申威处理器硬件数据预取技术的实现

硬件数据预取技术可以有效提升处理器的访存性能,是申威处理器性能优化过程中亟需突破的一项技术。硬件开销和处理器架构的制约是硬件预取技术实现中的主要难点。借鉴学术界对硬件预取技术的研究成果和工业界的应用现状,紧密结合申威处理器的结构特点,研究了申威处理器硬件预取技术的实现方法。以流预取为例,在处理器核心面积增加0.97%的情况下,硬件预取技术的应用可以将目前申威处理器的整数性能平均提升5.17%,最高提升28.88%;浮点性能平均提升6.39%,最高提升30.11%。

Hardware-based data prefetcher implementation in Shenwei microprocessors

The technique of hardware based data prefetching can improve microprocessors’ memory access performance efficiently. Implementation of this technique is urgently needed in Shenwei microprocessors to boost their performance, but it faces the restrictions from hardware cost and architectural characteristics. Based on existing research achievements and current industrial applications of prefetching technique, we propose a hardware based data prefetcher implementation framework, which is highly coalescent with Shenwei microprocessors’ architectural characteristics. Take the stream prefetcher as an example, the hardware data prefetching technique can improve integer performance of Shenwei microprocessors by 5.17% on average, 28.88% at most; and improve float performance by 6.39% on average, 30.11% at most, while the chip area is only increased by 0.97%.