Hello
I have a question regarding the L1 cache profiler for MI_example and MOSI_SMP_bcast. I am running a uniprocessor setup on GEMS and was looking to collect L1 cache statistics for a number of benchmarks that I am running. However, with MI_example, I found that all L1 statistics (misses etc.) were showing up as 0. Tracing through the code, it looks like the bool TwoLevelCache has been set to false which through the code causes there to be no L1 stats collection. I tried setting the bool to true in $GEMS/protocols/standard_1level_SMP_protocol.sm but when running my benchmark, I get assertion failures related to the file Vector.h. Is there a workaround to this?
The second question I had involves using the MOSI_SMP_bcast protocol. Since I ran into the problems stated above with MI_example, I have started using MOSI_SMP_bcast which appears to give L1 cache statistics. However, my question is the following: I am using a uniprocessor setting with this protocol and I assume that with one processor, this protocol defaults to a standard cache implementation. Is there any special behavior that I need to be aware of with this protocol when there is only one processor?
Thanks!
|