What data can we gather?

• What was the time spent exclusively and inclusively in each function?
• How many times was f1 called?
• How many profiled functions (subroutines) did f1 call?
• Whats the mean time/call for each function?
• What was the maximum(and minimum and mean and total) time spent in f1 over all nodes? contexts? threads?
• For each invocation of f1 what was the exclusive and inclusive time spent in it? (Trace)
• Can we replace "Time" by "flops"? Instructions issued? Cycles? (HW counters)
• Can we profile at a level at which the user understands the modules? Give a high level view of the application?
• Can we provide both a high level abstraction for profiling and allow low level hardware related quantities to be profiled?
• Can we profile only Communication functions? Comm + IO? (Selective Profiling)
• Can we profile a set of statements (finer granularity) instead of functions? Can we profile blocks such as for loops?
• Can we calculate statistics such as standard deviation of the exclusive time spent in each function?
• Does it work with Threads? pthreads? NexusThreads?