This preview has intentionally blurred sections. Sign up to view the full version.
Unformatted text preview: High spatial locality: larger blocks, because more local memory words are brought in. (3pts total, 2pts for correct answer, 1pt for explanation) High temporal locality: smaller blocks, because there will be more addressable blocks in the cache. c) (10pts total, 3pts loads, 3pts stores, 2pts branches, 2pts the rest) New CPI: 0.5*1+0.2*(0.05*30+0.95*1)+0.1*(0.05*60+0.95*30)+0.2*2 = or 0.5*1+0.2*(0.05*31+0.95*1)+0.1*(0.05*60+0.95*30)+0.2*2 = 4a) (5pts) The bypass paths from MEM to EX can be eliminated. b) (8pts total, 2pts each, -1pt for every answer not listed) 0x104 - $2 0x108 - $3 0x110 - $1 0x114 - $1 c) (7pts total, -5pts if semantics violated, -1pt per NOP) 0x100 0x10c 0x104 0x110 0x108 0x118 0x114 2) 5) (2pts each) PC2: 0x10c PC3: 0x108 PC4: 0x104 (2pts each) IR2: add IR3: add IR4: lw (1pt each) WE: 0x1 ALU_Asel: 0x1 RW: 0x1 ALU_Bsel: 0x1 Asel: 0x0 WBsel: 0x1 Bsel: 0x2 WriteEnable: 0x1...
View Full Document
- Spring '07
- CPU cache, Locality of reference, correct rounding, 16-bit FP, correct addition