VTune does not show L2 statistics

[Zhong_Liang_O_]

After running the VTune on Xeon Phi tutorial given here (http://software.intel.com/en-us/blogs/2013/01/08/ready-to-run-applications-from-multicore-platform-onto-intel-xeon-phi-coprocessor), we were unable to obtain a non-zero figure for the L2 cache hit/miss rates in the results, despite trying all possible amplxe-cl options for the xeon phi {knc-bandwidth, knc-general-exploration, knc-lightweight-hotspots}. We would greatly appreciate it if someone could help us shed light on this issue. 

Thank you!

Zhong Liang ONG

[Sumedh_N_Intel]

Can you please confirm that your application is running correctly: Pease check the terminal window in which you started Intel Vtune  to see if any errors have occured. If no errors have occured, then your can try the following:

During an Intel Vtune collection, the hardware events are sampled only after a certain number of hardware events have occured. This threshold value is after which the events are sampled is the "sample after" value in the Intel Vtune analysis. If the number of  sampled events are less than 1000, then these numbers may not be statistically valid. In which case you can try the following two things: 

1. Allow multiple runs from your project properties. This ensures that the events are not multiplexed and hence you do miss any events due to event multiplexing.
2. You can also try reducing the sample after value in your analysis. Empirically, your sample after value should be no lower than 10,000. To change the sample after value, select an analysis, for example, the general exploration analysis. Click on the "Copy" button. Change the events and sample after values as needed and, save the analysis with an appropriate name. This modified analysis should show up under the Custom Analysis tab. Run this analysis instead to use the newer sample after values. 

You can then use the formulae given at this link to calculate the metrics from the hardware event counts.

I hope this helps.