Efficient PairofReplacement Algorithms forLIandL2Cachefor Matrix Multiplication
ByRicha Gupta, Sanjiv Tokekar · 2009
fromLIistobeevicted tocreate aspace forthe Processors speedismuchfaster than required page.Various replacement algorithms, memory; tobridge this gapcache memoryisused. suchasLRU,FIFO,LFU [1,2]etcareusedto Thispaperproposes a preeminent pairof select thevictim page. replacement algorithms forLevel 1cache(LI) and LIishaving better temporal locality than Level2 cache(L2)respectively fortheMatrix L2,asL2isaccessed whenamissoccurs atL1. Multiplication (MM) application. Theaccess Thepageswhichareoften usedarelikely tobe patterns ofLIandL2aredifferent, whenCPUnot usedagain i.e. hotpagesshould remain inLIand getsthedesired datainLIthenitgoestoL2.Thus coldpagesshould betakenoffandareplaced in thereplacement algorithm whichworksefficientlythemainmemory. forLI maynotbeefficient forL2.Withthe Whatshouldbe thecriterion forL2 reference string ofMM,thepaperhasanalyzed the cache? Whichtypeofpagesshould remain inL2 behaviorof variousexisting replacement Cache? Whenever apageisevicted fromLIitwill algorithms at LI andL2 respectively. The beplaced onL2.Thenature ofpageswhichreside replacement algorithms whicharetakeninto onL2should beneither toocoldnortoohoti.e. consideration are.LeastRecently Used(LRU), moderate. PlaceforhotpagesisLIandthatfor LeastFrequently Used(LFU)andFirst InFirst coldpages isinthemainmemory[24].