Labs for VTune and Advisor
--------------------------

1. Run your code with APS. 

   check for MPI usage 
   What is the percentage of MPI usage?
   Which MPI function is using most of the time 
   Is it a Collective? 
   What is the imbalance shown?

   does you App use also OMP? 
   What is the OMP imbalance
   How much serial code is shown? 
   Do you know where OMP bottlenecks are expected (if it applies try later VTune for Details) 

2. Run Program with VTune. 
   For MPI program use VTune on single rank e.g. rank 0. 
   Run Hotspot analysis - if possible, use the HW drivers. 
   Check GUI for the "traditional" profiling 
   Check Flame Graph output. 

3. Run HPC performance analysis. 
   Start with single Rank only. 
   Does your application use OMP. Check of OMP details. 

4. Generate Advisor roofline chart 
   Do you see a loop that is a good candidate for Optimization.  
   Did you notice this loop already in VTune experiments?
   Check if roofline analysis in Advisor GUI shows more information that the roofline.html


Alternative
-----------
   
   Explore some Gromacs APS output generated on GNR with 86c processor
   Play with Poisson miniapp : Generate APS,VTUNE,Roofline