Synopsis

On this page and sub-pages we document the usage of a handful of tools that will help you profile your application.


!!!!ATTENTION!!!!

If you are not seeing the particular hardware event/counter you need special privileges in order to access some hardware events/counters. All you need to do on Snellius is submit a job with the constraint.....

#SBATCH --constraint=hwperf
#SBATCH --exclusive
  • It must be an exclusive job
  • You will get an allocation with Perf Event Paranoid set to 0

Documented tools

General information

Here is a open source tool list:

  • Extrae
  • Paraver
  • Dimemas
  • Score-P
  • Extra-P
  • Scalasca
  • Cube
  • TAU
  • pprof
  • cachegrind (part of valgrind)
  • kcachegrind/qcachegrind
  • gperftools

More tools listed: https://pop-coe.eu/partners/tools

Commercial/vendor based tool list:

  • Vampir
  • Intel Trace analyzer and collector
  • Intel VTune 
  • AMD µprof

Hardware performance counters

There are certain interfaces through which these counters are accessible. Apart from the call graph performance counters can provide the most detailed outlook of an application profile.

Interface list:

  • PAPI
  • perf (comes with the linux kernel)
  • likwid
  • No labels