Performance Analysis of Standalone and In-Fpga LEON3 Processors for Use in Deep Space Missions

When considering a new processor for a mission, one of the first questions that comes up is: “How does this processor compare with what we have used in the past?” Manufacturers use benchmarks to compare normalized performance from one product to another, but often these data are incomplete and are missing key parameters such as compiler version, compile options used, memory type, and many others. In this work, we present carefully documented benchmarking results for single-core and multi-core Gaisler LEON3 processors. We evaluate both standalone (UT699, UT700, GR712RC-multi-core)and in-FPGA soft-core (single-core LEON3, quad-core LEON3)processors. In the case of multi-core processors, we use the OpenMP API with RTEMS-5.0 real-time operating system to demonstrate performance gains with automatic parallelization of computationally intensive code segments. With increasing demand for high-performance FPGAs on space missions, the option to drop in a soft-core processor into the system design should be on the table, provided there are available FPGA resources. We show that the soft-core LEON3 implementation has comparable frequency-normalized performance and is a good alternative or supplement to a dedicated standalone hard-core processor. By adding one or more soft-core processors into the FPGA design, the data systems architect may eliminate additional hardware, save on overall power, and improve net system performance by optimizing HW-to-SW data transfers. On-board processing, communications, and data management tasks can benefit from a tightly coupled processor inside an FPGA. We present various architectural design considerations that impact the performance and resource utilization of LEON3 processors inside a Microsemi RTG4 FPGA.