An efficient FPGA-based function profiler for embedded system applications

A Composable Monitoring System for Heterogeneous Embedded Platforms

“…Aldham et al [4] presented LEAP, a HW profiler for FPGA-based embedded microprocessors, where gathered information is used to improve both performance and HW/SW partitioning. To improve this latter, Nadimpalli et al [29] proposed another profiler capable of identifying high computational loads to be moved to HW execution. Scheipel et al [39] present an approach for measuring execution time and events of an embedded system by integrating a dedicated performance monitoring infrastructure using HW/SW co-design techniques; on the other hand, Sadek et al [37] propose another infrastructure called STHEM, which is a collection of utilities for heterogeneous embedded image processing platforms.…”

“…The advent of FPGA-based SoCs opened up a plethora of new opportunities in the computing domain and, from the monitoring perspective, it introduced coprocessors as new elements to be monitored. SW monitoring could serve to identify parts of the application that benefit from a HW implementation [4,29] and some works address FPGA-based platforms [25,34,44], despite they do not give a comprehensive monitoring for such systems. To enable observability, FPGA vendors, such as Xilinx and Intel, are making available solutions to monitor SoC [5,45,46].…”

“…The monitoring components are built in a similar manner, thanks to the customization capabilities of the monitors that can be specialized for the events to be captured, but the APIs used to report the information at system level are the same. Moreover, designers can play with different customization options trading off monitoring overhead (in terms of resources, time and memory) versus the observation capabilities according to X X Seo et al [40] X X Seo et al [41] X X Rambo et al [35] X X X Aldham et al [4] X X Nadimpalli et al [29] X X Scheipel et al [39] X X Sadek et al [37] X X Patrigeon et al [34] X X X Doyle et al [13,25] X X X Valente et al [27,44] X X X X Kyung et al [26] X X Goeders et al [16] X X X X Hammouda et al [17] X X X X Fanni et. al [15,24] X X X X Lee et al [21] X X X Najem et al [30] X X X X X Zoni et al [11] X X X X X…”