Packet filters play an essential role for network traffic/ security management in today?s Internet. To make software-based packet filters so fast that could work even in suffering from DOS attacks, the authors are challenging to effectively combine both the higher-level optimization relating to algorithmic structure adapted to the input packet sequence and the lower-level one relating to acceleration techniques in compiler study. In this paper, we focus on the lower-level (machine code) optimization using software pipelining, and report preliminary experimental results indicating that our approach has great potential to accelerate packet filter performance. Our code optimization achieves more than ten times higher performance compared with a conventional interpreter-based packet filter for each of seven typical fragments of packet filter rules.