DISTRIBUTED SHARED MEMORY TYPE PARALLEL COMPUTER AND INSTRUCTION SCHEDULING METHOD

摘要

PROBLEM TO BE SOLVED: To improve effective performance by using a pre-fetch instruction in an NUMA type parallel computer. SOLUTION: When a pre-fetch instruction is output to a pre-fetch unit 7, an address designated by an operand is converted to a physical address by an address converting part 1, and it is determined by a pre-fetch type determining part 2 whether it is NUMA pre-fetch instruction or full pre-fetch instruction (conventional pre-fetch instruction). In the case of NUMA, it is determined by an address determining part 3 whether the physical address is local or remote, and when local is determined, the processing is ended by an instruction completion processing part 6. When remote is determined, cache check is made by a cache tag check part 4. In the case of cache hit, the processing is ended, and in the case of miss-hit, a pre-fetch request is given from the pre-fetch request part 5 to the request receiving unit. Thus, the data is pre-fetched from the remote main storage to the cache in the local main storage.