Is more training required on Performance Counters???

Is more training required on Performance Counters???

Just this week I received an email from a faculty member who asked if I would consider adding a module to our instructor lead faculty training course. The idea for the module would be based around performance counters. Speficially, the request read:
"Have you ever considered a module on performance counters?
perfex/PAPI/perfctr and the like? " - See -

My response is that I will take it under advisment and I'd like to post the question to the broader academic community. What's your take?

My 2nd response is that ISC has posted materials in the additional materials section of the Integrated Architecture & Tools course that addresses using performance counters with the VTune Performance Analyzer. So while ISC has not provided any training on PAPI nor tools such as perfex or perfctr - we have posted content on using the Vtune Analyzer (~ 4 hours worth) with an additional ~3-4 hours worth of training on using performance counters with VTune. Some of that training explores using core 2 counters to detemine if an app is experiencing false sharing, other labs we have posted use counters to explore effective cache utlization on codes such as matric multiply etc. The most in depth counter training tackles cycle accounting on core 2. Cycle Accounting is very powerful way to get at the heart of your application performance issues - no doubt about it. Such a module could easily fit into to any University course that uses "Computer Architecture: A Quantitative Approach - Hennessy, Patterson". With such tools concepts such as CPI (IPC), Cache Miss, and others become tangible quantities that any student can measure. So at your university - so you see the need for a in-depth Performance counters or cycle accounting module?

In the most recent released rev of our 3 day Facutly training (verison 2.0) course, the Integrated Architecture & tools module (3.5 hour module) uses VTune Analyzer& Performance counters to get at False Sharing & also cache utilization issues and this consumes maybe 30 minutes of lecture (then add extra time - maybe 30min - 1 hour for these labs). We had ommitted the deeper cycle accounting material for sake of time.

My question to the community is this:
Should ISC add a more in depth module on performance counters?

As noted, facultyhave access to much more in depth materialswhich are posted - just not taught live. Should we add more of this kind of content to our live training? - Note - link to this posted content is currently broken :-( but I have reported this and we are working on a fix.

Please give me your opinion!

Bob C

1 post / 0 new
For more complete information about compiler optimizations, see our Optimization Notice.