Steps for Profiling with Oracle SolarisStudio Performance Analyzer:

 

Steps for Profiling with Oracle SolarisStudio Performance Analyzer:

 

• Compile with optimization flags and the ‘-g’ flag

      

• Set up the environment:

 

setenv PATH "/path_to_profiler/solarisstudio-12.5b/bin:/path_to_jdk/jdk1.8.0_71/bin:$PATH"

 

setenv LD_LIBRARY_PATH "/path_to_jdk/jdk1.8.0_71/lib:~/jdk1.8.0_71/jre/lib"

 

• Run the application and collect samples in an experiment: The name of the experiment is printed in the console when the application starts, say /tmp/test.0.er. Running collect by itself would show the range of values for some of its parameters on the given machine.

 

For a serial application:

                                collect -p 1 -d /tmp

 

For profiling only the master of a MPI job, run it like this:

mpirun -f ./hosts -perhost 1 -np 4 ./prof.sh

 

where prof.sh is a script with content:

 

#! /bin/csh

if ($PMI_RANK == 0) then

   collect -p 1 -d /tmp $*

else

    $*

endif

 

• Analyze the experiment:
  analyzer

                                (In the GUI, load the experiment by browsing to it)

 

References:

Oracle Performance Analyzer Documentation