Estimating Configuration Parameters of Pipelines for Accelerating N-Body Simulations with an FPGA using High-level Synthesis

Tetsu Narumi, Akio Muramatsu


In the era of the IoT (Internet of Things) and Edge computing, SoC (System on Chip) with an FPGA (Field Programmable Gate Array) is a suitable solution for embedded systems because it supports running rich operating systems on general-purpose CPUs, as well as the FPGA’s acceleration for specific computing. One problem of designing an accelerator on an FPGA is that optimization of the logic for the accelerator is not automatic and much trial and error is needed before attaining peak performance from the SoC. In this paper we propose a method to reduce the development time of the accelerator using N-body simulation as a target application. Based on the hardware resources needed for several pipelines of the accelerator and their performance estimation model, we can estimate how many pipelines can be implemented on an SoC. In addition, the amount of memory each pipeline requires for attaining maximum performance is suggested. Our model agreed with the actual calculation speed for different constraining conditions.


Paper Citation