#Analyse: # #(jeweils 4-fach assoziativer cache, 512 bzw 256+256 k, #blocksize 16k, write-back, write-allocate, LRU-Strategie) # # non-unified cache: unified cache: words read 20036 22280 /demand fetches .8708 .9684 w copied back 5004 5004 /demand writes 2.5020 2.5020 total 25040 27284 /demand fetches 1.0883 1.1858 demand access: total 23008 23008 instr 17007 17007 data 6001 6001 read 4001 4001 write 2000 2000 #misses: total 5009 5570 instr 6 567 data 5003 5003 read 3752 3752 write 1251 1251 #(jeweils 4-fach assoziativer cache, unified cache, #blocksize 16k, write-back, write-allocate, LRU-Strategie) # 64 k 128 k 512 k words read 44008 32524 22280 /demand fetches 1.9127 1.4136 .9684 w copied back 8000 7004 5004 /demand writes 4.0000 3.5020 2.5020 total 52008 39528 27284 /demand fetches 2.2604 1.7180 1.1858 demand access: total 23008 23008 23008 instr 17007 17007 17007 data 6001 6001 6001 read 4001 4001 4001 write 2000 2000 2000 #misses: total 11002 8131 5570 instr 5001 2628 567 data 6001 5503 5003 read 4001 3752 3752 write 2000 1751 1251 #facit: #------ # #Grössere Caches bringen vor allem beim gut lokalisierten code etwas #(der wird offenbar dann nicht so leicht ersetzt). # #Vergleich der Strategien: #------------------------- # Random FIFO LRU words read 16136 22528 22280 /demand fetches 0.7013 0.98 .9684 w copied back 3876 5004 5004 /demand writes 1.9380 3.5020 2.5020 total 20012 27532 27284 /demand fetches 0.8698 1.1966 1.1858 #demand access: total 23008 23008 23008 instr 17007 17007 17007 data 6001 6001 6001 read 4001 4001 4001 write 2000 2000 2000 #misses: total 4034 5632 5570 instr 484 629 567 data 3550 5003 5003 read 2581 3752 3752 write 969 1251 1251