Charm++ parallel programming system

["Bohm, Eric J" <ebohm AT illinois.edu>]

You can also add [notrace] to disable tracing on entry methods you don't want tracked.  That is a compile time switch, so you can’t adjust that choice dynamically.

The best method I’ve found is to trace for a few iterations at a time.

This can be controlled by bracketing such intervals using traceBegin() and traceEnd()

e.g., 

if(iteration % startInterval ==0)

traceBegin()

 if(iteration % endInterval ==0)

traceEnd()

note: traceBegin() and traceEnd() control tracing within a PE, so for many cases it is easiest to manage those by putting the calls inside similarly named entry methods in a group that you trigger by broadcast.

You can also flip tracing on if some other detectable phenomena is going on (or will soon start) that you want to trace, like before and after load balancing, or changes in refinement.

On Dec 20, 2019 2:57 PM, Jozsef Bakosi <jbakosi AT lanl.gov> wrote:

Hi folks,

I would like to collect projections data from a long-running simulation.
I only would like to generate Timelines and Usage profile (under the
Tools menu). Also, I'm really only interested in the cost of a few specific
functions. What is the best way to go about this so data collection with
-tracemode projections do not interfere too much with my simulation?

Should I do +traceoff and instrument the code with
traceRegisterUserEvent() and traceUserEvent()?

How can I ensure that only specific functions are traced?

Thanks,
Jozsef