Skip to content

Latest commit

 

History

History
102 lines (83 loc) · 3.47 KB

2024-01-17.rst

File metadata and controls

102 lines (83 loc) · 3.47 KB
Raw

1/17/2024

Attendees:

  • Will Damon
  • Robert Cohn
  • Penporn Koanantakool
  • Alexey Kukanov
  • Maria Garzaran
  • Greg Lueck
  • Maria Kraynyuk
  • Rod Burns
  • Mehdi Goli
  • Timmie Smith
  • Ravindra Babu Ganapathi

Agenda:

  • Introductions
  • Opens
  • SYCL specification
  • Review Operational Procedures
  • Expanding participation beyond Intel
  • Retrospective
  • Publishing Infrastructure

Notes

  • Introductions
  • Opens
    • Participation
      • Robert invited members of spec working group mailing list. Next time invite steering committee so they can invite their organizations.
      • Some difficulties connecting to zoom. Are mailing list and zoom login the same? Do people know guest registration is available?
  • SYCL specification
    • Problem: oneAPI spec includes chapter defining required SYCL extensions. Not clear why we need it. Some members of Khronos SYCL working group feel that it is a competing specification
    • Robert proposed removing standalone chapter and components are free to list SYCL requirements in their spec.
    • Alexey wants important extensions with no line of sight to SYCL spec to be documented somewhere.
    • As an example, Alexey pointed out that DPL needs to document which std API can be used. This is mostly determined by SYCL support for use of std in kernels. He would rather SYCL/dpc++ document it. Greg said SYCL current spec does not require that any API, but SYCL working group acknowledges that a list is needed. There is a documented extension that lists API. oneAPI spec refers to that extension.
    • Not clear how to resolve this. Robert to follow up with Alexey, Greg, and other interested parties.
  • Review Operational Procedures
    • Operational procedures are being finalized now so it would be good to understand their impact on how we develop spec and propose changes if necessary.
    • Documents: UXLF Operational procedures and Robert's markup
    • We reviewed the highlighted sections. Most of the discussion was about how to interpret the document.
    • Robert called out the 7/14 day waiting period for accepting substantial changes. 3 required approvers. No one raised a strong objection at the meeting, team members will discuss it with their teams and provide feedback if there are issues. Rod proposed documenting the list of approvers.
  • Expanding participation beyond Intel
    • There needs to be a clear path for new people to participate in specification development. A steering committee member said that onemkl was clear, but it was harder to find for oneDNN. This is odd because oneDNN has a lot of non Intel participation. Rod and Alison suggested projects pro-actively looking for reasons for connections with other projects. Robert will look at project documentation for how to get involved in specification if you are not already a member of the project.
  • Retrospective
    • Now is a good time to make changes to the way we develop and publish specifications, so everyone is invited to raise issues and make proposals.
  • Publishing Infrastructure
    • Robert wants to stop using Akamai/AWS to publish specs and move to GitHub pages. Akamai/AWS was managed by Intel IT and would not work well for non-Intel employees.