next up previous
Next: HPMCOUNT and its options Up: Running HPMCOUNT Previous: Investigating serial code

Investigating parallel code

When investigating the performance of a parallel code named parallel_prog.x, the line
  poe parallel_prog.x
in the original LoadLeveler script should be replaced by the line
  poe /usr/local/packages/actc/hpmtk/pwr4/bin/hpmcount parallel_prog.x
Important: The ordering of the calls to poe and hpmcount is crucial. Having them in the wrong order or missing poe altogether will lead to HPMCOUNT examining the performance of poe instead of your application.

Joachim Hein