RTL IP abstraction into optimized embedded software

Modern SoCs gain a high level of parallelism by using both general purpose processors and a number of data processing entities (DPE), dedicated to certain heavy functionalities. As a consequence, most systems devote DPEs to executing functions with high performance rather than using dedicated hardware. Reusing already existing and pre-verified IPs through abstraction methodologies is a key idea to meet time-to-market requirements and to reduce the error risk. Rough abstraction techniques lead to non efficient software code, that results in being very limited by hardware communication protocols and data types. This paper proposes an abstraction methodology that produces optimized code. Protocol refinement and data redefinitions are exploited to increase software performance. The effectiveness of the methodology has been proven by applying it to industrial designs.