Mobile and Desktop Processors
Intel® Core™ processors, Intel Atom® processors, tools, and utilities
16766 ディスカッション

Cache replacement policy for Nehalem/SNB/IB?

idata
従業員
3,332件の閲覧回数

I am working on a simple cache simulator for recent Intel processors. It's used to provide profiling info for our cache optimization. Therefore, I don't need it to be very very accurate, but it can't be too different from the facts either.

I didn't find much information about cache line replacement policy. Does anyone know where I can look up these info? I am mostly interested in L2 and L3 cache replacement policies. Someone online said it's PLRU. Can anyone confirm that? If yes, is it a tree-PLRU or bit-PLRU?

Thanks.

0 件の賞賛
1 解決策
idata
従業員
2,021件の閲覧回数

Yes, it is PLRU: Three-LRU bit in specific.

元の投稿で解決策を見る

4 返答(返信)
idata
従業員
2,022件の閲覧回数

Yes, it is PLRU: Three-LRU bit in specific.

idata
従業員
2,021件の閲覧回数
idata
従業員
2,021件の閲覧回数

Wait, I just found a problem. How do you use 3 LRU bits to handle 16-way set associative L3 slices? I previously thought you were referring to using 3 bits in a 4-way set associative similar to SCC L2. (/servlet/JiveServlet/previewBody/5753-102-1-8879/L2cache.pdf http://communities.intel.com/servlet/JiveServlet/previewBody/5753-102-1-8879/L2cache.pdf)

Is it just a similar implementation where you use 5 bits to walk the tree?

idata
従業員
2,021件の閲覧回数

You may want to post this query to the Intel(R) Software Network forums: http://software.intel.com/en-us/forum Forums | Intel® Developer Zone

I am escalating this internally to see if this information is available through this support channel; still, please contact the Software Network forums.

返信