Notes 2021 01 07

MPI Tools WG Meeting 2020-12-17

Participants

• Marc-Andre Hermanns
• Bill Williams
• Joachim Protze
• Wesley Bland
• Martin Schulz

Topics

QMPI_Pcontrol

• Argonne customer wants an interface to communicate different phases of an applications

  • Caliper (LLNL Tool) should be the solution to that

• Existing MPI_Pcontrol semantics are terrible to begin with

• Maybe a key-value store (MPI_Info?, binary values?) could solve this

  • Binary keys and values could be a valuable
  • MPI_Info_key_create("name", &index) to create an integer ID for a string key?

• QMPI_Pcontrol could also be used for user instrumentation

  • Extraction of information from "info" objects needs to be fast (for small code regions)
    • Using hashes would only solve part of the problem
  • Add setter/getter with void* and
  • Could this be done via CVARs?
    • Some API needed to query the specific object a CVAR is bound to
    • One might bind a CVAR to an Info object
      • Requires multiple Info objects to identify as different instrumentation points.
      • Distinct CVARs might be needed to indicate begin/end

• Would a new handle type MPI_Tool coming from QMPI registration work better to convey information

• Non-blocking version?

  • How to make sure completion can be handled by the library?
    • Something along the lines of the MPI_Pready call to indicate that a tool that intercepts Qcontrol is done manipulating the info object

• Should a tool be required to pass on to next layer?

  • Is this info for me?
    • Exclusively?