A VLIW processor using both thread-level and instruction-level parallelism

摘要

VLIW processor's performance has been improved by using instruction-level parallelism. But it is difficult to achieve improvement in the performance of VUW processor which implemented many execution units. Because the instruction-level parallelism has a limitation of control dependencies. In this paper we propose a VLIW based instruction merge processor (VIM processor) which improves throughput. VIM processor uses not only instruction-level parallelism but also thread-level parallelism. We show the effectiveness of VIM processor by presenting experimental result.