Evaluation of Software Inter-Processor Synchronization Methods for the Zynq-UltraScale+ Architecture

Abstract

Current embedded systems provide diverse functionalities and their features are evolving constantly. This is the case of the Zynq-UltraScale+ (US+) MPSoC family, where it is possible to find a System-on-Chip (SoC) architecture with an Application Processor Unit (APU) containing up to four Cortex A53 processing units, as well as Graphics Processor Units (GPUs) or Real-Time Processor Units (RPUs) in the same device. Nevertheless, the synchronization among these different units is crucial whether tackling a multi-core approach, especially in applications in standalone mode, without an Operating System (OS). In this work, two methods of synchronization among the four cores of the APU in a Zynq-US+ MPSoC device are presented. One of them is based on sending interrupts using the InterProcessor Interrupt (IPI), whereas the other is based on the use of atomic instructions and mutual exclusion variables (mutex). Both methods are managed by the exchange of messages between the processors, previously defined by and for the application. The experimental results presented here allow both proposals to be compared in terms of running times, also considering the particular cases of either a cascaded synchronization among the different cores or a broadcast synchronization among processors.