Is l1tex__data_bank_reads.sum*128bytes the L1 traffic？
Databanks are used to access pieces of data that are smaller than cache lines. To help understand the cache architecture and access process, I recommend you check out these 2 GTC talks. Each bank is a 4 byte access so your metric would be closer if you used 4 instead of 128. Also, your metric is only counting reads, so you may want to include a write metric as well if you are interested in that.
So it only includes the times of access to data bank in shared memory instead of the whole L1 cache? Because I don’t find the information about data banks in the first talk.
databanks exist for all the L1 memory including shared and l1tex. Perhaps the concept was only described in one of the talks.
Get it! Thanks for your reply!
This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.