A High-Level Synthesis Design for a Scalable Hydrodynamic Simulation on OpenCL FPGA Platform
Changdao Du, Yoshiki Yamaguchi · 2019
FPGA shows its potential in heterogeneous reconfigurable computing systems in recent years. However, the traditional design method of FPGA has often been described as Hardware Description Languages (HDL), which blocked the way that adopting it to many applications. Nevertheless, both academia and industry attempt to improve the programmability of FPGA by using high-level languages as a development tool. In this paper, we use the Xilinx OpenCL platform SDAccel to accelerate a classic hydrodynamic simulation called Lattice Gas Cellular Automata (LGCA). We introduce a vectorization method of LGCA that binding the degree of execution and data bandwidth together by exploiting the flexibility of FPGA. This work also employs spatial and temporal parallelism to avoid simulation size restrictions and increase the simulation performance. We evaluate our technique on Xilinx VCU1525 board under FHP-III collision rule sets. The speed of LGCA simulation can reach to maximum 11073 million hits per second. Comparison reports to an Intel Core i7-6700 processor and Nvidia Quadro P5000 GPU are also given in this paper.