Understanding l2 subp* sector categories in computeprof

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi everyone,

(I recently posted a similar comment about the instructions issued 1/2 categories in computeprof. I decided to make this a separate post to separate the discussion about the different profiling categories.)

I’m doing some work with computeprof, and I’m trying to understand what the various l2 subp* sector categories mean. Do these represent the read/write accesses to the different banks of the L2 cache? If not, what do they represent (also, if this is explained in the documentation somewhere, that’d be great – I haven’t been able to find anything thus far)?

Thanks in advance for any help you may be able to provide!

Thanks,
Matt

2

Based on the nvprof --query-events, I got the following information:

l2_subp0_write_sector_misses:  Number of write misses in slice 0 of L2 cache. This increments by 1 for each 32-byte access.

Thus, it seems like these stats do indeed refer to the banks of the L2 cache. If someone knows that this is incorrect, please let me know.

Matt