I have some queries regarding theCache-Blocking method for utilizing data locality. I am refering to the following article.
I know thatthe main factors to be taken into consideration are
- the L2 processor's cache size.
- number of iterations / re-use.
- the data block size as compared to the L2 cache size.
Question # 1
However, does anyone know whether the the Sector-Mapping featuresand the Set-Associativityof the L2 cache plays any part ?
Does anyone have any data on this ?
Question # 2
If system has an onboard L3 cache, would it not be advisable to utilize the data localityinside the larger L3 cachethan inside the L2 cache ?
- Mithun Shanbhag
(ps : My system has a 8-way sector-mapped, 512 KB (Unified) L2 cache witha cache line size of 64 bytes. It does not have any L3 cache.)