Often times the best approach is to expand the "biggest red bar" on each level until
you find the methods that are driving the performance of the application.
Click image for larger view.
In this example the calls to Math.log(double), Math.abs(float),
and Math.max(double,double) in Logarithm.SS(..) are driving
the performance of the FFT command.
Before you go off thinking that those methods are "slow" look in the "Invocations"
column and note that they are being called millions of times...
Once you have identified the "slow" part of your primitive it is now time to ask
yourself, "How can I make it faster?"