cache miss rate is much lower as well (Alpha data is from
[1, 9]
3
, PowerPC data was collected using Aria [7]) .
On the other hand, the figures show that the data cache
miss rate is much higher for FP benchmarks, suggesting
that memory-hierarchy oriented research should put an em-
phasis on improving CFP2000 benchmarks as well. How-
ever, this isn’t the case: only 35% (11/31) of the memory-
hierarchy papers use a majority of CFP2000 or CFP95, in
contradiction to the 61% (19/31) of papers that use a ma-
jority of CINT2000 or CINT95. While we can’t claim that
every memory paper that neglects CFP2000 is remiss, we
feel that CFP2000 should be certainly given consideration
in papers that explore data cache performance.