The Improvement Of The Re-reference Interval Prediction Policy Of The Cache Replacement Policy

In recent years,with the development of hardware technology,the speed of the processor is faster and faster.However,the problem of the memory wall still exists,the speed of the memory does not match the speed of the processor.An important way to alleviate this mismatch is to use the cache as a bridge.The cache replacement policy has a direct impact on the performance of the processor.Therefore,the research of cache replacement algorithm has become a hot topic in recent years.In this paper,the research of cache replacement algorithm is mainly aimed at the shortcoming of re-reference interval prediction policy(RRIP),and propose the improvement policies.Aiming at the effect of RRIP on Cache,which can not eliminate the useless Cache blocks,this paper proposes a dynamic bypassing re-reference interval prediction policy(RRIPAB).The policy in the retention of re-reference interval chain and the insertion position of a new inserted block combination of dynamic bypassing policy which retain the advantages to reduce the loss rate for hybrid program mode and program thrashing mode,also increased bypass the useless block.When the Cache hit,RRIPAB uses the RRIP promoted policy.When the Cache is missing,RRIPAB chooses the dynamic insertion mode and dynamic bypassing policy,Which according to big or small of the bypassing probability to choose the dynamic RRIP policy.The RRIPAB policy proposed in this paper is to select the most suitable replacement policy of dynamic RRIP and dynamic bypassing policy for the current program in the running time.The use of LRU,RRIP and dynamic bypassing policy as a comparative experiment,the experimental results show that RRIPAB retains the advantage of RRIP and dynamic bypassing policy,reduce the loss rate of mixed program mode and thrashing program mode,Which is obviously lower than that of LRU and dynamic bypassing policy.Increasing the dynamic bypass has a lower loss rate and improve the IPC of almost all program type.According to the RRIP can not choose a more suitable insertion location according to the different types of program,this paper proposes a dynamic insert policy based on the mapping of the re-reference interval prediction(MIRRIP).The policy based on the temporal locality and design access memory map,according to each instruction will affect the reuse information of data,divided into 5 reuse possibilities.According to the spatial locality,the bypassing filter is designed,and the adjacent blocks are divided into one region,and the two states are set differently according to the information of each block.access memory mapping table and bypassing filter table to determine the reuse possibility of the new inserted block.Then,the new block is inserted into the corresponding position of the re-reference interval according to the reuse possibility.This policy enables Cache to better retain the useful Cache block.MIRRIP retains the RRIP policy to enhance the advantages of promoted fast,but also for different program types to change the insertion position,thereby enhancing the performance of Cache.The experiments show that the LRU,RRIP,and the dynamic insertion based on mapping policy reduce the missing rate and improve the IPC.