1 / 8

Large Counts in MPI

Large Counts in MPI. A Brief History. How it all started. HP-MPI users request 64-bit counts ( nat’l labs) HP takes issue up in Forum, Ticket #14, September 2008 Proposes new extended functions for inclusion in MPI 2.2 ( MPI_SendL , etc) Covers pt2pt, collective, datatypes , rma

jewel
Download Presentation

Large Counts in MPI

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Large Counts in MPI A Brief History

  2. How it all started • HP-MPI users request 64-bit counts (nat’l labs) • HP takes issue up in Forum, Ticket #14, September 2008 • Proposes new extended functions for inclusion in MPI 2.2 (MPI_SendL, etc) • Covers pt2pt, collective, datatypes, rma • Uses MPI_Aint for count type

  3. Forum Feedback • Too big a change for 2.2, delay to 3.0 • Define new MPI_Count type instead of using MPI_Aint • Leads to ticket #117, Feb 2009

  4. Ticket #117 • Modifies existing APIs • Covers pt2pt, collective, datatypes, rma • Uses MPI_Count for count type

  5. Forum Feedback • Lots of thrashing • Does anyone really needMPI_Count? • Does new functionality use MPI_Count? • What’s the impact on Fortran? • Is the MPI_Count version of all APIs too big of an API explosion? • HP sells HP-MPI, ticket lays idle for a while

  6. Ticket #220 • Focus shifts from communication to MPI-IO • Ticket #117 resurrected, changed in scope to just: • MPI_GET_COUNT_L • MPI_GET_ELEMENTS_L • MPI_STATUS_SET_ELEMENTS_L • Ticket #220 introduces _L version of MPI-IO routines. • Back to _L suffix!

  7. Forum Feedback • Discussion about suffixing rules • What happens when the next suffix is introduced? • i.e. timeout parameter • Generally rejected by forum as being too big of a change for too little benefit • Dave Gives up.

  8. Current Proposal Rip off the band-aid!! • Define MPI_Count as >= int • Define as int for back compatibility • Apply it to all APIs • No suffixing • Requires two libs • back compat • long counts • Can be handled in single codebase • Implementers decide which to ship/support

More Related