Hi all,
I just want to make clear how the sharing
miss account for the total memory access. I am using parallel benchmarks SPLASH-2.
So I want to make clear the size of the sharing access of the total memory
accesses. I have done some experiments like this:
1.
Construct a map< line_address,
vector<coreID> >. When a core makeRequest() to the ruby module, call a
function to sign the core number.
2.
After finishing the execution,
check the map to see whether a memory access has been called by different
cores. If yes, it is a sharing access, else, it is an individual access.
3.
Account the sharing proportion of
the total memory.
However, the results shows that the sharing
access account for >80%, which seems to be wrong.
So I wonder is there any advices for me to
get the correct proportion the sharing memory accesses take.
Thanks in advances!
shuchang