Speed up an x86 SAME simulator using synthesizable SystemC timing models

Jianfeng An,Yi (Estelle) Wang

DOI: https://doi.org/10.1109/MWSCAS.2016.7870148

2016-10-01

Abstract:There are many simulators to evaluate the performance of computer architecture, however they are commonly based on Software Architecture Model Execution (SAME). Due to hardly implement parallel execution on multicore platforms, their simulation speeds are slow. To solve the problem, we propose to improve simulation speed of SAME simulators on FPGA using SystemC synthesizable timing models. First, we need substitute the C++ timing model in the original x86 SAME simulator by our SystemC timing model. Second, we convert our SystemC timing model into Verilog timing model aided by Xilinx Vivado tool. Finally, we develop necessary components to support PCIe communication between the timing model running on FPGA (a Xilinx VC707 evaluation board) and the functional model running on the host machine, including a PCIe DMA controller and a PCIe driver. Our experiment results show that the proposed simulator is 47∼91 times faster than original x86 SAME simulator. In general, our work could provide a feasible solution for designers to improve the simulation speed of SAME simulators.