Application Performance Snapshot (APS), available as part of the Intel 2018 Parallel Studio suite, provides a broad characterization of various factors impacting performance, as well as an overall summary highlighting key performance issues.

Video Transcript
Key points:
  • The aps command instruments the code to produce the Application Performance Snapshot:
    mpiexec -n 128 aps --r ./aps_results_128 ./main.out
  • The Intel tool suite continues to expand to provide integrative and more user-friendly summaries of the broad set of performance information that it collects.
  • APS for this target application indicates that the application is MPI-bound, but this is not obviously the case, given that APS also reports a substantial fraction of back-end stalls, which would suggest problems with the memory pipeline. The presenter notes: "If the code is indeed MPI bound, the degree to which it is, is dwarfed by memory pipeline bottlenecks."
 
©  |   Cornell University    |   Center for Advanced Computing    |   Copyright Statement    |   Inclusivity Statement