Performance evaluation of floating point Differential Evolution hardware accelerator on FPGA

Abstract

Differential Evolution (DE) is a meta-heuristic algorithm widely being used for solving optimization problems. The computation time of the DE scales up with complexity of the problems. This limits its usage in real time embedded applications. This work compares the execution time of DE floating point algorithm implemented on a 32-bit PowerPC440 (PPC440) and a MicroBlaze (MB) processors. Further, the execution time of the DE hardware accelerator in SoC is compared with the execution time of floating point unit (FPU) enabled MicroBlaze and PPC440 processors operating at 125 MHz. For performance comparison, the DE hardware accelerator is used for solving three benchmark test functions of different complexities. It is observed that the hardware accelerator attained an acceleration of 6–8×, and 40–100× compared to PPC440 with hardware and software FPU respectively, whereas it attained an acceleration of 124–328× and 120–264× compared to MicroBlaze with software and hardware FPU respectively. Furthermore, resource utilization and power analysis of MicroBlaze and PPC440 with soft/hard FPU based SoC system on Virtex-5 ML507 platform are also reported.