MULHI Cache: An Instruction Cache Mechanism for VLIW Processors

摘要

VLIW (Very Long Instruction Word) processors, which are expected to be a next generation high performance microprocessor architecture, need a high-bandwidth, high-hit-rate instruction cache to fetch VLIWs and issue operations of each VLIW to function units quickly. However, when VLIWs including many nops (no operations) are stored in a conventional instruction cache, the cache utilization is not high, resulting in the performance degradation of VLIW processors. In this paper, a new instruction cache mechanism for VLIW processors, named MULHI (MULtiple HIt) cache, is proposed and evaluated using several programs in the SPEC95 benchmark suite. The experimental results indicate that the MULHI cache achieves 1.68 times higher performance than a conventional instruction cache that stores VLIWs with nops.