA monitoring framework that supports users in determining an effective HW/SW partitioning for SW application kernels. Our framework leverages Virtual Prototypes (VPs) to record HW metrics when executing the application kernels. The core of our framework is a novel monitoring approach using runtime code manipulation on the VP binary. The approach does not necessitate any modification of the VP or SW. We demonstrate the quality of our approach on a RISC-V VP platform running canny edge detection on video frames.