A Dynamic IQ-Capping Technique for Simultaneous Multi-Threading Processors

Simultaneous multithreading which is called SMT is a technique provides a platform to improve the overall performance of superscalar CPUs with hardware multithreading. SMT allows multiple independent threads to execute at the same time and shares key data-path to achieve better resource utilization. Because of the transient behaviors of threads, designing an effective architecture to better utilize critical resources among threads is still a challenge in SMT. Such as Issue Queue (IQ), which is one of the most critical shared resources in the pipeline. To better utilize shared IQ resource, we propose a simple dynamic algorithm to accommodate the transient behavior of different threads and reduce over-capping situation for each threads. This algorithm dynamically adjusts the cap value for each thread according to its activeness in terms of its IQ occupancy time. The simulation results show that the proposed technique achieves a significant improvement in terms of IPC throughput over the regular no-capping technique.