[mpiwg-hybridpm] mpi_memory_kinds default clarification (feedback window closing soon)

Rohit Zambre rzambre at nvidia.com
Fri Dec 13 14:49:17 CST 2024


Looks like we have closed on this. I’ll plan to bring forward the following text as an erratum proposal for the Jan 8 meeting (announcement deadline is Dec 25).
https://urldefense.us/v3/__https://github.com/mpi-forum/mpi-standard/pull/986/files__;!!G_uCfscf7eWS!YtUdnLkr5rdXwv5R6RSgP0CfhpIGsjL4eyrVvGjVYCszZRgoqdFq9GzzQYP9_TQ1YkQuDdfAyBfocOhcFKuyiBvKxyFM8eE$ 

Thank you,
Rohit

From: Rohit Zambre <rzambre at nvidia.com>
Date: Monday, December 9, 2024 at 6:16 PM
To: mpiwg-hybridpm at lists.mpi-forum.org <mpiwg-hybridpm at lists.mpi-forum.org>
Subject: Re: mpi_memory_kinds default clarification (feedback window closing soon)

Hello hybrid WG,

Thank you for the discussion at the MPI Forum meeting today. Summary points from the online discussion below.
The feedback window is closing soon as we are aiming to bring this into the January 8, 2025 Forum voting meeting as an erratum vote.
Please provide your feedback on the PR — https://urldefense.us/v3/__https://github.com/mpi-forum/mpi-standard/pull/986__;!!G_uCfscf7eWS!YtUdnLkr5rdXwv5R6RSgP0CfhpIGsjL4eyrVvGjVYCszZRgoqdFq9GzzQYP9_TQ1YkQuDdfAyBfocOhcFKuyiBvKXdiaVso$ 


  1.  Strong agreement on an explicit clarification about the MPI library returning `mpi` and `system` at minimum when defaulted.
  2.  Some agreement and some ambivalence on including the additional statement “Different sessions may return different default values”.
     *   No strong opposition on removing the statement.

Based on the online and offline feedback, I have revised the clarification text to include the additional statement:
“
When defaulted, the value returned must, at minimum, contain the kinds specified in “default" and may contain
additional implementation-defined kinds. Different sessions may return different default values.
"

The additional statement clarifies applicability to existing MPI implementations with non-standard ways of choosing memory kinds.

Thank you,
Rohit

From: Rohit Zambre <rzambre at nvidia.com>
Date: Tuesday, November 26, 2024 at 10:27 AM
To: mpiwg-hybridpm at lists.mpi-forum.org <mpiwg-hybridpm at lists.mpi-forum.org>
Subject: Feedback on mpi_memory_kinds default clarifcation

Hello Hybrid WG,

Could you please take a look at https://urldefense.us/v3/__https://github.com/mpi-forum/mpi-standard/pull/986__;!!G_uCfscf7eWS!YtUdnLkr5rdXwv5R6RSgP0CfhpIGsjL4eyrVvGjVYCszZRgoqdFq9GzzQYP9_TQ1YkQuDdfAyBfocOhcFKuyiBvKXdiaVso$ 
and provide any feedback you may have?

Here is the relevant issue — https://urldefense.us/v3/__https://github.com/mpi-forum/mpi-issues/issues/773__;!!G_uCfscf7eWS!YtUdnLkr5rdXwv5R6RSgP0CfhpIGsjL4eyrVvGjVYCszZRgoqdFq9GzzQYP9_TQ1YkQuDdfAyBfocOhcFKuyiBvKi044sy8$ 

Thank you,
Rohit
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mpi-forum.org/pipermail/mpiwg-hybridpm/attachments/20241213/bdeadd72/attachment-0001.html>


More information about the mpiwg-hybridpm mailing list