Memory Access Example
Repeatedly accessing 1MB of REAL*8 on the SP2
1MB = 244 pages * 4096 bytes;
4096 / 128 bytes = 32 lines/pg
128 bytes / 8 bytes = miss every 16 elements
32 misses/pg * 8 cycles/miss = 256 cycles/pg
36 cycle TLB miss per page
256 * (36 + 256) = 73728 cycles wasted per iteration